Filling in the Gaps: AI for Super-Resolution and Diagnostic Recovery

The Science

Tokamaks are donut-shaped devices that hold plasma within magnetic fields. Currently, scientists use them to research fusion. These devices will need to evolve for companies to use them for commercial fusion energy. Artificial intelligence (AI) is helping scientists make this leap. One effort is incorporating AI into plasma monitoring with measurement systems, called diagnostics. Diagnostics must function within the extreme environment required for fusion in a tokamak. These conditions put challenging constraints on the device's diagnostics. Diagnostics can also have limited resolution. These issues can prevent diagnostics from completely capturing rapid plasma events. To address this problem, researchers at the DIII-D National Fusion Facility (an Office of Science User Facility) developed an AI model. This model identifies correlations among existing diagnostics. It then produces synthetic data to recreate what a limited resolution or failed diagnostic would see at ultra-high resolution. This model created super-resolution synthetic data that captured the full dynamics of critical instabilities that occur in milliseconds, revealing physics that were invisible to current diagnostics.

The Impact

The Diag2Diag AI model provides a new commercially relevant tool for fusion devices. It can synthetically recover or enhance plasma measurements. By reducing the need for redundancy, this approach provides a fail safe for cost-effective fusion device designs. In addition, future fusion power plants will have limits on the number of sensors they can have. Diag2Diag's ability to reveal correlations among diagnostics will help scientists identify the minimum set of measurement systems that fusion power plants will need. These features are helping scientists harness the power of AI to enable detailed, reliable, and cost-effective monitoring of fusion plasmas. This advancement will accelerate progress toward commercial fusion.

Summary

Future commercial fusion devices will rely on a limited set of measurement systems to monitor plasma behavior and inform operational decisions. The incorporation of AI into tokamak diagnostics will help meet this need. AI makes it possible to produce higher resolution synthetic measurements and develop a deeper understanding of plasma physics. An international research team led by scientists at the DIII-D National Fusion Facility recently developed a new AI model, named Diag2Diag. This model ensures fusion devices can keep operating safely and efficiently despite sensor limitations or failures. It uses AI to fill in missing data and sharpen measurements from other existing sensors.

This model identifies correlations among multiple sensors, which allows it to infer outputs for diagnostics that fail or measure too slowly to completely capture a dynamic event. Diag2Diag generated synthetic data for the Thomson Scattering diagnostic. This is a key diagnostic that measures the electron profile but does not capture the full dynamics of plasma instabilities. Researchers used Diag2Diag to reconstruct large instabilities at the plasma edge, called edge localized modes. It produced super-resolution synthetic Thomson Scattering data. These data confirmed how magnetic structures in the plasma suppress edge localized modes, which is an important insight for the control strategies planned for future fusion devices. This study provides proof-of-concept evidence for the use of AI to augment fusion diagnostics, particularly in commercial designs. Furthermore, this approach has the potential to be expanded to other fields where missing data could likewise be detrimental, such as aerospace exploration or medical sensing.

Contact

Azarakhsh JalalvandPrinceton [email protected]

Egemen KolemenPrinceton [email protected]

Funding

This work was supported by the U.S. Department of Energy, Office of Science, Office of Fusion Energy Sciences, using the DIII-D National Fusion Facility, a DOE Office of Science User Facility. Additional support was provided by the Princeton Laboratory for Artificial Intelligence, as well as the National Research Foundation of Korea.

Publications

Jalalvand, A., et al., "Multimodal super-resolution: discovering hidden physics and its application to fusion plasmas." Nature Communications 16, 8506 (2025) [DOI: 10.1038/s41467-025-63492-1]

Related Links

• New AI enhances the view inside fusion energy systems, Princeton Plasma Physics Laboratory newsroom
• New AI technology revolutionizes visualization inside fusion energy systems, Bioengineer.org