Machine learning based spectral fitting and atomic physics corrections for the main ion charge exchange recombination spectroscopy system on the DIII-D tokamak (invited).

Rapid and accurate machine learning (ML) based analysis of the Doppler broadened and shifted Balmer-α spectra from the main ion charge exchange recombination spectroscopy (MICER) diagnostic on the DIII-D tokamak has been developed to provide raw measurements of the main ion (typically D+) temperature, toroidal rotation, density, and beam emission based measurements directly from the active Balmer-α spectra. The temperatures and toroidal rotations are typically within 150 eV and 10 km/s of those obtained from the standard least squares fitting approach. The raw measurements are then corrected for systematic errors using a second set of ML models that were trained as inverse surrogates for the FIDASIM three dimensional collisional radiative model. FIDASIM provides a comprehensive forward model of the raw measurements, including atomic physics and spatial smearing effects. At the pedestal top, corrections to the ion temperature are typically 150 eV but can be up to 300 eV. The standard deviation of the difference between the surrogate and the full model based correction is less than 50 eV, with some of this difference due to Monte Carlo noise from the FIDASIM model. 20 000 FIDASIM simulations spanning the relevant DIII-D parameter space were used for the training and validation of the ML model, incorporating variations in ion and electron profiles, as well as variations in the neutral beam injector parameters and magnetic equilibrium mapping. The complexity of the MICER spectra and associated corrections has historically limited analysis to high priority discharges, despite the wealth of information contained in the spectra. These advances have dramatically reduced the analysis time for single discharges from several hours to seconds. The speed-up now allows bulk analysis of thousands of historical discharges acquired by the diagnostic during its 12+ years of operation on DIII-D, enabling a variety of studies, such as those involving H-mode main ion pedestal profiles.