Description
Plasma tomography is a notoriously ill-posed problem, due to the sparse-view and limited-angle diagnostic coverage. Bayesian inference provides a statistically sound framework for addressing this challenge, naturally enabling multi-diagnostic approaches and allowing uncertainty quantification. In this contribution, we show that Bayesian statistics offers a unifying perspective on the most widely employed plasma tomography techniques [2]. We describe how modern Bayesian computational imaging techniques – including advanced sampling strategies and data-driven approaches – can be used to achieve critical physical insights from plasma tomography. We discuss quantitative insights into important radiation features of X-point target and doublet plasmas, which are currently the focus of intense research at the TCV tokamak. We then focus on real-time applications of our Bayesian tomography framework. Specifically, we address inference from TCV’s real-time capable bolometry system. We present a newly developed technique that enables efficient and robust estimation of the power radiated by arbitrary plasma regions of interest from bolometry measurements [3]. These estimates provide essential information for real-time plasma control, enabling feedback on the total, core and divertor radiated power. We demonstrate this capability in radiated power and divertor detachment control experiments at TCV.
[1] B. P. Duval and TCV Team, 2024 Nuclear Fusion 64 112023
[2] D. Hamm et al., 2025 Plasma Physics Controlled Fusion 67 115012
[3] D. Hamm et al., to be submitted to Nuclear Fusion