【摘要】 钼及其合金具有优异的高温性能和独特热膨胀系数，是极端环境下的关键战略材料。然而，在制备中，由于工艺问题会在材料表面形成凸起缺陷。针对该缺陷，传统测量多采用接触式方法，虽操作简便，但受材料硬度与表面形貌影响。提出基于FF-XRFI的钼片表面平整度测量方法，利用荧光成像中的几何阴影效应，实现表面凸起高度测量。实验对0.2～0.7 mm厚度的钼片凸起进行测试，结果显示图像阴影宽度可用于反演凸起高度，但因阴影边界模糊，导致测量高度与实际高度相对偏差较大，其相对误差范围是7.62%～47.81%。为提高测量精度，采用多级混合算法优化图像。通过对比不同插值算法，将图像空间分辨率从40.54μm减小至31.14μm;进一步优化基于像素强度与梯度的双重约束平滑算法参数，在获得稳定峰值信噪比的同时，将空间分辨率进一步提高至25.67μm;最终结合霍夫变换统计阴影像素，建立厚度-像素线性模型，将测量相对误差缩小至3.4%～14.5%。更多还原

【Abstract】 Molybdenum and its alloys possess outstanding high-temperature performance and unique thermal expansion coefficients, making them key strategic materials in extreme environments. However, during the manufacturing process, due to technological issues, protrusion defects form on the material’s surface. Traditional measurement methods for these defects are mostly contact-based, which are simple to operate but are affected by the material’s hardness and surface morphology. This paper proposes a method for measuring the surface flatness of molybdenum sheets based on FF-XRFI, which utilizes the geometric shadow effect in fluorescence imaging to measure the height of surface protrusions. Experiments were conducted on protrusions of molybdenum sheets with thicknesses ranging from 0.2 to 0.7 mm. The results showed that the width of the image shadow could be used to infer the height of the protrusions, but due to the blurred shadow boundaries, the relative deviation between the measured height and the actual height was relatively large, with a relative error range of 7.62% to 47.81%. To improve measurement accuracy, a multi-level hybrid algorithm was used to optimize the image. By comparing different interpolation algorithms, the image spatial resolution was reduced from 40.54 to 31.14 μm; further optimization of the dual-constraint smoothing algorithm parameters based on pixel intensity and gradient, while maintaining a stable peak signal-to-noise ratio, further improved the spatial resolution to 25.67 μm; finally, by combining the Hough transform to statistically analyze the shadow pixels and establishing a thickness-pixel linear model, the relative measurement error was reduced to 3.4% to 14.5%.更多还原