Automated visual inspection of cracks in cast and welded components

An automated inspection system for detecting external cracks in complex-shaped components has been developed. The components are scanned by a video camera and the digitized images are transfer, ed to a computer for defect identification. Heuristics for crack identification have been developed by studying the chain of reasoning used by human inspectors. These heuristics have been isolated, sorted according to their importance and implemented in a computer program. The program first generates some seed points at various edges and corners in the image and then tests each location to identify the cracks, if any. Compared to other reported investigations, this method can use raw images without any contrast enhancement or noise reduction. It does not depend on "ideal" images for comparison, any training before use or part-specific instructions. The inspection system has been successfully rested by digitizing the images of several castings. In each case, it took one to two seconds to detect and isolate the crack region. The speed of response makes it suitable for application in automated foundries with high rates of production. Other applications being considered include underwater inspection of cast and welded structures.