蒸汽发生器传热管中的高速气液两相流是高温气冷堆的关键热工水力状态，研究传热管中两相流状态对确保反应堆运行安全、提升工作效率至关重要。现有测量方法无法满足对小尺寸合金钢管内的高速两相流的流型和空泡份额的检测需求。本文提出了基于静态CT的高速两相流测量方法。针对传热管内两相流的测量要求，给出了静态CT系统的具体设计方案，研究了影响系统成像质量的关键参数，提出了适用该系统的稀疏角度CT重建算法和空泡份额计算方法。 本研究首先针对被检测传热管，设计了静态CT布置方案，并通过数值仿真验证了该系统对传热管内气液两相流实现断层成像的可行性。之后采用仿真实验的方法分析系统参数对图像质量的影响，针对两相流的气泡分辨目标，确定了射线源管电压和管电流、探测器单元尺寸等关键参数。 为获得更贴近真实情况的模拟投影数据，本文搭建GPU加速的蒙特卡洛仿真平台，为后续设备研制提供数据支持。通过仿真实验，明确了多源散射射线对系统图像质量影响较小。确定了系统曝光时间为1ms，并通过流动状态下的仿真得出系统对流速在1m/s以下气液两相流的断层图像能实现较好的气泡观测效果，流速不超过3m/s的泡状流实现气泡识别。 受到设备尺寸和系统空间布局的限制，上述系统可采集的投影数量不超过30，传统算法的重建图像存在严重的伪影。本研究设计了无监督的深度学习算法DRP（Deep Radon Prior），通过优化重建图像与投影数据在Radon域中的误差，利用神经网络对图像规律信息的学习能力，结合迭代算法的思想，直接从投影数据中重建出图像，适用于无法获得完备投影的实际应用场景。该算法有效抑制了图像的伪影和噪声，重建质量显著优于FBP及ADMM-TV算法，能实现对直径0.3mm小气泡的显示。 针对空泡份额测量，将其转化为分割问题，分析空泡份额的数学模型，推导流动状态下的权重矩阵与非线性因子，并验证时均非线性的影响可以忽略。采用TransUNet框架，并结合气泡连通域先验条件提出一种后处理模块，进一步提升分割网络的各项性能指标，根据图像分割结果计算空泡份额的相对误差小于7%。 本研究的相关成果为高速两相流测量提供静态CT这一新方法。根据高温气冷堆传热管的两相流测量需求，明确静态CT系统的具体设计方案，为后续实际设备的研发工作提供参考依据。

The high-speed gas-liquid two-phase flow in the heat transfer tubes of steam generators is a critical thermal-hydraulic state of High Temperature Gas-cooled Reactor. Investigating the two-phase flow state in the heat transfer tubes is essential to ensure safe operation and improve work efficiency of the reactor. However, conventional measurement methods are inadequate to meet the detection requirements for flow pattern and void fraction of high-speed two-phase flow inside a small-sized alloy steel tube. In this study, a high-speed two-phase flow measurement method based on static CT is proposed. A specific design for the static CT system is provided in accordance with the measurement requirements for the two-phase flow inside the heat transfer tube. The key parameters that influence the imaging quality of the system are studied, and a sparse-view CT reconstruction algorithm and a void fraction calculation method applicable to the system are proposed. A static CT layout is designed for the heat transfer tube, and numerical simulation is used to verify the feasibility of the system for achieving tomographic imaging of gas-liquid two-phase flow in the heat transfer tube. Subsequently, simulation experiments are conducted to analyze the influence of system parameters on image quality, and the key parameters are determined for bubble recognition target of the two-phase flow, such as the voltage and current of the X-ray source tube and the size of the detector unit. To obtain more realistic simulated projection data, a GPU-accelerated Monte Carlo simulation platform is developed, which provides data support for subsequent researches. Simulation experiments reveal that scattered ray of other X-ray sources have minimal impact on the system‘s image quality. The system‘s exposure time is determined to be 1ms, and simulations under the flow state indicate that the system can achieve good observation of bubbles in the tomographic images of gas-liquid two-phase flow with a flow rate below 1m/s and bubble recognition in bubble flow with a flow rate not exceeding 3m/s. Due to the limitations of equipment size and system spatial layout, the number of projections of the system does not exceed 30, and the reconstructed images of traditional algorithms exhibit serious artifacts. To tackle this issue, an unsupervised sparse-view CT reconstruction algorithm, Deep Radon Prior (DRP), is designed, which directly reconstructs the image from projection data by optimizing the error between the reconstructed image and the projection data in the Radon domain. This algorithm combines the learning ability of neural networks for image regularity information with an iterative algorithm and effectively suppresses image artifacts and noise, providing a significantly better reconstruction quality than that of FBP and ADMM-TV algorithms. This algorithm is able to display bubbles as small as 0.3mm in diameter. This study transforms the void fraction measurement into a segmentation problem and analyzes the mathematical model of void fraction. The weight matrix and nonlinear factor under the flow state are studied, and it is verified that the influence of time-average nonlinearity can be ignored. For the segmentation of bubbles, the TransUNet framework is utilized and a post-processing module is proposed to further improve the performance of the segmentation network due to the connected domain prior of bubbles. The relative error of void fraction calculated based on the image segmentation results is less than 7%. In conclusion, this study presents a novel method, static CT, for high-speed two-phase flow measurement. The specific design scheme of the static CT system is clarified based on the measurement requirements for gas-liquid two-phase flow in High-Temperature Gas-cooled Reactor heat transfer tubes, providing a reference basis for subsequent development of practical equipment.