- Promotor: Prof. Geert Verdoolaege
- Supervisor: Nastasija Petkovic
- 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, like tokamaks and stellarators.
A major challenge in tokamak devices is managing the extreme heat loads on the divertor (exhaust), which can damage wall components. Traditionally, infrared (IR) cameras are used to measure divertor heat flux, but they require complex post-processing and lack the ability to provide real-time predictions for control applications.
This thesis proposes an innovative method based on machine learning (ML) for predicting divertor heat flux profiles from electric probe measurements at the midplane, as well as global plasma conditions. By using interpretable ML models, this research will also provide help understanding the influence of the machine operational conditions on the heat flux profile at the divertor. This data-driven approach could improve tokamak operation and control of high heat loads on the divertor.
Illustration of the COMPASS tokamak (IPP Prague).
(a) Midplane measurements as input for the ML model, (b) neural network, (c) output: heat flux profiles.
Objectives
The key goals of this research are to:
- Develop a machine learning model that predicts the magnitude and shape of the divertor heat flux profile from upstream plasma measurements taken from the reciprocating probes in the scrape-off layer (SOL), as well as global plasma parameters. Of specific interest will be the peak heat flux, power spreading parameter and power fall-off length.
- Compare different machine learning models, starting with simple regression models and proceeding to neural networks, to predict heat flux parameters in order to find the most suitable model.
- Use machine learning interpretability techniques (such as SHAP values, feature importance and correlation analysis) to reveal how global plasma parameters and parameters at the midplane affect divertor heat flux.
- [optional] Explore the feasibility of using this method for real-time tokamak control and monitoring by leveraging time-series machine learning methods such as RNN or transformers.
The study will use experimental data from the COMPASS tokamak (IPP Prague), including:
- Langmuir and ballpen probe measurements in the midplane around the separatrix, providing the floating potential, plasma potential and ion saturation current. From these measurements, electron temperature and density can be calculated.
- Stationary plasma parameters, such as plasma current, line-averaged density, heating power, magnetic field, safety factor, plasma shape parameters, etc.
- Divertor electric probe measurements, used as reference data for validating heat flux predictions obtained from the machine learning models.
(a) Reciprocating probe head. (b) Langmuir and ballpen probe measurements.