Font Size: a A A

Research On Key Technologies Of Computer Vision For Tube Truss Structure Automatic Teaching Of Welding

Posted on:2016-03-30Degree:DoctorType:Dissertation
Country:ChinaCandidate:X T LinFull Text:PDF
GTID:1318330518471319Subject:Detection Technology and automatic device
Abstract/Summary:PDF Full Text Request
As the development of automated welding technology,especially the development and mature of computer vision technology,how to use these new technologies to expand the scope of application of automated welding and spun off manual from harsh environment and heavy work become a serious problem for industrial welding processing areas.Truss structure has been widely applied in marine engineering,bridge construction and load-bearing equipment,with its excellent properties.But intersecting structures can't automated welding for the complexity of the welding environment and high welding process standards,which is widespread in truss structure production.The way taking advantage of the latest advances in computer vision technology to solve key technologies of computer vision for pipe truss structure automatic teaching of welding become contents required in-depth study in this article.Based on the analysis of characteristics and welding process requirements of the pipe truss structure,three primary key technologies of computer vision for automatic teaching of welding need to be solved,which are 3D information detection of intersection structure,3D information detection of surface near teaching point and distribution information detection of the filler in bead section cross teaching point.A novel multi-view 3D reconstruction algorithm is proposed based on position limit dense patches,which is used to detect 3D information of various shapes,larger size,weak texture and occlusion tube truss structure.For having various shapes,larger size,texture and presence of weak occlusion tube truss structure features three-dimensional reconstruction,this paper analyzes and summarizes the existing three-dimensional reconstruction algorithm is proposed position limit dense patches of multi-view three-dimensional reconstruction algorithm.For the three-dimensional information detection problem teaching point where the local surface,based on the existing local surface reconstruction algorithm for analysis summary,we propose a non-calibrated light photometric stereo reconstruction algorithm based on closed-solver of GBR parameters.By GBR parameters solver to complete the calibration light source,camera and the surface positional relationship;combined with noise reduction,segmentation and repair steps to complete the three-dimensional reconstruction of the local surface to achieve the detection of local surface information.Because of the three-dimensional reconstruction of the algorithm does not require the local surface optimization,iterative processing steps,such as direct use points on the surface of the GBR parameters solver to improve the accuracy and speed of the surface reconstruction.By comparing the experimental results show that the algorithm advantage in the detection accuracy and speed to meet the self-teach the course of the torch pose correction needs.This article uses the proven line of structured light for a teaching point where the cross section of the weld filler distribution information for testing.In order to facilitate the evaluation of the centerline extraction algorithm,and a standard light bar contains images of known luminosity centerline unevenly distributed,intermittent noise and other content.By analyzing the shortcomings of the existing structured light stripe centerline extraction algorithms,we propose a centerline extraction based on anisotropic thermal diffusion algorithm.By comparing the experiment proved that the algorithm advantages in accuracy and speed.The key point of the center line detection using improved CPDA critical point detection algorithm,experiments show that the algorithm achieves better detection of critical points.The key point in the combination of the center line and get exactly the weld filler distribution information.After completing the research on three key visual technology,tube truss simulated autonomous teaching experiments.Through this experiment,each further confirmed that the proposed algorithm to solve the three-dimensional information on the primary structure of the teaching process node detection,teaching point where the local surface distribution of the filler on the bead section where the three-dimensional detection and teaching points visual detection of these three key technical problems of validity of the information.
Keywords/Search Tags:Computer vision, Automatic teaching, Multi-view 3D reconstruction, Photometric stereo reconstruction, Centerline extraction
PDF Full Text Request
Related items