Research On Intelligent Identification And Repair Method Of 3D Laser Point Cloud Holes In Storage Tank Wall Based On Multiple Information Fusion

In recent years,tank farm accidents occur frequently,which has caused huge economic losses and a bad social impact on enterprises.Personnel and property losses caused by tank farm accidents shall be avoided.The main way to avoid tank accidents is to detect the state of the tank,that is,to monitor whether the tank is deformed and make corresponding decisions according to the deformation detection results.Because of its advantages of high precision and automation,three-dimensional laser scanning technology can realize the all-around and continuous measurement of the storage tank in the field of storage tank deformation detection,and can more comprehensively evaluate the deformation of the storage tank,which has a good application prospect.However,with the development of large-scale storage tanks,the three-dimensional scanning point cloud not only brings all-around and high-precision detection results to the deformation detection of storage tanks,but also brings the point cloud processing process with high data capacity,high computational power requirements,and high labor consumption.To optimize the complex point cloud processing process,this thesis carries out the following three aspects in combination with the special shape of storage tanks and the advantages of computer image processing:(1)Aiming at the problems of the large volume of the storage tank,a large number of scanned point clouds,time-consuming and laborious manual segmentation of point clouds,and strong subjectivity,this thesis proposes an intelligent cleaning method of three-dimensional point clouds of storage tank wall based on model fitting.Based on fitting the parameters of the cylindrical model by the least square method,the threshold is determined by extracting the distribution of point clouds based on a normal vector,which makes the cleaning of three-dimensional point clouds of storage tank walls more accurate and time-efficient.Taking two large and small storage tanks in petroleum and petrochemical enterprises as an example,this thesis applies the proposed method.Compared with the traditional manual point cloud segmentation method,the accuracy and recall rate of the proposed method are more than 99.9%;In the case of threedimensional point cloud data cleaning of two storage tanks,the time efficiency of this method is 3 times,and 15 times higher than that of manual segmentation of point cloud,respectively.(2)Aiming at the problem that the traditional point cloud hole recognition method based on scattered distribution is greatly affected by the density of the point cloud,there is much noise extracted from the hole boundary,and the running time is long,which can not classify the holes,this thesis proposes an intelligent recognition method of threedimensional point cloud holes on the tank wall based on the image point cloud.By mapping the three-dimensional point cloud of the storage tank into a binary image,detecting and clustering the hole boundary through the edge detection algorithm,and then mapping the coordinates of the hole boundary back to the three-dimensional point cloud,the three-dimensional boundary coordinates of the point cloud hole are obtained,which solves the problem that the extraction of the point cloud hole boundary is noisy,the running time is long,and the point cloud hole can not be classified.Taking two large and small tank walls with point cloud holes as an example,using this method,the traditional method can not identify the point cloud holes of the tank wall.In the identification cases of point cloud holes of two large and small tank walls,the hole recall rate and accuracy rate are 100%,and the time efficiency is 137316 times higher than that of the traditional method.(3)Aiming at the problem that the tank wall hole repair method based on interpolation BP neural network needs to artificially identify the hole boundary,train multiple models and the low prediction accuracy of a single axis,a tank wall point cloud hole repair method based on adjacent axes is proposed.Taking the data of two dimensions as the input,the GA-BP neural network model is trained and the point cloud missing point cloud is predicted.Taking the tank walls of two large and small tanks with point cloud holes in petroleum and petrochemical enterprises as an example,compared with other methods,the point cloud hole repair method based on the adjacent axis proposed in this thesis improves the accuracy by 1.7 times and 4.2 times,and the time efficiency by 99 times and 88 times respectively.