Research On Technology Of Automatic Extraction And Identification For Weld Defects In X-Ray Image

As one of non-destructive testing technologies for weld defects, X-ray inspection has become animportant method to ensure the quality of welding structure. For the test results, manual assessmenthas a big subjectivity. And more misjudgments are easy to be produced. With the rapid developmentof computer technology and electronic technology, the technology of computer aid assessment basedon image processing has become possible. As the problems of low noise-signal ratio and largebackground fluctuation, the automatic identification and classification of welding defects remaindifficult after digitization of X-ray film. In this paper, the application of technology for X-ray weldingimage processing has been explored, and the automatic extraction and identification technology ofwelding defects has been realized.First of all, on the basis of hardware system which can meets the requirements for X-ray digitalimage acquisition and detection, the framework of technical solutions has been built in this paper. Andthe process of automatic identification and classification of welding defects has been divided intothree parts: pre-processing of weld image, segmentation and extraction of weld defects, identificationof weld defects.Secondly, based on the technical solution framework, and aimed at the features of severalcommon welding defects in project, the pre-processing of weld image has been separated into severalimage processing technologies:256color image to8-bit gray image, extraction of weld seam, noisereduction of weld image, fuzzy enhancement of weld image. Using the technologies of ideal weldseam construction, binarization of X-ray image, gray projection and photographic subtraction, theweld seam area has been extracted successfully and the efficiency of subsequent image processing hasbeen improved. Owing to enhancement of the image noise-signal ratio and contrast, good quality weldimage has been provided for the extraction and identification of weld defects.By the analysis of problems including large background fluctuation, complex texture and defectedge blur in X-ray image, b-spline curve has been used to fit the column gray curves of weld imagesto eliminate redundant curve peaks. Defect edges are determined by a boundary threshold α and afluctuation threshold β defined from curve extreme point set which has been corrected twice. Then theedge and size of defects extracted have been amended using mathematical morphology. In addition,aiming at the fact that strip defects in X-ray image always break into several connected areas afterbinaryzation using automatic image processing, the paper has used the technology of fitting area center line to realize the automatic growth. Finally, the accuracy of segmentation and extractionalgorithms has been analyzed by comparing the characteristic parameter values extracted from welddefects and the actual values.Additionally, considering the causes of affecting weld defect characteristics, such as X-ray imagecharacteristic, welding procedure and structure, this paper has designed and realized expert systemwhich is using forward inference engine technology to identify weld defects automatically. Idea basedon inexact reasoning technology has been used by the expert system, and a more user-friendlyknowledge base management module has been designed to realize the function that technicists canadd and modify their own rules for defect identification.At last, an automatic extraction and identification system for weld defects in X-ray image hasbeen designed and implemented. Through the computer multi-threading technology, this paper hasrealized function including continuous and single frame weld image capturing, image displaying andstoring, automatic processing of image and so on. Thereby, the system can be applied to practical.