Research On Underwater Positioning Method Of Refueling Mechanism Based On Machine Vision

Nuclear energy is a high-radiation and clean new energy source,commonly used in fields such as nuclear power generation,medical treatment,radiographic testing,and seawater desalination,with broad application prospects.The refueling mechanism designed in this thesis mainly loads the nuclear fuel assembly into the nuclear reactor,so that the core can operate normally for nuclear power generation.In this process,accurate positioning of underwater nuclear fuel assembly is the key technology for the normal refueling of nuclear reactors.At present,traditional underwater fuel precise accurate location has the problems of high labor cost,visual recognition,low degree of automation,and low positioning accuracy.In recent years,machine vision technology has been widely used in underwater fuel assembly positioning,but it still faces some problems,such as spatial mismatch,scarce data set,poor underwater image quality,low working efficiency,and low positioning accuracy.Based on this,because of the above existing problems,this thesis carries out the research on the underwater positioning method of refueling mechanism based on machine vision based on artificial intelligence technology,improves the work efficiency,positioning accuracy,and intelligence level,and achieves the accurate positioning requirements of the underwater nuclear fuel assembly.The main research tasks of this thesis are as follows:(1)Aiming at the problems of spatial misalignment and data scarcity of underwater and air data sets,this thesis proposes an underwater pin image data set generation method based on deep convolution adversarial neural network.In this method,the codingdecoding structure and attention mechanism are used to form a generation network model to generate pin images in different environments;then the fully connected structure is used to construct the discriminator network model to distinguish the true and false pin images;finally,the antagonistic training strategy of discriminator and generator is used to improve the generation ability of the generated network model.In addition,a positioning experimental platform based on machine vision was built,and the ability of network model generation was evaluated by FID and MMD indexes.Experimental results show that this method can realize the correction of the spatial position of the underwater and air pin data set and the expansion of the data set,and has a good practical effect.(2)Aiming at the problem of poor quality of underwater images,this thesis proposes an underwater image restoration method based on physical priori.Firstly,the structure restoration network model and physical prior model are used to restore the structure of underwater images;then the multi-scale attention network model is used to correct the color of the underwater structure restoration images.In addition,this thesis uses residual jumper connection for feature map fusion to enrich feature information,and uses PSNR,SSIM and UIQM indexes to evaluate the quality of the generated image.Experiments show that this method can realize structure restoration and color correction of underwater images,and obtain clear images.(3)Aiming at the problems of low efficiency and poor positioning accuracy,a selfattention mechanism-based underwater fuel assembly positioning method is proposed in this thesis.Firstly,the pin detection lightweight network model is used to select the pin area quickly,that is,pin rough positioning;then subpixel edge extraction algorithm is used to extract the pin edge coordinate points;finally,the nonlinear least square algorithm is used to fit the edge points of the pin,and the center coordinates of the pin are obtained,that is,the pin precise positioning.In addition,this thesis uses recall rate and accuracy index to evaluate the pin detection lightweight network model.Experimental results show that the proposed method can accurately identify pins with various shapes,extract pin center coordinates and speed up detection efficiency,to realize underwater fuel assembly positioning operation and meet the actual requirements.(4)This thesis designs an underwater fuel assembly positioning system based on machine vision.According to the intelligent hardware facilities in the real scene,this thesis develops a set of easy-to-operate software systems based on the industrial scene.The software system is mainly used for algorithm integration,real-time positioning display,control diagram drawing,and emergency control,which greatly improves the intelligent operation of the underwater fuel assembly positioning system.The system test shows that the software system can make the underwater fuel assembly positioning accuracy within 0.1mm,which meets the practical demand.