Detection Method Of Weld Quality Defect Based On 3D Vision

Traditional weld quality inspection mainly relies on inspectors through visual inspection or the use of measuring tools such as weld inspection rulers and endoscopes to achieve,these inspection methods are not only inefficient,but also the accuracy of detection is limited by the experience of the measuring personnel themselves.With the continuous development of industrial vision,two-dimensional vision to locate and analyze weld quality defects gradually replaced the manual detection method,which can largely improve the detection speed and accuracy,however,because two-dimensional vision can only extract two-dimensional information about the weld can not get its depth information,so for the weld burn-through,weld tumor and other defects with greater depth is difficult to accurately locate and analyze In response to the above problems,this paper designs and develops a 3D vision-based weld quality defect detection system,which obtains the initial point cloud data by scanning the workpiece with a robotdriven line laser contour sensor,and then the host computer extracts the weld area and analyzes the defects existing in the weld.The specific work includes the following aspects(1)A weld quality defect detection system based on 3D vision is designed and built.The hardware of the system includes robot,laser sensor and industrial computer.The software of the system integrates the commonly used point cloud processing algorithm,and the whole software can realize the initial data acquisition,weld area extraction,defect location analysis,realtime processing display.(2)The weld quality defect detection system was calibrated,including the robot Tool Center Point(TCP)calibration and eye-in-hand calibration between sensor and robot.We analyzed and compared the four-point method and the six-point method commonly used in TCP calibration,and conducted TCP calibration experiment using the six-point method for the experimental scenario and accuracy requirement of this paper.In order to accurately collect the initial point cloud data of the arc-shaped workpiece,this paper carries out eye-in-hand calibration of the system based on the principle of spherical calibration method,and the experimental results show that the system calibration accuracy can reach 0.05 mm.(3)In view of the problems such as time-consuming increase and inaccurate defect location analysis caused by direct detect defects of weld point cloud data,based on the characteristics of line laser profile sensor,this paper proposed a weld point cloud segmentation algorithm integrating contour features,the algorithm uses the separation-combination,mapping-reflection method to process a single contour,and then recombine all the weld area to reflect back to threedimensional space.Experimental results show that this algorithm can accurately extract the weld area.(4)For the problem of inaccurate analysis of weld defect localization due to under-segmentation and over-segmentation when using area growth segmentation algorithm to extract defects such as burn-through,weld tumor,etc.,based on the curvature and normal vector characteristics of the point cloud,this paper proposes an improved area growth segmentation algorithm,which is based on the principle of greedy algorithm to achieve data search clustering under different conditions,thus avoiding the above problem;for the width and height of weld property,nibbling,spatter defects,this paper proposes a new formula and detection method;experimental results show that the improved area growth segmentation algorithm and detection method can extract nibbling,weld tumor,spatter,burn-through,etc.,and the recognition accuracy can reach0.05 mm.