A Study On Pipe Skeleton Reinforcement Cages' Defect Detection Based On Machine Vision

Welding is playing an important role in the development of modern industry and widely used in many fields such as building and bridge. That welding quality is good or bad may directly affect the service life of product and reliability, so the detection on the welding defect appears particularly important. Currently, the welding defect detection of reinforcement cages is mainly detected by labor, but manual operation has such disadvantages of heavy workload, low efficiency, high cost and detection and the result of defect detection is more easily affected by subjective factors. So, we need to develop a more effective method of automatic welding defect detection to realize the standardization and intelligence of welding defect detection.This thesis gets reinforcement cages' welding defect images as research objects, which are captured by CCD(Charge Coupled Device) camera. Characteristics of the the reinforced cages' defect are extracted by image processing, and then types of welding defects are identified.According to the welding environment of reinforcement cages, a set of image acquisition device will be developed based on machine vision. Combined with the image requirement of welding defect of reinforcement cages, the CCD camera and image acquisition cards are chosen.The captured reinforcement cages' images are processed. The weld region is extracted by using differential detection method to remove the background regions; Median filtering method is adopted for noise reduction of the extracted weld region, which can reduce noise interference and keep the edge of the image;Fuzzy enhancement method is adopted for contrast enhancement of the filtered images,which makes the image welding regions more clear and details more prominent The segmentation of the enhanced images is conducted by optimal Otsu. This optimal Otsu takes the influences of Intra-class variance and Inter-class variance on the image segmentation into consideration, mean information is replaced with variance, the quality of segmentation images is improved and the rate of real-time processing is raised; The small region background of segmentation images is removed and scanned to extract the defect region of reinforcement cages, which lays the foundation for the next extraction and defect recognition.The geometric features of the welding defect image are calculated quantitatively. Eight connected region labeling method is used for target marking, chain code table is used for edge defect tracking, the defect region in the image is distinguished, and then its edge information is stored. Then, the defect area perimeter, area, long and short axis ratio and circularity are calculated.Through the research on the recognition method of the reinforcement cages' welding defects, The defect detection procedure is studied. According to surface cracks and blowhole defects, the types of defects including solder off, surface crack and surface porosity are identified by BP neural network to determine if the welded reinforcement cage meets the production requirements.