Research On X-ray Digital Imaging Based Weld Defect Detection

In this dissertation, a weld defect detection system is developed for satisfying the actual demands, in which X-ray digital imaging device is used to generate weld digital images. By applying image processing algorithms, the weld seam is extracted for detecting weld defects.With the development of computer and pattern recognition technology, particularly the breakthrough of X-ray digital flat panel detector technology, X-ray digital detection has been more significant in application of industrial defect detection, of which the digital image based weld defect technology plays the key role in nondestructive testing. Comparing with the traditional film based detection method, X-ray digital image approach has notable advantages in security detection, time efficiency, cost-effective, convenient operation and preservation photos.For the purpose of preprocessing weld images with large contrast and strong noise, Wiener filtering algorithm is applied to reduce noise and index transform method combing with the segmentation method to enhance images. Moreover, comparing with other existing algorithms, wavelet transform method has a good response in weld edge detection.The procedure of obtaining the weld areas can be divided into two steps. First, the amount of X-ray transmission should be converted into digital images. Second, the digital images should be segmented. There exist several parameters in X-ray digital imaging system to be determined including focal length, tube voltage, stacking frames. Reasonable parameters have been adopted by using the optimal transmission parameters selecting approach. After preprocessing the weld digital images in pertinence, a gray vertical slope stepping based digital subtraction technology is proposed. Furthermore, image rotation method is utilized for deriving the weld, which is simple, convenient, accurate, effective and the background redundant information is reduced.Gray-scale, geometry and texture are the main features of the weld images. In dealing with the gray scale of weld defects, the Gaussian kernel-based Meanshift method is applied to extract the defect gray scale, which leads to a good result in weld image segmentation. Based on this, the Meanshift peak algorithm is used to mark defects in satisfying the criteria of artificial selection. After series of artificial defects are obtained, the SVM algorithm is used to achieve machine learning for detecting new defects in new weld. Experimental examples are given to show the effectiveness and feasibility of the proposed methods.Finally, by summarizing this dissertation briefly, the prospects of the main stream in weld defect detecting for industrial automation are discussed.