Welding Seam Recognition And Defect Classification Of Pressure Vessels Based On Vision

According to the relevant national regulations,large-scale pressure vessels in service need to be inspected regularly.At present,traditional manual detection has problems such as time-consuming,labor-intensive,low efficiency,and so on.However,the use of detection robots for intelligent detection can effectively solve these shortcomings.During the visual inspection test of the pressure vessel,the internal weld seam suffered different degrees of wear and corrosion due to long-term use,which interfered with the welding seam identification;when the TOFD ultrasonic detector was used for defect classification,it is influenced to the subjective judgment of inspectors.In order to ensure that the robot runs along the established weld and accurately classify the weld defects,this paper studies the above two problems.The research of weld image recognition,industrial cameras are used to collect weld images.Through image enhancement,image smoothing,edge detection and other algorithm technologies,the acquired images are preprocessed to eliminate interference factors,enhance practical image features,and improve the weld identification.However,the Hough Line Transform of the processed pictures is used to identify and adapt its parameters,and the effect is still not obvious.The comparative analysis of welding seam image recognition methods using gray gradient and gray distribution is studied,then a visual recognition method of inspection robot based on LBP operator texture feature is proposed.In this method,the LBP histogram is compared with the Bhattacharyya distance and the threshold segmentation method is adopted to distinguish the weld area from other areas by taking advantage of the texture features such as weld moire and so on,which solves the problem that the identification of different types of welds on large pressure vessels is not very accurate due to the interference of rust and light spotsIn the aspect of welding seam defect classification,by describing the categories of welding seam defects and the currently used defect detection methods,based on the image classification technology of convolutional neural network,a new model based on improved ALEXNET framework is proposed to classify the images obtained from TOFD ultrasonic detector.Compared with the traditional detection methods that mainly rely on empirical judgments and the SVM support vector machine classification based on texture features,this research can solves the disadvantages of low efficiency and low accuracy of weld defect recognition and classification better.Finally,the above two improved methods are tested by wall climbing robot,and the experimental results show that the two schemes are feasible.