Description
Information density of plasma diagnostics is becoming increasingly important given the competition for reactor wall space with tritium breeding blankets critical to sustainable fusion power plants. Ion Cyclotron Emission (ICE) may enable a versatile plasma diagnostic due to its ubiquity in magnetised fusion devices and high technical readiness having been detected on Langmuir probes and ICRH antennae. ICE measurements have been promoted as a candidate to inform advanced alpha-channeling techniques through energetic ion spatial and velocity distribution information. However, its potential as a diagnostic is dependent upon accurate modelling of plasma parameters against experimentally-analogous emission spectra.
The underlying mechanism for ICE is the Magnetoacoustic Cyclotron Instability (MCI), a kinetic effect arising from a resonance between the gyromotion of a non-thermal fast ion population and a fast Alfven wave propagating predominantly perpendicular to B0. Simulating the MCI requires a high temporal resolution kinetic code, such as Particle-In-Cell (PIC), running for >10 ion cyclotron periods to realise the nonlinear regime of the instability which would be observed by experimental diagnostics. This presents a computational challenge even in the one-dimensional case given constraints on the numerical parameters needed to resolve MCI behaviour while suppressing unphysical heating. This work presents the generation of a dataset of 100 simulations simultaneously scanning reactor-relevant ranges of background magnetic field strength, density, and alpha particle velocity pitch (v-parallel/|v|) and concentration in the fully kinetic 1D3V regime. Finite 2D parallel wavenumber is included by running each simulation at four propagation angles close to perpendicular. Simulations are performed in EPOCH, a well-established PIC code for this type of problem.
Power spectra of magnetic and electric field components exhibiting MCI-like behaviour are used to train specialised Time-Series Extrinsic Regression (TSER) algorithms from Aeon, a time-series analysis toolkit. This novel application of TSER demonstrates Tier 3 HPC resources can produce a dataset capable of inferring B0, density, fast fusion or NBI ion velocity pitch and concentration with an R2 of at least 70-90%.