Research On Recognition And Location Algorithm Of Nuclear Fuel Rod Based On Machine Vision

As a clean,low-carbon and efficient energy,nuclear energy has been attracting much attention.With the development of nuclear power,demand for the standard fuel rod bundles(heavy water reactor fuel assemblies)used in CANDU-6 reactors has soared.The fuel assembly of heavy water reactor is made by adjusting different types of nuclear fuel rods to correct positions,intubating and fixing,and then welding end plates.My research group developed the first generation of automatic nuclear fuel rod type recognition and position positioning equipment by using machine vision instead of manual labor.In order to improve the working efficiency of the equipment and the positioning accuracy of the nuclear fuel rod,we plan to develop the second generation of nuclear fuel rod type identification and positioning equipment and algorithm.The second generation of equipment adopts the form of dual stations,the first station is responsible for the type identification and initial positioning of nuclear fuel rods,and the second station is responsible for the precise positioning of nuclear fuel rods.In this paper,the image processing algorithm for two stations in the second generation nuclear fuel rod type recognition and location equipment is studied.The main research work of this paper is as follows:(1)Research on traditional image processing algorithm for nuclear fuel rod type Recognition and initial location.Firstly,the mechanical center is obtained by calibration of the center.Then the image is spliced and the complete end face of the isolation block is obtained.Then,binarization,morphological processing and Canny edge extraction were used to get clear and noise-less contour map of isolated block.Then,the waveform is smoothed and the features are extracted to obtain the isolation block information and realize the identification and initial location of the nuclear fuel rod type.It takes an average of 0.812 s and has an average accuracy of 3.013° to process images of 1500*1500 pixels.(2)Research on YOLO algorithm for Nuclear fuel Rod type Identification and Initial Location.In order to improve the efficiency of the algorithm,enhance the robustness of the algorithm and save the camera cost,we try to use YOLO target retrieval algorithm.In order to remove the interference of the support pad,the initial position images and rotating position images are splicing as training and testing samples,and the splicing images of nuclear fuel rods at various positions are selected as far as possible,so as to better deal with enviro nmental interference and other problems in the image collection process,so as to increase the robustness of the network.Respectively,YOLOV3 and YOLOV4 target retrieval algorithms were used to identify the type of nuclear fuel rods.Through comparison and analysis of detection results,YOLOV4 target retrieval algorithm was selected.Early at the same time to reduce the image stitching error,and improve the positioning precision of calculating base isolation block in machinery center as the origin of coordinate system in the practical point of view,counterclockwise to standard position Angle calculation,if the base isolation is a patchwork,then find the nearest base isolation piece of isolation,as the benchmark,calculates the fuel rods at the beginning of the orientation Angle.The experimental results show that the average time of YOLOV4 algorithm is 0.379 s and the initial positioning accuracy is 3.146°,which is 0.03 s faster and 0.884°higher than the traditional image processing algorithm when processing 256*256pixel images.(3)Research on the algorithm of precise location of nuclear fuel rod.Taking the type of nuclear fuel rod at the first station and the location information of the isolation block as priori information,the reference isolation block area was quickly located,and the center correction algorithm was proposed to get the actual center of the detected nuclear fuel rod.In order to remove the noise interference of the support pad at both ends of the nuclear fuel rod and get the accurate Angle of the reference isolation block,a threshold optimization method based on prior information was proposed.Due to the equipment error,the precise positioning method of multiple rotation was adopted.The experimental results show that the average accuracy of the algorithm is 0.098° and the average detection time is 0.7060 s.The positioning accuracy is 0.037° higher than that of the first generation.The efficiency and precision of the second generation algorithm meet the requirements of the second generation equipment.