Multimodal super-resolution: discovering hidden physics and its application to fusion plasmas

Area of Science:

• Physics
• Plasma Physics
• Machine Learning

Background:

• Complex physical systems require integrating data from multiple diagnostics with varying resolutions.
• Edge-localized modes (ELMs) in fusion plasmas pose risks to plasma-facing materials.
• Existing diagnostics may suffer from limited resolution, coverage, or measurement failures.

Purpose of the Study:

• To develop a machine learning framework for multimodal super-resolution of diagnostic data.
• To reconstruct high-temporal-resolution synthetic data for Thomson Scattering diagnostics.
• To investigate the dynamics of edge-localized modes (ELMs) and the role of resonant magnetic perturbations (RMPs) in their suppression.

Main Methods:

• A machine learning framework was developed to reconstruct synthetic high-temporal-resolution data for a target diagnostic using information from other diagnostics.
• The technique, termed multimodal super-resolution, does not require direct target measurements during inference.
• The framework was applied to fusion plasma data, specifically reconstructing Thomson Scattering data from complementary diagnostics.

Main Results:

• The framework successfully generated high-fidelity synthetic Thomson Scattering data with high temporal resolution.
• Fine-scale plasma dynamics related to edge-localized modes (ELMs) were uncovered.
• The role of resonant magnetic perturbations (RMPs) in suppressing ELMs via magnetic island formation was validated, supporting plasma profile flattening observations.

Conclusions:

• The multimodal super-resolution framework enhances diagnostic robustness and enables monitoring even with degraded measurements.
• The approach provides valuable insights into ELM dynamics and RMP-induced suppression mechanisms.
• This technique is broadly transferable to other scientific domains with sparse or incomplete data, aiding in future fusion reactor development like ITER.