Temperature Field Reconstruction of Tungsten Monoblock Divertor on EAST using Physics-aware Neural Operator Transformer

2026-06-30 · Source: Artificial Intelligence · Field: Science & Research — Artificial Intelligence & Machine Learning, Engineering & Applied Sciences, Physical Sciences & Chemistry · Depth: Expert, quick

Summary

A Physics-aware Neural Operator Transformer (PNOT) has been developed to enable real-time temperature field reconstruction of tungsten monoblock divertors on the EAST fusion device. This innovation addresses the computational expense of conventional numerical methods like the Finite Element Method (FEM), which are unsuitable for real-time applications in fusion devices, leading to potential material melting and damage. PNOT models boundary heat-flux relations as a structured graph, employing graph attention to explicitly capture spatial physical dependencies. It further integrates a physics-aware neural operator module to aggregate query points with similar physical conditions via slicing and model heat diffusion. A gradient-constrained Sobolev regularization loss is also applied to enforce consistency between function values and their derivatives. Experimental results indicate that these physical constraints enhance prediction accuracy while maintaining physical consistency, making it suitable for real-time control. The source code will be released on https://github.com/Event-AHU/OpenFusion.

Key takeaway

For Research Scientists developing real-time control systems for complex physical phenomena, you should consider integrating Physics-aware Neural Operator Transformers. This approach offers superior accuracy and physical consistency compared to traditional numerical methods. It enables faster decision-making and prevents material damage in critical applications like fusion devices. Your team can utilize the open-source PNOT code to accelerate development of robust, real-time predictive models.

Key insights

Physics-aware neural operators can accurately reconstruct complex spatiotemporal fields in real-time, outperforming traditional methods.

Principles

• Explicitly model physical dependencies.
• Aggregate points with similar physical conditions.
• Enforce consistency with gradient constraints.

Method

PNOT models heat-flux relations as a structured graph, uses graph attention for spatial dependencies, and a physics-aware neural operator for heat diffusion, applying Sobolev regularization.

In practice

• Apply PNOT for real-time control.
• Improve fusion device service life.
• Use graph attention for physical relations.

Topics

• Tungsten Divertor
• EAST Fusion Device
• Temperature Field Reconstruction
• Physics-aware Neural Operator
• Real-time Control
• Graph Attention Networks

Code references

• Event-AHU/OpenFusion

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

Related on AIssential

• Towards Data-Efficient Cross-Device Generalization of Grad-Shafranov Equilibria via Transfer Learning Neural Operator — Neural operator pretraining enables data-efficient, real-time, cross-device generalization of Grad-Shafranov equilibria for magnetic confinement fusion.
• Material-Agnostic Zero-Shot Thermal Inference for Metal Additive Manufacturing via a Parametric PINN Framework — A parametric PINN framework enables zero-shot thermal modeling across diverse materials in metal AM with high efficiency.
• LithoDreamer: A Physics-Informed World Model for Multi-Stage Computational Lithography — LithoDreamer is a physics-informed World Model for multi-stage computational lithography, achieving state-of-the-art performance.
• Physics AI research that’s shaping the industry. - Mistral AI — Physics AI, utilizing neural surrogates and transformers, advances industrial engineering simulations.
• Learning Physical Operators using Neural Operators — A physics-informed framework improves neural operator generalization and temporal continuity for PDE solving via operator splitting.
• Physics-Informed Machine Learning for Pouch Cell Temperature Estimation — PIML integrates physics equations into neural network loss for accurate, efficient temperature estimation in battery cells.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.