Machine learning driven correlation studies: Alfvénic and sub-Alfvénic frequency chirping at NSTX

Magnetic perturbations in a very broadband range (<30 kHz to >1 GHz) are commonly measured on tokamaks such as NSTX by using Mirnov coils. The spectral behaviour can be categorised as quiescent, fixed-frequency, chirping, or avalanching. Here, ‘chirping’ modes experience a time-dependent frequency shift due to non-linear effects – in some cases, multiple plasma modes chirp in a near-concurrent fashion (mode 'avalanching').

Mode avalanching is typically correlated with fast ion loss. However, transition to this phase of mode behaviour is not fully understood. Human characterisation of mode behaviour is labour intensive - studying dependences of plasma parameters on mode character proves difficult.

Here, results are presented from machine learning driven studies of correlations between different mode character and weighted averages of plasma parameters obtained from TRANSP (e.g., v-fast/v-Alfvén and β-fast/β-Alfvén). The weighted averages allow for correlations to be drawn (i.e. shear versus mode character). These results yield similar correlations to previous work by Fredrickson et al. [1]. An overall framework is presented to utilise this tool for generic tokamaks, for possible future use on MAST-U and DIII-D.

 [1] E. D. Fredrickson et al. 2014 NF 54, 093007