Design And Implementation Of Weld Defect Detection System Based On Convolutional Neural Network

In the field of weld defect identification,the main method is manually inspecting the radiograph of the weld defect.This traditional method is not only inefficient but also difficult to control.At present,traditional machine learning methods can't detect the defects of things that are not fixed or difficult to move.It is difficult to identify.Therefore,a suitable process needs to achieve this type of weld seam inspection.This paper collects a dataset which contains 13006 digitalized x-ray images of weld seam.For the defect type of these samples have uneven distribution,and all those origin pictures are low signal to noise ratio,so the system cut defect part and part without defect to label pixel.In the mean time,the system provides multiple methods to cut off the weld seam part of origin picture as the detection part to improve the detection result.A complete weld seam defect detection procedure is important for weld seam defect detector and weld seam researcher.The paper provides a system which can afford development for researchers and defect detection for normal users.And this paper provides a function to manage different user role.And for developers,provide a systematical research workflow,for normal users,provide a defect detection function on weld seam picture.Firstly,for the research procedure of developers,the system provides a user friendly front page to do the filter of sample data,judgement of weld seam cut result,data preprocessing and choice of CNN model.Then a model which is built to train and then test the weld images cropped from x-ray images is constructed base on convolutional neural network.In the meantime.the system provides a front page to show the results which demonstrates what kind of preprocessing method is better and to do the classification of this dataset.This system provides a complete workflow from origin data processing to the result of convolutional neural network output,and improve the efficiency of the following up researches.Secondly,for the weld defect detection person,it provides a well-trained model and standardized process,so that the weld defect detection person can quickly and efficiently detect the weld defects.At the same time,the system provides the recognition function for normal user to detect weld seam defects so that user can get a judgement from only input a single image.This article mainly provides a systematic inspection method for large and fixed weld defect 'detection,including inspection process,model selection,result comparison,and so on.Using the trained model to detect the sample data and output the results intuitively,using this system can more easily and quickly obtain the data monitoring results and the defect results of sample data.