Study On Non-destructive Defect Detection System Based On Weld Film

It is well known that welding technology is the base of modem industry because the service life of weld products depends on the quality of welding. X-radiography is an important method of non-destructive testing (NDT). But there are a lot of shortcomings in X-radiography method, such as needing a large quantity of films, uneasy to save films, and so on, which is unsuitable for querying, modifying or reviewing the films. At the same time, evaluating film of manual method has disadvantages such as more workload, low efficiency and so on. Different workers, who evaluate film, can get different results on the same film, which heavily influence the real result of weld image with defect.In order to reduce the workload of examiners, computer aided method is used to evaluate film,from which the efficiency of evaluating film is improved. So, influence of subjective factor is decreased and the process of evaluating film is more scientific and normative. However, because of influence generated from outside noise, there are a lot of disadvantages in digitalizing weld film image such as indistinct edge of weld, low contrast of film, and so on, which will create a large quantity of difficulty in extracting feature characteristics of weld, segmenting defect, identifying image and so on.Based on the current development status of home and abroad, this paper analyzes summaries all kinds of methods of dealing with defect image. And then, the key technology and theory, working in identifying and detecting weld film of defect, is studied more, in which a kind of detecting weld defect system is designed. The main contains of this paper are introduced as the following parts.1)The current problems and solving methods are analyzed and compared based on weld image of defect according to check a large quantity of books, which is the key focus of this paper.2)From comparing current algorithm of pre-processing image, segmenting image and detecting edge of image, subtraction method is used to extract weld area, which is the basis for the subsequent method for extracting feature of weld defect and identifying defect. At the same time, the treatment process of the algorithm based on detecting weld film with defect is given in this paper.3)The common defect in weld film is analyzed and the feature characteristics of different defect are confirmed. The feature characteristics are circularity, length-width ratio, relative position of defect, gray-value difference between defect and background, gray-value difference of defect itself, and so on, which provides the basis for identifying weld defect.4)RBF neural networks is finally used to identify the image according to analyze the principle of neural network and compare common neural network. The feature characteristics of identifying weld defect are normalized treated and the feature vector is extracted, whose samples are trained in this paper. It is proved by experiments that identifying weld defect can be accomplished by RBF neural network.5)Mixed program technology in VC and MATLAB is used to design automatical detection system of weld defect. In the system, the powerful ability of processing image on MATLAB and high efficient and simple interface between human and machine on VC are full used in this design, from which the software is designed quickly and successfully.6)According to diversity of film defects, auxiliary module for detecting defect is designed, which is used to automatically identify the unrecognized weld image and help people to detect the aim defect. Based on the feature of weld image, a lot of algorithms of processing image and extracting defect are given, from which it is easy and convenient for worker to evaluate film.7)Data management module based on weld film is designed according to traditional managing film, from which film information is integrated effectively and it is easy to modify.Finally,200film images with defect are tested by this system, among which164film image are tested correctly with automatical detection method. The accuracy of detecting in automatical method is82.0%. The rest of36film images are tested by auxiliary module with the help of professional workers, among which35film images are detected correctly. The accuracy of detecting in auxiliary method is97.2%. The filed tests show that the system can achieve user’s requirements, which can reach the index for evaluating weld image.