Robust impurity detection and tracking for tokamaks.

Area of Science:

• Nuclear Fusion Engineering
• Plasma Physics
• Machine Learning Applications

Background:

• Tokamak devices are crucial for fusion energy research.
• Impurity particles, like dust, can negatively impact tokamak performance and diagnostics.
• Accurate detection and tracking of these impurities are essential for operational control.

Purpose of the Study:

• To develop and validate a machine learning-based code for detecting and tracking dust particle tracks from tokamak camera footage.
• To assess the code's performance using data from multiple major tokamak facilities.

Main Methods:

• The study presents a machine learning code designed for robust impurity detection and tracking.
• The code processes camera footage to generate large sets of dust tracks.
• Performance was evaluated using data from Joint European Torus, Doublet-III-D, and Magnum-PSI.

Main Results:

• The code achieved a high dust track classification accuracy, ranging from 65% to 100%.
• It can detect up to approximately 1000 dust particles from a single camera image.
• The machine learning approach demonstrated effectiveness across different tokamak environments.

Conclusions:

• The developed code offers a powerful tool for analyzing dust impurities in tokamaks.
• Future work should focus on expanding training datasets and addressing selection biases for improved performance.
• The code's adaptability allows for potential application in stereoscopic reconstruction and tracking of non-dust impurities.