Research On Path Planning Strategy Of Oil And Gas Pipeline Robot Based On Obstacle Image Perception

As the main transportation channel for crude oil,gas and other materials,oil and gas pipelines are widely used in chemical industry,nuclear industry and many other fields,and play an important role in people’s production and life.However,in the long-term operation,the oil scale and other obstacles in the pipe will gradually accumulate and become more serious with the increase of the age,which will cause the blockage and corrosion of the pipeline,and seriously affect the normal operation of the oil and gas pipeline.In order to prevent the blockage and corrosion of the pipeline,the oil and gas pipeline robot came into being.Oil and gas pipeline robots have the advantages of safety and intelligence,and have gradually become one of the key equipment for pipeline inspection.In oil and gas pipelines lacking light,how to improve the detection effect and path planning accuracy of robots has become the focus of research.Based on the existing results,this dissitation studies the image detection algorithm and path planning strategy of oil and gas pipeline robots.Firstly,the reconstruction effect of image super-resolution is improved by improving the structure of attention network.Secondly,a higher precision target detection model is proposed to detect the position of obstacles in the image,and further measure the distance of obstacles from the robot.Finally,according to the position of obstacles in the oil and gas pipeline,the existing path planning strategy is optimized so that the oil and gas pipeline robot can reach the target position more quickly and accurately according to the planned path.The main research contents of this dissitation are summarized as follows:(1)To improve the resolution of obstacle images captured by robots in oil and gas pipelines lacking light,this dissitation proposes a Structured Fusion Attention Network(SFAN)to achieve image super-resolution(SR)reconstruction.Through the structured processing of attention module,the improvement of residual module,the information enhancement of channel merging,and the design of high-frequency residual module,the super-resolution reconstruction effect of in-pipe images is improved.The comparison with other SR models verifies that SFAN can reconstruct higher resolution and maintain fewer model parameters(only 1.6M).The actual experiment also proves that SFAN can effectively improve the resolution of obstacle images in the pipeline.(2)The existing object detection and ranging methods in images have problems such as lack of local spatial modeling ability and insufficient edge detection,which affect the detection effect and ranging accuracy of obstacles in oil and gas pipelines.To this end,a high-precision detection model(RODFormer)using Transformer for rotating obstacle targets and an improved SGBM(CSGBM)algorithm based on the improved canny operator are proposed.RODFormer is used to improve the effect of object detection in obstacle images.RODFormer uses a structured architecture to improve the collection range of image obstacle feature information.The constructed space-FFN model solves the problem that the multi-layer perceptron lacks the ability to model in local space,and can effectively alleviate the loss discontinuity of the rotating frame.The CSGBM algorithm is used to improve the accuracy of stereo matching,its improved focus mainly on using Canny operator instead of horizontal Sobel operator to overcome the defects of traditional operator in the edge measurement,using a bilateral filtering instead of gaussian filtering to reduce the loss of boundary,using the standard Euclidean distance instead of the Euclidean distance to balance the scale between different dimensions of data.Compared with 10 rotating target detection methods and actual experiments,the high-precision effect and generalization performance of RODFormer for obstacle target detection in pipeline images are proved.Compared with SGBM,it is proved that the image matching time and false matching rate of the CSGBM algorithm are reduced(30.867 s and 8.83%,respectively),and the ranging accuracy between the robot and the obstacle is controlled within 0.5%.(3)Aiming at the low efficiency of path planning caused by random sampling of robots in oil and gas pipelines,this dissitation proposes a path planning strategy for robots in oil and gas pipelines is proposed,including the improved B-RRT*(RB-RRT*)algorithm,the improved artificial potential field(APF)method and the PID controller with improved RBF neural network.RB-RRT* improves the search efficiency of B-RRT*algorithm through the proposed adaptive sampling adjustment method and fast search algorithm.The introduction of the potential field factor in the APF method can effectively avoid unnecessary repulsion of the robot during the obstacle avoidance process so that the robot can avoid possible obstacles.The improvement of the RBF neural network combined with the PID controller improves the control ability of the robot path planning so that the pipeline robot can follow the planned path through the pipe as much as possible.The experimental results show that the above improved algorithm can effectively improve the efficiency and effect of the robot’s path planning and control in the pipeline.The research results of this dissitation provide theoretical and key technical support for the visual image detection and path planning control of oil and gas pipeline robots,and have important application value for the detection,cleaning and maintenance of oil and gas pipelines.