| In this study,aiming at the industry status and needs of prefabricated structures in the manufacturing and construction process,combined with 3D laser scanning technology and computer vision algorithms,an intelligent dimension detection of steel cages before pouring and completed prefabricated components was established.Based on BIM and robot technology,the automatic obstacle avoidance and assembly of prefabricated components on the construction site are realized,thereby realizing the intelligent construction of prefabricated structures.The research work is mainly divided into the following four points:(1)This study uses 3D laser scanning to obtain point cloud models of steel cages and prefabricated components,and proposes data preprocessing methods such as denoising,coordinate conversion,and downsampling to improve the quality of point clouds.Aiming at the current situation of scarcity of on-site scanning data sets,a synthetic point cloud model generation method for prefabricated components based on geometric features and coordinate matrices is proposed,which realizes the rapid,automatic,and batch generation of prefabricated component point clouds.Besides,it can also realize the automatic labeling and custom density adjustment of point clouds of subcomponents(rebar,concrete).This method greatly improves the acquisition efficiency of deep learning point cloud datasets,and can be used as a supplement to real scanning point cloud datasets.(2)An automatic recognition method for prestressed pipeline line shape of prefabricated components before pouring is established.A segmentation framework combining the radius neighbor covariance feature and DBSCAN is proposed to recognition the point cloud of the steel cage and the prestressed pipeline of the prefabricated component,and the accuracy reaches 96.95% compared with manual segmentation.For the point cloud of the prestressed pipeline,a line fitting method that combines the point cloud slicing function and the Newton iterative circle fitting algorithm is constructed,which realizes the high-precision extraction of the axis center of the prestressed pipeline section and the automatic detection of the overall line shape.Compared with the field measured data,the average error is 1.84%,which verifies the effectiveness of the method.(3)Aiming at the problem of insufficient generalization ability of classical machine learning,an automatic dimension detection method for precast components based on Precast Component Recognition Net(PCCR-Net)was established.A PCCRNet is proposed to recognize and segment four types of typical precast component point clouds.Combined with machine learning algorithms,automatic dimension inspection of subcomponents is realized.Compared with the measured data on site,the average deviations of concrete size,rebar length and rebar spacing are 1.6,1.1 and 1.0 mm,which are far smaller than the allowable manufacturing tolerance.It shows that this method can accurately measure the dimensions of various precast components(precast columns,beams,panels,walls),saving labor time and measurement costs.(4)An automatic obstacle avoidance and assembly method for prefabricated components based on BIM and robot technology is proposed.This method establishes a component library for prefabricated structures based on BIM,and determines a reasonable installation sequence and location of prefabricated components.Based on the RRT-Star algorithm that introduces Minkowski difference and the numerical solution of inverse kinematics,the prefabricated component installation path and obstacle avoidance algorithm is established.Furthermore,the trajectory parameters of the automatic installation and operation of the robot arm are calculated,to realize the automatic obstacle avoidance and installation of prefabricated components.The effectiveness of this method in terms of component installation sequence and positioning accuracy,obstacle avoidance performance,trajectory smoothness and time efficiency is verified by simulation experiments. |
PDF Full Text Request