Research On Detecting Technology Of Radiographic Image Of Welding Seam Of Forestry Equipment

The boom of a forestry truck crane is an important part of forestry cranes.The forestry crane uses the pulley combination at the top of the boom to lift heavy objects such as logs,raw bamboo,lacquer,and natural resin.Different boom structures and technologies make the crane performance and efficiency different.The welding seam of the boom has an important influence on its structure and performance.A good welding seam structure can make the crane work safer and have a longer life.Therefore,as the working life of cranes increases,it is necessary to inspect the welds of the boom,and the radiographic image inspection technology is an important method for non-destructive inspection of the welds.Radiographic inspection has the characteristics of fast imaging,accurate positioning,and low cost.It is a very suitable method for detecting weld defects.In this paper,through the research on the radiographic principle and weld defect image detection and other related content,the inspection process of the weld radiographic image is explained in detail.The main body of the article is divided into four parts:preprocessing,radiographic image repeatability detection,weld area segmentation,and weld defect detection.The main contents are as follows:(1)By organizing the existing GD Xray image library and combining the images with the images taken by the author on the spot,a new and richer ray image library is constructed.According to the points of the weld radiographic image in the database,the image is preprocessed.The preprocessing process is:grayscale,image denoising,and image enhancement.The purpose of preprocessing is to improve image quality,remove irrelevant factors in the image,and lay the foundation for subsequent detection links.(2)Carry out repetitive detection on the weld image library used in the article,detect and eliminate the three types of duplicate images in the image library:copying,number tampering,and repeated shooting.In this link,SVM algorithm and hash algorithm are used for detection.Among them,in the SVM algorithm system,the image will be converted into support vectors and distributed on both sides of the hyperplane in the SVM space,and then the features of the image will be extracted through the HOG operator,and finally the classification of the repeated images will be completed in the fitcecoc training model.By comparing with the detection of the hash algorithm,it is found that the SVM algorithm has a more accurate detection effect,and its comprehensive accuracy rate has reached 95%.(3)Extract the weld area of the image.In order to extract the target weld area,the article uses a variety of image segmentation algorithms to extract the weld area,such as:Roberts algorithm,Canny algorithm,Sobel algorithm,threshold extraction method,etc.,analyze and compare the performance of each algorithm,and finally choose Threshold method is used as the method of extracting weld area in the article.(4)Defect detection on weld image.In this part,the principles of deep learning are used to intelligently detect defects through neural network models.The Faster-RCNN neural network model is selected as the framework to detect weld defects.In this chapter,the model structure and principle of Faster-RCNN are introduced in detail.Compared with traditional defect detection methods,Faster-RCNN neural network detection has the characteristics of smarter and more accurate detection.Through the training results of the model,it is found that the average accuracy of this method for defect detection reaches 77.39%,which can effectively detect defect.And by optimizing the model parameters,the accuracy of the scheme has been further improved,and the final average accuracy has reached 80.60%.Experiments prove that the algorithm used in the article can meet the needs of practical applications.