Research On The End Face Of Thick-wall Steel Tube Defect Detection Based On Machine Vision

Thick-walled steel pipes are mainly used as liquid and gas conveyor pipes and some construction pipes.The quality of the end face of thick-walled steel pipes will directly affect the assembly accuracy,thus affecting the service performance of steel pipes.At present,most domestic and foreign steel tube manufacturers complete defect detection of steel tube by manual visual inspection,which will lead to low detection efficiency,slow speed,missed detection and high cost.Machine vision technology,with its advantages of fast speed,high accuracy and strong stability,has become an indispensable trend to replace manual surface defect detection in the future.In view of the characteristics of thick-walled steel pipe end face,a method of defect detection,recognition and classification for the end face of thick-walled steel pipe is designed by using machine vision technology,and a set of defect detection device for the end face of steel pipe is built for experimental verification.The main work of this article is as follows:Firstly,according to the characteristics of thick-walled steel pipe end face and the requirements of inspection,appropriate lighting scheme and image acquisition equipment are selected.Secondly,Analyzing the characteristics of the end-face image,formulating the corresponding image segmentation algorithm,and finally using the two-level segmentation processing method.Firstly,the chamfer region is segmented by a fixed threshold and the circle contour of the chamfer region is fitted by the least square method.The chamfer eccentricity is judged by the Euclidean distance.Then,OSTU algorithm and Canny algorithm are combined to extract the contour of defects in the end plane.Thirdly,in view of the discontinuity of scratch defects,a new algorithm for discrete defect merging is proposed.The main idea of this algorithm is to judge whether the minimum enclosing rectangle of adjacent discrete defects intersects and then decide whether to merge the defects.At the same time,a new regional connection is proposed for the case that the nature of the defect itself changes greatly after morphological operation of scratch defects.The main idea of the algorithm is to find the shortest distance between adjacent defects and connect them with straight lines.At last,according to the selection principle of feature description,the feature description of defect contour is screened and extracted from the aspects of discrimination and correlation.By comparing the classification effect of classification model based on BP neural network and support vector machine,the optimal support vector machine is selected to classify the defect.