Integrated data analysis using Bayesian probabilistic methods for real-time plasma diagnosis and control in DEMO

Promoter: Prof. Dr. Geert Verdoolaege, Ghent University

Home university: UGent

Inference of plasma conditions from diagnostic measurements can often benefit from an integrated analysis, combining measurements from multiple diagnostics. This can be achieved in a Bayesian probabilistic framework, which allows determining the probability distribution of the physical quantities of interest that is compatible with all available data. The approach, sometimes referred to as integrated data analysis (IDA), therefore takes care of data fusion and error propagation at the same time, incorporating constraints on the inferred quantities by means of prior distributions. In fusion demonstration power plants (DEMO), the number of diagnostics will be limited due to space restrictions to maximize fusion power. Furthermore, on top of the usual errors and uncertainties accompanying measurements in fusion plasmas, several diagnostics in DEMO will suffer from additional sources of uncertainty due to the harsh conditions (neutron flux). The role of joint diagnostic analysis and uncertainty estimation will therefore be critical in DEMO. Moreover, in DEMO the uncertainty of the plasma response to actuators will also need to be quantified and all analysis will have to be carried out in real time.

This PhD topic will be the first that is directly aimed at IDA for DEMO. The goal is to design and implement an innovative strategy for sensor fusion and uncertainty quantification toward real-time plasma diagnosis and control in DEMO. Bayesian inference will be used as a probabilistic framework, allowing the study of uncertainty propagation through complex, nonlinear models that describe the physics of the measurement process and the plasma response. In addition, the framework will allow contribution to optimization of the design of machine components, diagnostics and actuators, in an early stage. Concretely, it is envisaged to target estimation and control in the following two specific cases:

• Plasma current and shape based on magnetic measurements by inductive coils and Hall probes;
• Electron density profiles based on interferometry and reflectometry.

Hence, the PhD will focus on some of the most fundamental plasma quantities and their related diagnostics in a fusion device, which directly impact the plasma equilibrium and the fusion performance. This will require models for the diagnostics, including the primary sources of uncertainty. In addition, models will be constructed of the relevant actuators, such as control coils, gas puffing and pellet injection. Furthermore, techniques will be implemented to carry out the inference of measured and controlled quantities in real time. This will require suitable approximation techniques of the posterior distributions and/or the entire inference process. Sensitivity studies will be carried out to identify the design characteristics of diagnostics and actuators, as well as the main uncertainty sources, that have the greatest impact on the measurement and control of the plasma quantities of interest. The developed techniques will be tested on fusion devices like ASDEX Upgrade and COMPASS Upgrade. If still feasible within the scope of the project, dedicated optimization techniques aimed at Bayesian experimental design will be used for design optimization. Finally, proposals will be made toward integration of the IDA approach for diagnostics and actuators into the control strategy for DEMO.