Method Research Of Weld Defects Inspection Based On Machine Vision

In the modern machinery manufacturing field, as a basic process technology, welding process is used more and more widely in the field of machinery manufacturing, nuclear industry, aerospace industry, energy and transportation, oil chemical industry, building and electronic industry, etc. That the welding quality is good or bad may directly affect the service life of product and reliability. So it is very important to inspect the weld defects in target image. At present, welding image is gotten mainly through X-ray nondestructive. However, welding image evaluation is basically using artificial detection and judgment, and it has many shortcomings, such as subjective standard is not consistent, labor intensity and low efficiency. Therefore, modern production needs a kind of effective and automatic method of weld defects inspection to replace artificial detection, so that the detection work will be objectification, standardization and intelligent.This paper gets the film which through X-ray nondestructive detection of welding member as research objects. First, let the film to be digital through CMOS industrial digital camera; Second, research the algorithm of weld defects inspection; last, do the identification of the weld defects which is already inspected. Through these steps, we can get a weld defect inspection and identification system which including the function of image acquisition, processing, identification and assessment. The specific content is as follows:(1) Build the acquisition platform of X-ray weld film based on machine vision, decide the image acquisition method according to the inspection requirement and choose the camera and light source for image acquisition.(2) Do the preprocessing of the collected image, including the extraction of effective area and median filtering, then use the SUSAN algorithm for weld defects inspection, and with other defects inspection methods are compared.(3) Understand the different kinds of defects and their characters, analyze the selection and calculation method of defects characteristic parameters. At the same time, get sixty images which are already known of crack defects, lack of fusion defects and stoma defects to do feature parameters extraction and classification, and to establish sample library for defect recognition using.(4) Based on the Support Vector Machine (SVM) method to do classification and identification of weld defects. Know the theory of SVM, train the simple library combined with LIBSVM library to get training model. Finally, do the identification and classification of unknown defects based on the training model.(5) Using the Visual C++6.0software to design the interface of weld defect inspection, and testing the defect inspection and classification algorithm in this platform. The correct segmentation rate and identification rate is about87%and82%respectively compared with artificial observation, the detection time of each weld is about2seconds.