Welding Defects Modeling And Recognition Algorithm Reasearching Based On Back-Propagation Neural Network

The use of oil and gas transmission pipeline buried inside the arc welding defects directly affects the pipeline strength and service life,when serious causes major production accidents,so the welding defects detection is an important technology to ensure safe operation of the pipeline.Welding detection images based on X-ray becomes a hot research topic in recent years,with the rapid development of machine vision and artificial intelligence technology,makes the computer automatic recognition of welding defects had great progress.BP(Back Propagation)neural network not only has strong adaptability and robustness,but also only needs online or offline training dispense of the need of mathematical modeling in control applications,the advantages of welding defects are convenient and quick recognition,so this article uses the BP neural network algorithm to research the modeling and identification algorithm of welding defects.This article uses welding seam X-ray detection images of the submerged arc welding pipe as the research objects.In view of the cracks,porosities and noises may exists in the welding zone,it makes image pretreatment,feature description,classification,recognition and so on,to complete image recognition on the X-ray welding detection.Firstly it gets the image of weld ing boundary by denoising,enhancement,dajin(Ostu)segmentation method and Sobel edge detection and so on,then uses Hough transform line extraction method to obtain the information of welding boundary line.In the part of image enhancement it puts forward a kind of welding seam image enhancement method based on Hopfield neural network.Through constructing energy function it tranforms image enhancement problem into an optimization problem,to avoid the normalization process of grey value,at the same time makes the process of image enhancement is not restricted by the size of the image.Then it uses the gray density clustering method to complete the segmentation of welding defection areas and noises,obtains6 kinds of parameters described the characteristics of them through calculating the defects of shape features.Thus it establishs the corresponding eigenvectors,completes welding defect feature extraction.Finally using feature vectors as inputs,the BP neural network system completes the classification and recognition of cracks,porosities and noises.Through simulation experiments,this paper compares the recognizing results when usingdifferent activation functions in the hidden layer and output layer of BP neural network.The simulation experimental results show that the BP neural network algorithm can make the recognition accuracy to 92.457%,achieving rapid and accurate identification of welding defects and noises.