2026年“核理安邦”联合博士生学术论坛

基于X射线正交双投影的圆形管材截面参数在线检测系统研究

Not scheduled

1h

地址：清华大学校内

海报展示 海报展示 海报展示

Speaker

煜昌 孙

Abstract

The geometric cross-sectional parameters of circular tubes including inner and outer diameters, wall thickness, and eccentricity are critical indicators for manufacturing quality evaluation. Existing X-ray digital imaging-based inspection methods typically rely on the localization of edge transition points (ETPs) in attenuation curves. However, their performance is fundamentally limited by the detector’s physical pixel size and noise, leading to insufficient localization accuracy and, consequently, degraded precision and stability in parameter inversion. To address these limitations, this paper proposes a novel cross-sectional parameter reconstruction framework that integrates detector sub-pixel motion with a closed-loop residual-driven optimization (CRO) algorithm. First, a sub-pixel detector motion scheme is introduced to enhance the effective sampling resolution of the projection signal. Analytical simulation results demonstrate that, compared with the no-motion case (detector pixel size of 0.2 mm), reducing the motion step size to 0.1 mm, 0.05 mm and 0.04 mm decreases the mean absolute error (MAE) of the inner contour major axis by 38.10%, 40.01%, and 47.25%, respectively. Furthermore, to mitigate the influence of noise and other practical disturbances, a closed-loop iterative optimization algorithm based on residual sign feedback is developed. In the proposed CRO framework, the residual between measured projections and analytically computed projections from a parametric contour model is evaluated at the ETPs to guide parameter updates, enabling iterative refinement of contour parameters. Monte Carlo (MC) simulation results demonstrate that taking a motion step size of 0.04 mm as an example, the introduction of the CRO algorithm reduces the MAEof inner contour major axis by 69.44%, with a maximum MAE of only 0.03 mm. In summary, the proposed approach provides an efficient and paratically feasible solution for high-accuracy industrial online inspection.

摘要

圆形管材（如电缆和胶管）的几何截面参数（内外径、壁厚及偏心）是表征制造质量的关键指标。现有基于X射线数字成像的检测方法通常依赖于射线衰减曲线中边界突变点的定位，然而受限于探测器物理像素尺寸及系统噪声，导致定位精度不足并进一步影响参数反演的准确性与稳定性。针对上述问题，本文提出一种结合探测器亚像素微动与闭环差值驱动优化（Closed-loop residual-driven optimization, CRO）的截面参数重建方法。首先，通过引入亚像素级探测器微动采样策略，获取多帧位移采样数据并进行拼接，从而在不改变物理像素尺寸的前提下提高投影信号的等效采样分辨率，实现突变点的亚像素级定位。解析仿真结果表明，相较于无微动情况（探测器像素为0.2 mm），当微动步长分别缩小至0.1 mm、0.05 mm和0.04 mm时，内轮廓长轴参数的平均绝对误差（MAE）分别降低了38.10%，40.01%和47.25%。进一步地，针对实际测量中存在的噪声等干扰，本文构建了一种基于突变点处差值符号反馈的闭环迭代优化算法（CRO）。该方法以测量投影与参数化轮廓模型解析投影在突变点处的差值为反馈信号，通过差值的符号驱动更新策略对轮廓参数进行修正，从而提升轮廓参数的精度并增强算法对实际干扰因素的鲁棒性。Monte Carlo（MC）仿真数据表明（以0.04 mm微动步长为例），引入CRO算法优化后，轮廓参数的MAE下降了69.44%，最大误差为0.03 mm。综上，本文提出的协同硬件与软件算法的方案在不依赖复杂成像重建的前提下，实现了对圆形管材截面参数的高精度反演，为工业在线检测提供了一种高效且具有工程可行性的解决方案。