Detection Method of X-ray Fuel Rod End Plug Defect Based on Deep Learning

Amidst the global expansion of nuclear power generation, ensuring the integrity of nuclear fuel rods is crucial for the safe operation of nuclear power plants. As a vital component of fuel rods, the detection of defects in the end plugs is a key step in ensuring nuclear safety. Traditional manual detection methods are not only time-consuming and inefficient but also susceptible to subjective influences. To address these issues, this study proposed an automatic detection method for defects in fuel rod end plugs based on deep learning X-ray imaging, aiming to enhance the accuracy and efficiency of detection. The research began by collecting a large number of X-ray images of fuel rod end plugs and preprocessing these images, including single-rod segmentation and extraction of effective evaluation areas, to optimize image quality. Subsequently, an improved YOLOX model was adopted as the core detection algorithm, with adjustments made to the network structure and loss function to address the characteristics of small target defects. The introduction of a coordinate attention module enables the model to more accurately locate and identify tiny defects. Additionally, the CIoU loss function was employed in place of the traditional IoU loss function to improve the model’s localization precision for small targets. During the model training phase, data augmentation techniques such as Mosaic, Copy and Paste, and Mixup were implemented to enhance the model’s adaptability to new scenarios. The experimental results demonstrate that the improved model excels in the task of end plug defect detection, with significant enhancements in detection accuracy and speed compared to traditional methods and unimproved deep learning models. Tests on industrial datasets show a notable increase in the model’s mean average precision (mAP) while maintaining a fast detection speed, meeting the requirements of actual production. The model also performs well in detecting various types of defects, including accurate identification of porosity, swelling, incomplete welding, tungsten inclusion, and plug abnormalities. In summary, this study successfully develops an efficient and accurate automatic detection method for defects in fuel rod end plugs based on deep learning X-ray imaging. This method not only improves the level of detection automation but also provides strong technical support for the safe management and maintenance of fuel rods. Future research will continue to explore the potential for model optimization to better adapt to a wider range of industrial applications.