A novel technique for in situ calibration of the C-2W electromagnetic neutral particle analyzer utilizing machine learning.

In TAE Technologies' current experimental device, C-2W, neutral beam injection creates a large fast ion population that sustains a field-reversed configuration (FRC) plasma. Diagnosis of these fast ions is therefore critical for understanding the behavior of the FRC. Neutral Particle Analyzers (NPAs) are used to measure the energy spectrum of fast ions that charge exchange on background or beam neutrals and are lost from the plasma. To ensure correct diagnosis of the fast ion population, a calibration check of the NPAs was performed. A novel, generally applicable method for an in situ relative calibration of diagnostics on an unknown source with a small dataset was developed. The method utilizes a machine learning technique, Generalized Additive Models (GAMs), to reconstruct the diagnostic source distribution, and Stochastic Gradient Descent (SGD) to determine the NPA channel calibration factors. The results on both synthetic and experimental datasets are presented.