Research On Automatic Calibration Method Of High Shielding Core Load

Since nuclear reactors change their loads regularly,in order to ensure the safe operation of the reactor,it is necessary to accurately determine the types and the installation locations of newly replaced loads.At present,the proofreading of nuclear power reactor loads mainly relies on manual work(multi-person,multi-group,batch proofreading),and the low proofreading work efficiency(a single shift takes an average of 5 hours)seriously affects the production efficiency of military experiments.In civil use,the production value of civil nuclear power plants is about 5 million yuan per hour when the single reactor is shut down for operation and maintenance.Automatic load calibration can greatly improve the production efficiency of civil nuclear power plants.Therefore,the safety proofreading of reactor operation and maintenance is of great significance.Various intelligent methods can greatly improve work efficiency in the manufacturing and operation and maintenance stages,and machine vision can also be applied to the intelligent proofreading of the loads in the heap.In view of the serious occlusion of tunnels and control rods,it is difficult to directly obtain the complete information of the bottom core target through a single perspective.A accessible method to figure out the issue is to identify and splicing the target information from multiple perspectives.However,underwater scenes from various perspectives suffer from the large parallax and high occlusion,strong similarity of target image features,and serious underwater thermal disturbance.Existing image recognition and stitching technologies are tough to solve such complex working conditions.In response to the above issues,a method of automatic core loads proofreading based on machine vision is proposed,and a set of automatic core loads proofreading system is designed and implemented.The main research content of this article is as follows:1)A method for accurately identifying the type of core load in a thermal disturbance environment is proposed.An image based target detection network is used to output the load identification results of sequential frames,and then a clustering algorithm is used to cluster the load identification results of each sequential frame to obtain the final load category identification results.2)A method for reconstructing the distribution and classification of core loads is proposed.In this paper,The proofreading method first establishes global and local virtual two-dimensional(2D)coordinate mapping models according to the distribution relationship of the installation positions of the loads.Then,during identification,each partial sequence diagram is photographed,the position of the loaded object in the partial picture is identified by the time series feature aggregation algorithm,the local virtual2 D coordinate mapping model is calibrated by the position of the loaded object,and the position of the loaded object in the partial picture is consistent with the calibration.Afterwards,the position coordinates of the local virtual 2D coordinate mapping model are then judged by spatial geometric distance,the position labels and categories corresponding to the loads are screened out,and the distribution of the load types is reconstructed.3)A reliable panoramic mosaic method for core images is proposed.The panoramic stitching method in this paper is based on the prior information in the category reconstruction results.According to the reconstruction results(such as the same load in the middle image and the surrounding image),coarse positioning is achieved,and then combined with a rotating template matching algorithm to achieve fine positioning of the load.At the same time,the Deeplab v3+deep learning network is used to remove the obstructions.The segmented images from each perspective are used and a matching algorithm for coarse positioning and fine positioning is used,Using the central image as a benchmark,the surrounding image is combined with it to obtain a panoramic core mosaic image.The panoramic mosaic image and the category reconstruction results are mutually corroborated to achieve the purpose of intelligent recognition and manual proofreading.Experimental research shows that the proofreading method in this study can effectively detect the type and installation position of the load,with a load recognition rate of better than 99%.The panoramic stitching result is stable and reliable,with a matching speed of less than 1 second,a matching accuracy of better than 2.5 pixels,and a matching accuracy of more than 96%.Category reconstruction and panoramic stitching are mutually supportive,which can achieve a 100% accuracy rate for core load proofreading,and have great application potential in core load proofreading.