Development Of Groove Defect Detection System For Nuclear Fuel Rods Based On Deep Learning

Under the background of carbon emission and carbon neutralization,as a clean and high energy density energy,nuclear power has a very broad market prospect,and the demand for nuclear fuel rods has also increased sharply.The groove quality of nuclear fuel rods after processing directly affects the welding quality of nuclear fuel rods and the risk of nuclear leakage in future operation,which has a very important impact on the safe and reliable operation of nuclear power plant.Three years ago,the research team developed the first-generation groove defect detection system of nuclear fuel rods based on machine vision,which greatly improved the detection accuracy and efficiency.However,with the further growth of the demand for nuclear fuel rods,enterprises have put forward higher requirements for efficiency of defect detection,and it is urgent to develop a second-generation groove defect detection system of nuclear fuel rods that takes into account both high efficiency and accuracy.Based on this,the second-generation detection system is developed in this paper,and the effectiveness of the system is verified.The main research work of this paper is as follows:(1)Based on an overview analysis of the groove defects of nuclear fuel rods and the technical indicators of the second-generation groove defect detection system,the efficiency and feasibility of the first-generation groove defect detection system from imaging,auto focusing and groove defect detection algorithm are analyzed respectively.An auto focusing algorithm based on a priori model is designed,which can realize auto focusing by collecting only one image,and the effectiveness and efficiency of the algorithm are verified by experiments.(2)Aiming at the current situation that the first-generation groove defect detection algorithm takes a long time,a two-stage defect detection algorithm is designed.In the first stage,traditional image processing and different deep learning image classification methods are used to realize groove segment filtering,and the effectiveness of the methods is compared.In the second stage,in the process of abnormal fragment determination based on template matching,corresponding improvement methods are proposed for different problems.The two-stage defect detection algorithm meets the requirements of enterprise technical indicators in defect detection rate(recall rate)and detection efficiency.(3)The defect detection algorithm of nuclear fuel rods groove based on target detection network is deeply studied.Try to use Rep VGG to replace the feature extraction network of YOLOv5,and build an improved Rep VGGYOLOv5 detection network;YOLOv5,Rep VGG-YOLOv5 and Faster R-CNN networks are used to detect the groove defects of nuclear fuel rods respectively,and the advantages of the network are obtained.(4)Based on the architecture of the upper computer software of the first-generation groove defect detection system,a more flexible programming architecture based on message queue management mode is selected to realize the compilation of the upper computer software of the second-generation groove defect detection system of nuclear fuel rods,and the corresponding functions are developed and realized.