The processing quality of the grooves of a nuclear-fuel rod will directly affect the quality of the finished nuclear-fuel rod. Due to the highly reflective, microscopic, and annular characteristics of nuclear-fuel rod grooves, it has been quite challenging to realize imaging and microscopic defect detection for these grooves. In this work, a machine vision-based defect detection system was developed for nuclear-fuel rod grooves. Through the performance improvement and application of the self-reference template defect detection method, efficient online inspection of nuclear-fuel rod grooves was realized. In the developed system, a combined-light-source imaging system was first designed by combining a coaxial light and a ring light, which realized the clear imaging of a groove. After that, an image expansion strategy was employed to expand the annular groove into a strip-shaped region of interest (ROI). Then, according to the turning processing characteristic of the nuclear-fuel rod groove, the large-size defect detection effect of the self-reference template method was improved by eliminating the anomalous columns prior to generating the self-reference template. The experimental results indicated that the average inspection efficiency of the developed system was 8.026 s/rod, the average false detection rate was 0.183%. The accuracy of the self-reference template method was 87.6%, higher than that of YOLOv2 and Faster R-CNN. The developed system exhibits high inspection efficiency and accuracy, so it can meet the actual detection functions and requirements of production lines, and now it has been successfully applied to actual production.

核燃料棒凹槽的加工质量将直接影响成品核燃料棒的质量。由于核燃料棒凹槽的高反射性，微观和环形特性，实现这些凹槽的成像和微观缺陷检测非常具有挑战性。在这项工作中，为核燃料棒凹槽开发了基于机器视觉的缺陷检测系统。通过性能改进和自参考模板缺陷检测方法的应用，实现了核燃料棒槽的高效在线检测。在开发的系统中，首先通过组合同轴光和环形光来设计组合光源成像系统，从而实现了凹槽的清晰成像。在那之后，采用图像扩展策略将环形凹槽扩展成带状的感兴趣区域（ROI）。然后，根据核燃料棒凹槽的转弯加工特性，通过在生成自参考模板之前消除了异常列，提高了自参考模板方法的大尺寸缺陷检测效果。实验结果表明，所开发系统的平均检查效率为8.026 s / rod，平均误检率为0.183％。自参考模板方法的准确性为87.6％，高于YOLOv2和Faster R-CNN。开发的系统具有很高的检测效率和准确性，可以满足生产线的实际检测功能和要求，