The midterm objective of global fusion plasma physics research is to design and construct a fusion power plant capable of producing electricity for the grid. Such a future fusion power plant will generate energy by utilizing deuterium-tritium plasma heated to 100 million degrees Celsius. However, it is crucial that this plasma, which is contained within a magnetic field, does not accidentally come into contact with the metal walls of the fusion reactor. Another important consideration is that the components designed for helium ash removal, known as divertors, must not be subjected to excessive heat loads, as they could melt.

Infrared cameras can be used to determine the temperature and monitor the time evolution of the temperature on these elements. However, it would be advantageous to predict temperature rise well before it becomes critical.  While cameras operating in the visible range cannot directly measure temperature, an increase in light intensity can be the first indication of plasma-material interaction under certain conditions. Therefore, it is important to investigate and understand the physical interplay between the change of brightness caused by the particle transport and the temperature increase of the plasma facing components (PFCs). By using recording from both visible and infrared cameras observing the same part of the plasma facing componentsPFCs, a neural network could be trained to predict the parts that are about to heat up before the infrared cameras can detect them, thus providing an early warning of potentially dangerous high-temperature events.

The student's task is to develop an AI architecture that can be optimally trained to predict these temperature-rising events from visible camera data, using infrared data for training purposes. To achieve this, data from Wendelstein 7-X, the world’s largest stellarator device, is available for both visible and infrared systems. The student should identify the best architecture for this purpose by training multiple versions, analyzing, and validating the results. Using this architecture, the students should investigate the predicting capabilities of the model developed on the operational scenarios of Wendelstein 7-X.