Neural Operator-Based Surrogate Model for CFD:Helical Coil Steam Generator in Small Modular Reactor

2026-05-28 · Source: Machine Learning · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Computational Fluid Dynamics · Depth: Expert, quick

Summary

This study introduces an integrated framework for real-time thermal-hydraulic simulation in Small Modular Reactor (SMR) digital twin applications, specifically targeting the helical coil steam generator (HCSG) of the System-integrated Modular Advanced Reactor (SMART). The framework combines reduced-order models (ROM) with neural operators to overcome the high computational cost of traditional CFD. Researchers compared two ROM strategies: an MLP-based autoencoder for unstructured mesh data and a convolutional autoencoder for structured mesh data, both coupled with DeepONet to create latent DeepONet (L-DeepONet). A Fourier neural operator (FNO) was also evaluated. A multi-scale technique was integrated into both frameworks to enhance Kármán vortex street prediction. The multi-scale L-DeepONet accurately captured instantaneous periodic vortex dynamics, whereas the FNO and its multi-scale variant provided reliable time-averaged mean flow and pressure drop estimates. This work, published on 2026-05-28, offers model-selection guidelines based on CFD data type and required flow resolution.

Key takeaway

For Machine Learning Engineers developing digital twin solutions for Small Modular Reactors, this research offers a clear path to real-time CFD simulation. You should integrate reduced-order models with neural operators; L-DeepONet captures instantaneous flow dynamics, while FNO provides reliable time-averaged pressure drop estimates. Tailor your model choice based on the specific CFD data type and the required flow resolution for your application. This approach significantly reduces computational overhead.

Key insights

Neural operator-based surrogates, combined with ROMs and multi-scale techniques, enable real-time CFD for SMR digital twins.

Principles

• Neural operators can surrogate CFD for transient analysis.
• Multi-scale techniques mitigate spectral bias in flow prediction.
• Model selection depends on CFD data type and resolution.

Method

Integrated framework combines ROMs (MLP-AE for unstructured, CAE for structured) with DeepONet to form L-DeepONet, or uses FNO. Multi-scale technique incorporated.

In practice

• Use L-DeepONet for instantaneous vortex dynamics.
• Use FNO for time-averaged flow and pressure drop.
• Tailor ROM strategy to CFD mesh data type.

Topics

• Neural Operators
• CFD Surrogate Models
• Small Modular Reactors
• Digital Twin Technology
• DeepONet
• Fluid Dynamics Simulation

Best for: AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• Accelerate Clean, Modular, Nuclear Reactor Design with AI Physics — AI-augmented simulations with surrogate models significantly accelerate nuclear reactor design by overcoming computational bottlenecks.
• Ten Years of Trying to Learn from a Mesh That’s Too Large — Large CFD meshes pose representational challenges for surrogate models, requiring specialized architectures to capture localized physical features.
• Droplet-LNO: Physics-Informed Laplace Neural Operators for Accurate Prediction of Droplet Spreading Dynamics on Complex Surfaces — PI-LNO uses Laplace transforms and physics-informed neural networks to model droplet spreading dynamics efficiently.
• FLAME 2023: Diving into the Future of Fluid Dynamics and Machine Learning — AI techniques are transforming simulation by integrating physics-based modeling with data-driven machine learning.
• Deep neural operator for free boundary problems — The Free Boundary Neural Operator (FBNO) solves PDEs on evolving, unknown domains by utilizing a fixed conjugate system.
• Physics AI research that’s shaping the industry. - Mistral AI — Physics AI, utilizing neural surrogates and transformers, advances industrial engineering simulations.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning.