Surface Defect Detection System For Welding Product Based On Machine Vision

As one important step in production welding, welded components appearance testing has becomean important measure of high quality welded components quality. For the weld appearance evaluationprocess, artificial evaluation are influenced greatly by subjective factors and caused misjudgmentinformation easily, lead the accuracy of assess results low. With the rapid development of thecomputer, pattern recognition and electronic machinery, the intelligent auxiliary detection based onmachine vision technology has become possible. But it is still exist the problem of low signal to noiseratio and low contrast after CCD camera, so the welded components appearance quality automatic andintelligent test remain more difficult. In this paper, the applications of acquisition, processing andrecognition technologies for images of the welded components have been explored, and the extractionand identification technology of defect area have been realized.First of all, according to the production environment and the requirements of mechanical deviceof welded component detection system based on machine vision recognition system, the automaticrotation device and light compensation device of CCD camera were built in this paper, and weldedcomponents appearance collection can be realized by this device. At the same time, considering basicprocess of image recognition, this system can be divided into five modules, such as weldingappearance acquisition module, image preprocessing module, image segmentation module, weldproduction appearance defects recognition module and information maintenance module.Secondly, on the basis of the characteristics of welded components appearance collected by CCDcamera, the image preprocessing module of this system is divided into four basic functions, such asimage gray-scale transformation function, image restoration function, image noise reduction functionand image enhancement function. Among them, fuzzy images that objects in motion state have beentoken photograph can be recovered by wiener filtering method in order to get clear images. Imagesegmentation is subsequently done to provide good quality images, the images can be processed byadopting the method of de-nosing and enhancement in order to improve signal-to-noise ratio and thecontrast of image.Also, considering the analysis the image appearance, especially including complex appearanceand so on, the images can be segmented by the methods based on improved kernel fuzzy k-means. Byintroducing nuclear binding β and membership degree adjustment coefficient λ, ROI (Region ofInterest) will be accurately segmented. Moreover, segmented image could be marked by the regional marker method. By adopting the technology of image silhouette, the extraction of the welding seamand defect zone will be realized. In addition, the blank of the defect zone will be filled by seed fillalgorithm, in order to improve the calculation accuracy of defect area characteristic parameters.Finally, with the analysis of the relevant defect recognition standard, the characteristics of thewelded component appearance defects parameters can be summarized. Expert system based onrepository was developed by considering both the parameters and weld processing factors, and theappearance defect of welded components will be automatically identified. Expert system which adoptsthe ideal of the fuzzy reasoning to carry out the defect identification and add the detection informationmanagement database, realized the store of information, therefore the information of product qualitycan conveniently be traced by technicians.The welded components appearance detection system has been designed and implemented in thispaper. Through the use of computer program technology, realized function including figure collection,processing and analysis of the weld appearance so that the system can be applied to practicalapplication.