Seeing Through the Gaps: AI for Super-Resolution and Diagnostic Recovery in Fusion Devices

In this talk, we present a breakthrough in fusion diagnostic capabilities through the development of Diag2Diag, a machine-learning framework that synthesizes super-resolution Thomson Scattering (SRTS) data at MHz frequencies. This method enables real-time, single-event capture of entire Edge Localized Mode (ELM) cycles within a single discharge on DIII-D—illuminating key phases of ELM onset, crash, and recovery without the need for hardware upgrades. We will show that such model can be trained and adapted using low computation resources and generates synthetic diagnostics in real-time. Leveraging correlated diagnostic signals, Diag2Diag also delivers the first direct experimental evidence of pedestal flattening caused by Resonant Magnetic Perturbation (RMP)-induced magnetic islands near the plasma boundary. Beyond physics insights, we discuss the application of this approach to address the challenge of diagnostic failure, allowing reconstruction of critical data with high fidelity validated against ECE and interferometry. Additional applications to present and discuss include (1) super-resolved density tracking in pellet-fueled plasmas that enables detailed tracking of the rapid, localized density changes induced by pellet injection, and (2) investigating the discrepancies in electron temperature measurements between ECE and TS by closing the gaps between the temporal resolution of ECE and TS measurements on different tokamaks such as DIII-D, WEST. These results represent a transformative step toward cost-effective, robust diagnostics and stability control strategies for ITER and future reactors. By enhancing our ability to observe and interpret fast edge dynamics in high-performance plasmas, Diag2Diag paves the way for optimized fusion operation and sustained energy production.