Accelerated Main-Ion Charge Exchange Recombination analysis leveraging Machine Learning

The development of a neural network (NN) based framework for Main-Ion Charge Exchange Recombination (MICER) data analysis on DIII-D has enabled faster and more comprehensive insights into plasma behavior. This framework supplements the time-consuming FIDASIM Monte Carlo collisional radiative code traditionally used to correct profile measurements obtained from MICER. The MICER diagnostic utilizes charge exchange between neutral beams and the main ion species (typically deuterium) to measure temperature, velocity, and density profiles, providing crucial data for magnetic confinement fusion research. Despite acquiring raw data on almost every DIII-D shot over the last decade years, only high priority experiments are presently processed due to the complex and extensive FIDASIM-based analysis required to accurately account for atomic physics effects. To streamline this process, a database based on the parameter space of DIII-D pedestals was created, and FIDASIM simulations were used to generate training data for the neural network.Once trained, the NN corrects MICER measurements to within 5% accuracy of FIDASIM, eliminating the need for extensive simulations for initial observations. This automated analysis greatly expands MICER’s availability and accessibility for fusion research.