Laser Vision-based Fusion Welding Additive Forming Quality Inspection Technology

With the rapid development of manufacturing industry,in the field of fusion welding and additive manufacturing,the traditional manual visual judgment of weld quality has been unable to meet the needs of large-scale industrial production.In terms of intelligent welding,with the complexity and variety of welding methods,the traditional welding seam tracking image processing scheme is no longer universal.In this paper,the detection of roughness in the forming quality of fusion welding additive is studied,and the accurate classification of weld roughness in online welding is realized.At the same time,in the process of feature extraction,the deep learning method is applied to the intelligent field of industrial welding.The main research contents of this paper are as follows:(1)A line laser-based push-broom 3D reconstruction system is built.Through the selfcalibration of the 3D system and the joint hand-eye calibration with the robot,this set of 3D vision systems can accurately obtain the 3D contour information of the weld formation during the welding process for welding quality inspection provides reliable data support.The optimal assembly distance of line structured light device relative to welding gun of welding robot is obtained through a large number of experiments,which provides important references for intelligent detection of weld quality and online parameter correction.(2)A dynamic vision texture roughness detection algorithm based on laser vision is proposed.This algorithm combines the Tamura texture roughness in the image field with the workpiece surface roughness theory in the field of mechanical industry,and realizes the dynamic realization of the weld surface information significantly.The correct detection and classification of the surface roughness of fusion-welded additive forming was introduced.A series of experiments such as theoretical analysis of the algorithm,roughness comparison specimens,and actual welding tests finally verified the high robustness and accuracy of the roughness classification algorithm.(3)This paper proposes a welding feature segmentation extraction scheme based on ERFNet.There are two major problems in the traditional algorithm applied to the welding seam tracking visual image processing,which are unable to extract the laser fringe submerged in environmental noise and cannot be applied to a variety of welding types.In this paper,the idea of segmentation of deep learning is introduced into the laser stripe feature extraction of welds,and the accurate extraction of the centerline of the welds and the detection of feature points are achieved for a variety of welding types.Different from the traditional method of extracting lines first and then solving feature points,this solution can directly obtain the two types of features required for weld tracking at the same time.At the end of the article,a lot of comparative experiments verify the superiority and universality of the algorithm proposed in this paper.