The deep radon prior-based stationary CT image reconstruction algorithm for two phase flow inspection.

Area of Science:

• Nuclear Engineering
• Fluid Dynamics
• Medical Imaging

Background:

• Accurate measurement of two-phase flow is vital for reactor safety and efficiency.
• Existing methods struggle with high-velocity flow in small pipes, limiting void fraction and flow pattern identification.
• Stationary computed tomography (CT) offers a potential solution but faces challenges with sparse data acquisition.

Purpose of the Study:

• To develop and validate a novel method for measuring high-velocity two-phase flow in small-diameter pipes.
• To address the limitations of sparse data in CT-based flow measurements.
• To improve the visualization and analysis of two-phase flow patterns and void fractions.

Main Methods:

• Proposed an unsupervised deep learning algorithm, Deep Radon Prior (DRP), for image reconstruction from sparse projection data.
• DRP optimizes errors in the radon domain, integrating neural network learning with iterative algorithms.
• Compared DRP's performance against traditional Filtered Back Projection (FBP) and ADMM-TV algorithms.

Main Results:

• DRP significantly suppressed image artifacts and noise, outperforming FBP and ADMM-TV.
• Achieved superior image reconstruction quality, enabling visualization of 0.3 mm bubbles.
• Demonstrated the algorithm's effectiveness with sparse projection data inherent in the CT system's physical constraints.

Conclusions:

• The DRP algorithm provides a robust and effective solution for high-velocity two-phase flow measurement in small pipes.
• DRP enhances visualization capabilities for small flow structures, crucial for safety and efficiency.
• The algorithm shows broad applicability for various fluid flow patterns and bubble flow measurements.