- Promotor: Prof. Geert Verdoolaege
- Mentor: Jeffrey De Rycke
- Study programs: Master of Science in Engineering Physics, Master of Science in Physics and Astronomy, Master of Science in Teaching in Science and Technology (Physics and Astronomy), European Master of Science in Nuclear Fusion and Engineering Physics
- Location: Technicum, at home
Problem setting
The worldwide research on controlled thermonuclear fusion aims to realize a source of energy on earth that is limitless, clean and safe. It is a large-scale enterprise that mimics the power source of the stars. Ghent University is involved in fusion R&D by developing data science methods and numerical models for understanding and predicting the complex science and technology of fusion devices based on magnetic plasma confinement.
The first European demonstration fusion reactor (EU-DEMO) will require highly accurate and real-time measurements to ensure optimal operation and plasma control. Currently, our research unit employs Bayesian inference for integrated data analysis (IDA) using measurements by magnetic coils to reconstruct the current flowing through the plasma. This is a crucial piece of information for ensuring efficient and safe plasma operation. However, we face two key limitations:
- Lack of internal plasma information: Magnetic diagnostics are insufficient for accurately reconstructing the full current density and magnetic flux distribution.
- Computational bottlenecks: The Bayesian inference takes in the order of seconds, making it unsuitable for real-time plasma control.
Right: cross-section of a DEMO design with multiple fusion diagnostics—pick-up coils, flux loops, and saddle coils—used to infer plasma current, centroid position, and shape.
Objectives
To address these challenges, this thesis will focus on improving our present Bayesian code for current tomography. Specifically, the following goals are envisaged:
- Improving plasma current reconstructions by integrating measurements from additional diagnostics;
- Optimizing sensor placement using Bayesian experimental design (BED);
- Accelerating inference methods through approximation techniques and machine learning.
Enhanced plasma reconstruction will be achieved by extending the Bayesian IDA framework to incorporate internal plasma diagnostics. This includes the use of interferometry and polarimetry for electron density and internal magnetic information, as well as electron cyclotron emission for electron temperature and internal magnetic data. The integration of these additional diagnostics is expected to enhance current density accuracy and reduce uncertainties in plasma boundary estimation. Furthermore, physical relationships between different plasma parameters will be incorporated, for instance, by using the Grad-Shafranov equation to improve consistency in the reconstructions.
To optimize real-time performance, the research will explore faster Bayesian inference techniques. This involves employing lower resolution inference methods, combining two parallel low- and high-fidelity integrated estimates, and developing neural network surrogates to approximate the complex Bayesian models efficiently. These approaches are aimed at significantly reducing the computational time, making real-time plasma reconstruction feasible.
Additionally, Bayesian experimental design will be applied to determine the optimal sensor arrangement for DEMO, maximizing information gain while respecting reactor constraints. This includes performing sensitivity analysis and investigating sensor failure scenarios to ensure robust diagnostic systems. By leveraging these advanced probabilistic and computational methods, the research aims to provide more accurate and faster plasma reconstructions, contributing to improved control and safety of future fusion reactors.