The Research About Automatic Defection Of Welding Defects From X-ray Image

X-ray detection is one of the important methods of industrial non-destructive testing.Its detection principle is to use different media to attenuate the X-ray energy to different degrees,so that the energy difference of the X-ray passing through will appear,and the internal defects of the object will be displayed on the negative film or fluorescent screen.Nowadays,with the continuous development of modern production technology,welding technology in industrial production has gradually realized automation and intelligence,greatly improving the production efficiency of welding products.However,as the most important guarantee for the safety of welding products,the quality inspection still needs the inspector to evaluate the weld image based on experience accumulation.Because of the large number of weld image,the use of manual evaluation greatly limits the production efficiency of welding products.At the same time,for inspectors,long-term observation of images will inevitably lead to wrong or missing judgment of weld images,which may reduce the safety factor of welding products and cannot fully guarantee the quality of products.Therefore,the intelligent evaluation of weld image has become an urgent problem.With the development of artificial intelligence,big data,cloud computing and other computer technology,image processing technology provides technical support for the intelligent recognition of X-ray image detection.Now the intelligent detection of defect image has become a hot topic in the field of flaw detection.This paper focuses on the in-depth study of intelligent identification of crack defects and incomplete penetration defects in the actual welding detection process.The specific research contents are as follows:1.Because of various factors in the image,there will always be noise that affects the observation image.Image processing can reduce the impact of noise on our observation image,so image processing is essential in the process of image processing.In the paper,according to the characteristics of the noise in the image,the comprehensive influence is analyzed,and the median filter is proposed to filter the image.2.In this paper,we study the characteristics of the gray-scale curve of the weld defect image,segment the image based on the gray-scale gradient change of the image,and extract the effective weld defect.On the basis of image segmentation,the advantages and disadvantages of several quadratic gradient function methods are discussed.3.Based on the extraction and calculation of seven features of weld defects as the criteria of defects.Establish the competitive neural network,input the characteristic parameters of the weld defects to train the neural network,and carry on the pattern recognition to the crack defect characteristics and the incomplete penetration characteristics.The recognition rate of the crack and the incomplete penetration is more than 80%,and the result is good.