Part Point Cloud Segmentation Method Based On Model Registration

The inspection of mechanical parts for form errors is a particularly important part of mechanical engineering.The common methods of coordinate measurement and laser scanning can inspect the surface of mechanical parts,but cannot inspect the internal geometric elements of complex internal cavity parts.Industrial Computed Tomography scanning can get the 3D measurement model of complex internal cavity parts,and then compare the 3D measurement model with the CAD model to get the overall manufacturing error of mechanical parts,but because of the common point cloud semantic segmentation often at the boundary of the part surface transition smooth,there is obvious ambiguity.Accurate segmentation of the 3D measurement model scanned by industrial CT technology to detect the form error of the geometric elements of the part is still a problem that has not been fully solved.To address this problem,this paper proposes a point cloud segmentation method based on model alignment and develops a prototype system for mechanical part form error detection based on this method,with the following details:(1)Reading and pre-processing of part CAD models and 3D measurement models.The CAD model in B-rep representation and the 3D measurement model in point cloud form are read and displayed using the OCC geometry core library.To address the shortcomings of the traditional surface sampling method,a sampling method based on surface triangle meshing is proposed to first triangulate the CAD model to obtain a preliminary guide point cloud,and then use the point cloud up-sampling technique to make the guide point cloud and the 3D measurement model similar in sample size to meet the subsequent model alignment requirements.(2)Alignment of CAD model and 3D measurement model.First,the FPFH-based alignment algorithm is used to roughly align the guiding point cloud and the 3D measurement model to ensure that the two models mostly coincide,and then the ICP algorithm is used to finely align the two models.(3)Point surface attribution of the CAD model of the part and the 3D measurement model.Firstly,the shortest distance from the points in the 3D measurement model to each surface in the CAD model is calculated by using the combination of subdivision and elimination techniques,and then the 3D measurement model is partitioned into subsets of point clouds belonging to different surfaces according to the relationship between the shortest distance from each point in the 3D measurement model to each surface in the CAD model.(4)Development of a prototype system for shape error detection of mechanical parts.Based on the QT platform and OCC geometry core library,the point cloud segmentation method proposed in this paper is combined with the actual engineering requirements to develop a prototype system for detecting common geometric errors in mechanical parts.It realizes the reading and pre-processing of model data,the alignment of part CAD model and 3D measurement model,the segmentation of 3D measurement model,and the calculation of common geometric errors.The research results of this paper have been partially applied to the project of remanufacturing and optimization analysis of complex internal cavity workpieces.