Research On Feature Inference Of Mechanical Workpieces 3D Point Clouds Based On Analytical Tensor Voting

Mechanical workpieces are the fundmental parts of the major equipments. Precise extraction of the size of the mechanical workpieces plays important roles in improving the utilization of raw materials, improving product quality and increasing the rate of qualified products. Thus, the research on feature inference of mechanical workpieces plays importantant roles in intelligent manufacturing. Based on the strong noises, feature undersampling and absence of localized characteristics of the mechanical workpieces 3D point clouds, utlizing tensor voting theory, the problem of accurate, rapid and reliable feature inference of mechanical workpieces were researched indepth.Firstly, basic thoughts of tensor voting theory were investigated, shortcomings and corresponding reasons were analyzed, laying the foundation for improving tensor voting algorithm.Secondly, a novel analytical tensor voting algorithm was proposed to reduce the complexity and heavy computational burden in traditional tensor voting. Owing to the analytical stick tensor voting being independent of particular reference coordinate system,mechanisms for plate tensor voting and ball tensor voting were proposed and the analytical solution was also solved. Thus, the problems of iterated numerical approximation, complicate computational process and the confliction between accuracy and efficiency in traditional tensor voting, which were caused by the lack of analytical solutions, were soundly solved.Thirdly, the mechanical workpices 3D point clouds were divided and dense voted based on Single Sub Voxel Cmarch algorithm. The features of mechanical workpieces 3D point clouds were inferred after eigen-decomposition for dense voting tensor.Finally, The experimental results of voting fields reveals that the proposed method was correct. The mechanical workpieces 3D point clouds were got by 3D scanner. Then,the experiment is constructed based on the strong noises, feature undersampling and absence of localized characteristics of the mechanical workpieces 3D point clouds. The experimental results reveals that the accuracy, rapidity and reliablity of the proposed methods.