Point Cloud Segmentation Of Deep Learning-based Deformable Materials During Stator Assembly

With the increase of labor costs,the automatic assembly of industrial,micro and special motors is a difficult problem in the motor industry,and the identification and assembly of wires is a core problem.A large deformation occurs,which brings huge difficulties to the assembly.This thesis mainly studies the identification of easily deformable materials in the assembly process of micro-motors.Taking the wire assembly of micro and special motors as the research object,firstly,a series of deformation problems caused by deformable materials during the assembly process are deeply analyzed;secondly,for the data redundancy problem in the point cloud data of deformable materials,a key point-based point is proposed.Cloud preprocessing algorithm;finally,according to the characteristics of point cloud data,a space boundary Transformer point cloud segmentation algorithm is designed.The main research contents of this thesis are as follows:1.This paper introduces the research background and significance of this topic,introduces the development process of 3D vision technology,summarizes the development status of point cloud segmentation,and analyzes the development status at home and abroad.2.By analyzing the assembly process and system design in the industrial production of micro-motor stator assembly,it is concluded that the core key in the process is the identification of easily deformed wires.Design and hardware equipment of 3D assembly vision system.3.Aiming at the problems of large amount of point cloud data,many noise outliers,and less important point clouds in the total point cloud,a new data preprocessing method based on key points is proposed.The point preprocessing method increases the point cloud identification,improves the ratio of the number of deformable wire point clouds to the overall number of point clouds,and can minimize the loss of boundary features.4.Through the analysis of the assembly process in the production process of the micromotor,it is proposed to use a 3D camera to collect point clouds,and use the improved spatial boundary Transformer method for 3D point cloud data to convert the 2D depth to the problems of occlusion and winding of easily deformed materials.The Transformer network structure in the learning network is combined into 3D deep learning,and finally the deformable wire material part in the stator point cloud is segmented through network learning.Finally,in the experimental part,the experimental results of the deformable wire preprocessing algorithm and point cloud segmentation algorithm proposed in this thesis are analyzed.The results show that the preprocessing algorithm can reduce the number of point clouds by about 90%,and also increase the point cloud of the deformable wire.The ratio to the total point cloud reduces the difficulty of algorithm learning and helps to improve the accuracy of subsequent algorithms,and the recognition rate of the spatial boundary Transformer segmentation algorithm can reach 87%.