Enabling data-driven NTM studies with advanced mode labeling

We present the application of automatic Neoclassical Tearing Mode (NTM) labeling on Alcator C-Mod and DIII-D. Previous work on DIII-D has labeled onset of 2/1 modes in multiple ways, including extrapolating backwards from maximum N1RMS [1] and training a classifier on a set of human-labeled shots before applying to all shots. NTMs on Alcator C-Mod are rare by comparison, so we leverage computer vision to segment spectrograms from high-frequency Mirnov probes [2]. Data fetching and spectrogram calculation is done using the open-source DisruptionPy library [3]. If an NTM is identified, we perform spectral analysis combining multiple probes to infer the mode’s structure, and use filament tracing to calculate perturbed current and island width [4]. These detailed labels will enable data-driven approaches to prevent NTMs, such as cross-machine database analysis for designing safe operating scenarios and real-time trajectory optimization with model predictive control.

[1] L. Bardóczi et al 2023, Nucl. Fusion 63 126052

[2] E.D. Zapata-Cornejo et al 2024, Plasma Phys. Control. Fusion 66 095016

[3] Trevisan et al, Zenodo (2024) 10.5281/zenodo.13935223

[4] R. Sweeney et al 2017, Nucl. Fusion 57 016019