中法核工程与技术学院核声论坛（总第235期）

讲座摘要：

This lecture focuses on the integration of artificial intelligence with advanced nuclear thermal hydraulics, addressing three key challenges in complex two-phase flow systems: difficult extraction with limited data, unstable AI simulation, and uncertain CHF safety-boundary prediction.

First, for intelligent observation and data augmentation, a YOLO-based bubble detection and tracking framework incorporating attention mechanisms and multi-object tracking is introduced to automatically extract bubble morphology, trajectories, and thermal-hydraulic parameters from two-phase flow experiments. A physics-constrained BF-GAN model is further developed to generate high-fidelity two-phase flow images under prescribed operating conditions, thereby expanding the available experimental data space.

Second, for AIphysics-informed simulation, a PINNs-based framework integrating adaptive residual sampling, dynamic loss weighting, and temporal domain decomposition is proposed to improve the stability and interpretability of AI-based fluid dynamics and transient two-phase flow simulations.

Third, for CHF safety-boundary prediction, deep learning, Transformer, and transfer learning models are developed for critical heat flux prediction. These models improve cross-condition and cross-geometry generalization, supporting more reliable assessment of thermal safety margins in advanced nuclear systems.

Overall, this research establishes an integrated route from intelligent observation and data augmentation to physics-informed simulation and CHF safety-boundary prediction. It provides data foundations, modeling tools, and safety-criterion support for trustworthy digital twins and reactor safety assessment in advanced nuclear energy systems.

主讲人简介：

Wen ZHOU is a Project Researcher at the Department of Nuclear Engineering and Management, the University of Tokyo. His research focuses on artificial intelligence for fluid mechanics, two-phase flow, thermal hydraulics and nuclear engineering, covering data-driven AI, physics-informed AI, generative AI, and intelligent safety assessment of advanced nuclear systems.