Study On X-ray Weld Image Defect Detection System Based On AI

In recent years,the rapid development of artificial intelligence technology has promoted the intelligentization of various industries.Especially in the field of non-destructive testing,computer automatic identification technology has gradually replaced the traditional manual evaluation method with its advantages of high efficiency,standardization and intelligence.For the defect detection based on X-ray weld image,the traditional manual evaluation method is inefficient,the defect detection standards are different,and it is subject to the subjective influence of the staff.Therefore,in order to improve the detection accuracy and work efficiency,it is of great significance and application value to study the automatic defect detection technology based on image processing technology and computer vision.In this paper,the X-ray weld images of spiral submerged arc welding and girth welds are used as research objects,and welding defect detection and recognition algorithms are studied,which mainly include three aspects: weld image preprocessing,weld defect segmentation,and defect recognition.According to the actual needs,a set of X-ray weld defect detection system was designed and developed.The specific content is summarized as follows:(1)In the image preprocessing process,aiming at the problems of high noise,low contrast,and poor image quality in X-ray images,the noise model and defect characteristics in the image are analyzed,and the average filter method is used to remove noise,and the sin function is used to enhance the algorithm Improve the contrast of welds and background areas;(2)In the image segmentation process,the Otsu method is used to combine Sobel operator edge detection and Hough transform to achieve accurate extraction of ROI regions.When segmenting SDR,the concept of gray density was introduced to improve the accuracy of clustering segmentation defects and achieve good segmentation results;(3)In the process of defect recognition,a deep convolutional neural network algorithm is used for image classification.A 6-level 10-layer CNN network structure is determined.A tanh function is selected as the activation function.The convolution kernel is set to 5 * 5 and 3 * 3,the average classification accuracy rate is 98.97%,and the recognition accuracy rate is high;(4)Designed and developed X-ray weld defect detection system,including server and client,to meet the needs of staff remote work.