Synthetic magnetic diagnostic integration on SPARC and C-Mod for MHD mode identification

This work presents progress towards automatic MHD mode identification tools for the SPARC tokamak, with a focus on Alfven Eigenmodes. Using existing high-frequency magnetic signals from Alcator C-Mod, we benchmark machine-learning classifiers against the well-known approach of fitting toroidal mode number n against the temporal vs geometric phase difference between toroidally separated Mirnov probes. For the ML identifier, we present results from on a “hybrid” model, using an image segmentation model trained on synthetically generated spectrogram images to identify regions of interest to perform the aforementioned n# fitting. Finally, progress is presented on building a fully ML classifier trained on synthetic magnetic signals using known probe locations and a reduced model of nearby conducting structures. The latter allows the calculation of realistic confounding eddy currents via the ThinCurr[1] code (part of the OpenFusion toolkit). We present signals generated from analytically defined “filament” models of the MHD modes vs simulations from FAR3D and M3D-C1. The comparison to existing C-Mod data helps validate the corresponding predictions for SPARC. Progress towards future work integrating additional diagnostic input such as synthetic ECE signals into the ML model is presented.

[1] C. Hansen et. al, 2025, Comp. Phys. Comm.