| The spherical tank is widely used in industrial production as a pressure gas storage device.Regular maintenance of the weld area of the tank is an important guarantee for safe production.The traditional maintenance method is to set up temporary scaffolding inside and outside the storage tank,and the workers climb the scaffold for maintenance.This maintenance method has low efficiency and high cost.In this context,domestic and foreign researchers have developed many tank weld detection robot systems,but most of these robots are detected by manual remote control,and the detection effect depends on the experience and technology of the operator.Therefore,it is an inevitable trend for the development of non-destructive testing of storage tanks to develop a detection system for intelligent identification and autonomous navigation of welds to improve the efficiency and quality of tank maintenance.In this context,this paper carried out the following related research work:Firstly,the overall design of the weld navigation system of the wall-climbing robot is carried out,including the construction of the hardware platform and the development of the software system.In this paper,the semantic segmentation model based on deep learning is used to identify and extract the weld path.The navigation problem of the complex environment of the tank is simplified to the visual patrol tracking problem,which significantly reduces the difficulty of image processing and navigation control.The navigation system is designed and the software and hardware system are developed.The functions of robot image acquisition,remote monitoring of ground station,communication between robot and ground station are realized.On this basis,a graphical operation interface is designed to simplify the operation process of the navigation system.Secondly,the image data set of weld seam on the surface of storage tank is made and the Seg Former semantic segmentation model is improved.Using the image acquisition system and manual shooting method,the weld image of the tank to be detected was collected,labeled and data enhanced,and the weld image data set of the tank surface was produced.Aiming at the problem that the wall-climbing robot is difficult to identify the weld on the storage tank,which leads to the low efficiency of weld detection,this paper proposes a weld image semantic segmentation model based on Seg Former,combined with Patch Expanding up-sampling method and recursive cavity self-attention mechanism.The experimental results on the weld image dataset show that the improved weld segmentation model is superior to the mainstream segmentation model,and the average intersection over union and average pixel accuracy are improved by 2.41 % and 1.37 % respectively compared with the baseline algorithm model.Then,the weld path is extracted and the kinematics and deviation correction model of the wallclimbing robot are analyzed,and the weld navigation method is designed.The mask image generated by the improved model after identifying the weld is used to extract the center line of the weld as the weld path,and the offset and angle of the robot relative to the weld path are calculated.The motion model of the wall-climbing robot is analyzed,and the PID control technology is used to control the robot to correct the deviation and realize the weld navigation.Finally,the weld navigation experiment was carried out on the simulated tank and the real tank,and the experimental data were analyzed.An indoor simulated tank scene was built and the navigation system was tested to verify the correctness of the kinematics and weld deviation correction model of the wall-climbing robot.Finally,the weld navigation test was carried out on the real 1000m3 storage tank,which verified that the weld recognition and control system had good response ability and the real-time and accuracy of weld recognition on the real storage tank.The experimental results show that the error of weld autonomous navigation is kept between ± 5mm,which meets the requirements of tank maintenance. |
PDF Full Text Request