| In the field of aviation manufacturing,the workpieces often have the characteristics of various shapes and variable curvature.Moreover,there are many kinds of evaluation indexes and high standard requirements for coating process.The introduction of robots for spraying operation meets the actual need and development trend of aviation industry.At present,the robot manual teaching painting method for these workpieces exists some drawbacks,such as time-consuming operation,insufficient precision,the influence of the paint film quality from workers’ experience and so on,which restricts the wide application of painting robots.The off-line programming method can automatically generate the spraying trajectory with the help of virtual simulation system,but the workpiece models often have own deviation and the pose measurement is not accurate enough.In view of this,the high-precision sensing equipment based on the principle of line structured light is used to study the workpiece modeling,pose calibration and trajectory planning method of painting robots through the point cloud processing technology in the paper.And the automatic programming system for complex structured aviation parts is constructed.The specific research contents are as follows:Aiming at the problem of insufficient accuracy and uncertain installation position of aerial workpiece model,a two-stage rough and fine matching method based on local descriptor and clustering ICP algorithm is proposed.Firstly,the line-structured laser sensor is introduced to scan the scene to obtain the measured point cloud,which is preprocessed by Pass Through filtering,voxel sampling and statistical filtering to achieve denoising and compression.Secondly,according to Poisson-Disk distribution,the existing STL model is sampled to acquire the reference point cloud with uniform spacing.Thirdly,the fpfh descriptors are extracted from the measured point cloud and the reference point cloud respectively,and the initial pose of the workpiece is calculated by combining with the evaluation function of truncation error.Finally,considering the set inclusion relationship between the reference point cloud and the measured point cloud,the target recognition and pose estimation results are obtained through the ICP fine registration algorithm after clustering segmentation.The final surface reconstruction accuracy reaches sub millimeter level,the calibration translation error is less than 1.6 mm,and the rotation error is less than0.2 degrees,which can meet the requirements of subsequent spraying trajectory planning.In order to solve the problem that the structure of aviation workpiece is complex and the overall spraying planning is difficult to meet the process requirements,the regularized segmentation method based on geometric feature learning is proposed.In this paper,two different ideas are proposed to divide the structural components into sub pieces(plane,cylinder,cone and sphere)with moderate area and obvious regularity:(1)Unsupervised clustering algorithm based on surface fusion features: the depth residuals and normal deviation angles are constructed as well described fusion histogram features,and the 3D points with high similarity are clustered by using the improved extended growth algorithm to solving the problem of structure segmentation.(2)Instance segmentation algorithm based on Point Net++: Point Net++ can classify every three-dimensional point in the point cloud into four types of regular faces(Semantic Segmentation).In this paper,this network is extended by the optimal matching theory of bipartite graph to realize end-to-end instance segmentation.According to these two segmentation methods,the RANSAC algorithm is adopted to complete the task of geometric parameter estimation(such as principal axis,radius,etc.).On the basis of the above research,four types of templated trajectory planning methods for regular sub patch point clouds are proposed respectively in the paper.At first,the coating accumulation rate model of spray gun for flat / curved surface is established by Gauss and distribution,and the map is built to pre cache and optimize the calculation.Then,based on the single path generation method of point cloud slice,the targeted full coverage path is designed according to the shape and symmetry of four types of regular surfaces.In addition,in order to meet the process requirements of expected film thickness and coating uniformity,the robot speed and travel distance of four regular surfaces are optimized through the modeling method of micro element analysis by discrete point sampling.Furthermore,based on the optimization of sub path connection by ant colony algorithm,the dynamic factor of path reversal is added under the operation experience of the nearest transition.In the end,the simulation experiments and coating quality analysis are carried out to verify the effectiveness and practicability of the relevant algorithms. |
PDF Full Text Request