Research On Robot Grinding Planning Method Based On Casting Flash Point Cloud Information

In the field of casting manufacturing,flash grinding is an essential process.In my country,casting flash grinding is still mainly manual.Traditional manual grinding has shortcomings such as low grinding efficiency and harsh working environment.Due to the high efficiency and good stability of industrial robots,it is an inevitable trend to replace manual grinding with robots.This topic is mainly based on the casting point cloud information of the horizontal parting surface,and studies the grinding planning method of the casting flash.The main contents are as follows:(1)A digital model of the casting point cloud is constructed.The 3D point cloud information of the target is obtained by building a visual scanning platform,and a complete casting point cloud model is obtained by combining the point cloud registration algorithm.In addition,in order to extract key information more purposefully,the deep learning DGCNN network is further used to segment the workpiece point cloud,which is mainly divided into three modules: casting body,casting flash and environmental background.Among them,the casting flash is the overflow of the metal sheet due to the failure of the mold locking force caused by various factors during the casting process.Therefore,the shape of the flash is uncertain and is a free-form surface in space,which makes it difficult to determine the grinding position of the robot..In view of the above problems,it is proposed to use the edge detection based on normal estimation combined with the K-D tree radius search algorithm to segment the inner and outer edges of the flash,and retain the inner edge.Determine the grinding path height of the robot.(2)Extraction of casting grinding points.Aiming at this problem,a single-side projection algorithm of point cloud is proposed to flatten the spatial point cloud of the casting body.The selected projection plane is the plane where the inner edge of the flash is located,that is,the horizontal parting surface.After obtaining the plane point cloud of the ontology,combined with edge detection and DBSCAN clustering algorithm,the grinding points on the edge of the casting were separated.The points extracted by the algorithm are compared with the points of artificial labels,and the results show that the coincidence rate of the two is 93.13%,which verifies the feasibility of the method.In addition,in view of the selection of the fitting points of the robot grinding path,this thesis uses the improved voxel downsampling method to select the separated edge grinding points.The improved algorithm uses the nearest neighbor of the centroid point in the voxel as the representative point of the voxel for downsampling,to ensure that the selected points are all from the initial input point cloud,and to avoid introducing interference for the subsequent polishing path fitting.(3)Robot grinding path planning.For the extracted path fitting points,firstly,the grinding path is fitted with cubic non-uniform rational B-splines.The average distance and root mean square error between the evaluation point set and the nearest neighbors of the fitting curve are used as the evaluation of the fitting effect.Then,the relationship between interpolation step length,bow height error and curvature is further analyzed,and the interpolation method of grinding path point is proposed.Through analysis and calculation,the pose information of the robot end manipulator during the grinding process is determined,and the RRT algorithm is used to avoid obstacles during the movement from the robot's original position to the grinding initial point.(4)Construction of robot grinding system platform.Aiming at some problems of casting point cloud data processing,a user interface software with the functions of point cloud format conversion,point cloud visualization,casting flash inner edge extraction and casting edge grinding point extraction was developed.During the experiment,the experimental object of this thesis selects the gray cast iron casting,and extracts the grinding point of the casting on the horizontal parting surface through the developed software,performs path fitting,and interpolates the grinding path point.The grid occupied by the point cloud projection before and after flash grinding in the two-dimensional plane coordinate system is used as the grinding evaluation index.Through the experiment,it can be seen that the flash removal rate of the project scheme can reach 79.17%,which can greatly reduce the workload of manual grinding.