Neural Operator-Based Surrogate Model for CFD:Helical Coil Steam Generator in Small Modular Reactor
Minseo Lee, Seongmin Oh, Chaehyeon Song, Bumjin Cho, Shilaj Baral, Sangam Khanal, Minseop Song, Joongoo Jeon
Read on arXiv →Key claim
Multi-scale L-DeepONet captures vortex dynamics effectively.
This study introduces a novel framework that combines reduced-order models with neural operators for real-time thermal-hydraulic simulation in small modular reactors. The multi-scale L-DeepONet effectively captures vortex dynamics, while the Fourier neural operator provides reliable pressure drop estimates. This work offers a practical guideline for model selection based on specific objectives and data types.
In plain English
The authors developed a new framework that combines reduced-order models with neural operators to improve real-time thermal-hydraulic simulations for small modular reactors. Unlike previous methods that struggled with the high computational costs of detailed fluid dynamics simulations, this approach allows for faster and more efficient analysis of complex systems like helical coil steam generators. The multi-scale version of their model, called L-DeepONet, effectively captures the dynamic behavior of swirling flows, while another model, the Fourier neural operator, provides accurate estimates of pressure changes. This advancement is significant for builders because it enables safer and more efficient reactor operations by allowing for quicker decision-making based on reliable simulations. Understanding these models can help builders select the right tools for their specific simulation needs.
The integration of neural operators with reduced-order models for CFD-level transient analysis is a significant advancement in the field.
The study compares multiple ROM strategies and provides a practical model-selection guideline, supporting its claims with experimental results.