| Storage tank is an important facility for storing oil,gas,chemical raw materials and other dangerous chemicals.It is widely used in petrochemical,military,transportation,aviation and other fields.Large petrochemical enterprises have dedicated storage areas for storage tanks,which can reach hundreds or even thousands,which poses difficulties for enterprise management.The storage tank is inevitably subject to electrochemical corrosion,material aging and stress deformation during long-term use,which leads to a certain deviation between the tank volume and the standard volume during construction,resulting in inaccurate trade handover and energy measurement.Therefore,the safety management and maintenance of storage tanks have received more and more attention in the petrochemical industry.In recent years,3D laser scanning technology has been applied to the modeling and detection of tank body with its non-contact,high-precision,digital,automatic and other characteristics.The existing tank point cloud3 D modeling methods have problems such as inaccurate tank wall shape,inaccurate tank model construction and lack of consideration of tank wall thickness.In addition,due to occlusion,limited resolution and environmental impact,the point cloud at the tank bottom collected by the laser scanner has missing shape,and the measurement is inaccurate or even the algorithm fails due to the lack of measurement point data when calculating the tank bottom capacity using the equal area method.At present,when the inspection robot detects the defects at the bottom of the tank,the path planning algorithm is prone to fall into local minima.At the same time,the localization technology of the robot in the closed environment of the tank is limited,which affects the effect of path planning.In addition,the research on the determination of defect location and size judgment of the automatic inspection method for the defects at the bottom of the tank is not perfect.Aiming at the above problems,this paper conducts research from three aspects: three-dimensional modeling of tank point cloud,point cloud completion algorithm,path planning and defect detection method of inspection robot.The main work completed is as follows:(1)To solve the problems of incomplete tank wall shape segmentation and inaccurate model design in the existing tank point cloud threedimensional modeling methods,a tank body point cloud segmentation algorithm based on MSAC algorithm is proposed to completely segment the tank wall point cloud,and further separate the tank roof and tank bottom point cloud according to the height threshold.Then,the tank wall and the tank roof are modeled by building a layered circular plate based on the circular truncated cone model.Then,the inner diameter of the ring is obtained by fitting the inner and outer diameters of the ring based on the least square method,and the circular truncated cone model is optimized to obtain an accurate tank body model.The real data experiment shows that this method can effectively improve the accuracy of three-dimensional modeling of tank body point cloud,and the average error rate of volume measurement is only 0.85‰ with known bottom volume.(2)The missing of the point cloud at the tank bottom will lead to inaccurate calculation of the volume and bottom volume.However,the existing point cloud completion algorithms have the problems of insufficient utilization of local geometric information and weak learning ability of the offset of the point cloud generation model.To solve this problem,a Skip DGCNN feature extraction network based on skipconnections is proposed to share local and global geometric information for the prediction of shape details.Then,a TSMB network inference module based on Taylor approximation is proposed,and a theoretical model of offset learning is given.Based on this model,a TSMB module that can be directly used for network training is designed.Then,Skip DGCNN and TSMB are integrated into the design of point cloud completion network to form a PST-Net neural network structure.Next,Tank Bottom,a small sample dataset,is constructed for tank bottom point cloud completion to train and tune PST-Net to complete the shape of the missing tank bottom point cloud.After that,a calculation software for tank volume and bottom volume is designed to realize tank point cloud visualization,volume measurement convenience,and volume statistics automation.The experiment shows that the error rate of calculated bottom volume is only 0.14‰,and the average error rate of volume calculation is only 0.85 ‰ after the point cloud at tank bottom is completed by PST-Net neural network.(3)Aiming at the problems that the path planning method based on the artificial potential field method is easy to fall into local minima,the accuracy of the existing positioning technology inside the tank is limited,and the defect feature analysis is insufficient,a spiral path design method based on geometric description is proposed to adapt to the defect detection of the circular tank bottom.Then,four improved methods of the artificial potential field method are proposed,which are combined with the improved repulsion function to effectively avoid the local minimum trap.Then,a three-point localization algorithm based on sonar ranging is proposed to ensure that the robot moves in the tank according to the path planning results.In addition,a defect detection method based on magnetic flux leakage signal is proposed.The extracted components of magnetic flux leakage signal are determined by simulation analysis of magnetic flux leakage distribution.Moreover,the defect location and relative size are determined by signal strength distribution.The experiments show that the improved path planning method can effectively avoid the local minimum trap to complete the planning and obstacle avoidance.In addition,the average positioning error is only 0.0748±0.0032.On the platform of the tank bottom detection robot system,the magnetic flux leakage signal shows the obvious defect location and has the relative size resolution. |
PDF Full Text Request