Research On Weld Surface Defects Detection Method Based On Machine Vision

As one of the important forming technologies in the field of machinery manufacturing,welding is widely used in automobile manufacturing,construction,shipbuilding etc.The quality of the weld is the core of the safety guarantee for the use of weldments.It affects the performance of the final product,and even leads to product failure in severe cases.Due to the influence of the environment and welding process,defects such as cracks and holes often appear in weld after welding,which affects the structural performance of the weldment.Therefore,it is very necessary to inspect the surface quality of welds.At present,weld surface defect detection is mainly performed by the visual inspector.It is susceptible to subjective factors to cause problems such as missed detection and misdetection,and can't meet the industrial needs of modern society.Machine vision is a detection method with advantages of rich information,intuitiveness,and high precision,etc,which has been widely used in the field of defect detection.This paper is based on machine vision weld surface defect detection research,the main contents of the research are as follows:(1)Aiming at the problems of noise in the collected weld surface image flat butt welding and the similar contrast between the base metal and the weld area,a series of preprocessing methods such as grayscale,bilateral filtering,and edge detection are used to achieve the "weld area" Rough extraction" and proposed a weld extraction algorithm based on PCA,which can recognize two welds and three welding methods.(2)Aiming at the problems of different shapes,irregularities,micro-targets and inconspicuous features of holes defects on the surface of welds,some existing studies use their geometric features to identify,and there are serious problems of missed detection and false detection.This paper presents a holes detection algorithm based on momentum.The algorithm is used to identify the holes in the extracted weld area,and the flood fill algorithm is used to fit.Finally,the pore size is obtained by the camera calibration parameter conversion,and the recognition rate is 87.8%,and the holes error is less than0.5mm~2.(3)Aiming at the singleness and low detection speed of the traditional visual weld surface defect detection algorithm,the YOLOv4 model and the problems in the application of weld defect detection are analyzed,and an improved YOLOv4 algorithm is proposed for the weld surface defect detection method.Reduce the number of parameters of the backbone extraction network and improve the real-time performance of the system.Using the collected images to create a data set,model training,testing,and comparison experiments of multiple models,the method in this paper performs better,with the average recognition rate 83.07%.(4)An experimental platform was built to conduct related experiments to verify the rationality of the algorithm.The standard welding defect samples from the Engineering Training Center of Xinjiang University were used as the experimental objects to verify the feasibility and effectiveness.The average recognition rate obtained was 81.03%.