Description
The upcoming ST40 campaign will include a series of experiments testing the performance of various PFC (Plasma Facing Component) materials including both lithium and molybdenum. Part of assessing the effectiveness of the PFCs will be measuring the edge to core transport of the first wall material which requires an understanding of the plasma composition. Since no single diagnostic can precisely measure the impurity content there is a need for an integrated analysis that can encompass all the available diagnostic data. Bayesian inference is the natural choice for an integrated analysis framework as it can robustly capture the effect of both modelling uncertainties and measurement noise in the probabilistic model.
This poster will discuss a workflow called BDA (Bayesian Diagnostic Analysis) that has been used to infer the probabilities of the likely plasma compositions for a series of shots in the most recent ST40 campaign [1]. BDA uses Bayesian Optimisation as the workhorse for performing model inference over a wide range of input parameters. The diagnostic systems that were used include Thomson Scattering, two SXR (Soft X-Ray) cameras, two bolometers, and a CX (Charge Exchange) spectrometer for measurement of the carbon 6+ density as well as the background bremsstrahlung. Forward modelling of the diagnostic systems was done with the InDiCA (Integrated DiagnostiC Analysis) python library [2]. The transport of core impurities was included with Aurora [3]. The effect of neutrals is also investigated given the relatively small size of ST40 and the absence of strong electron temperature pedestals.
[1] Michael Gemmell 2025, Bayesian Optimisation as a Tool for Solving the Inverse Problem with Plasma Diagnostics [Poster presentation] EPS
[2] Marco Sertoli et al 2024 Plasma Phys. Control. Fusion 66 095011
[3] Sciortini F. et al 2021 Plasma Phys. Control. Fusion 63 112001