Using Deep Learning for the Detection of UFOs within the JET Tokamak
Components on the inner wall of fusion reactors are exposed to extremely high heat loads, and significant care must be taken to minimise the damage to these components during operation. UFOs, also known as Transient Impurity Events (TIEs), are small particles of dust within the tokamak vessel which can lead to plasma disruptions. These can cause serious damage to the device and remain a significant challenge for safe operations. This study presents a novel approach to track these UFO events by utilising visual camera data from the Joint European Torus (JET) and a Convolutional Neural Network (CNN) to identify UFOs within the tokamak. The model attained an accuracy of 95.1%, a precision of 95.4% for UFO detection, a recall of 95.1%, and a Receiver Operating Characteristic curve area (AUC) of 0.99. These metrics underscore the model’s competence in reliably identifying UFOs with minimal false identifications. Consequently, this suggests the model’s suitability for integration into UFO detection systems in tokamak environments, potentially improving monitoring and efficiency, and providing an avenue for real-time UFO detection in future research.