An open source fusion machine agnostic standard for the exchange and processing of infrared videos and video annotations

Abstract

The exploitation of the infrared diagnostics of long pulse fusion machines requires sophisticated and automated intelligent models for effective machine protection. To achieve this, it is imperative to train these models on high-quality and diverse datasets derived from different fusion machines, such as WEST and W7-X, as well as from simulations for future projects such as ITER. To address the need for a standardized approach to exchange infrared videos of the inside of vessels and video annotations within these scenes, we propose two open-source standards accessible on GitHub. These standards aim to facilitate the sharing and use of data for the training of machine protection models across various fusion machines. The proposed standards address the challenge of creating a common platform for sharing infrared video data as well as video annotations, ensuring compatibility among different machines. This initiative serves as a pivotal step towards establishing a unified framework for collaborative research in fusion research and development. The ThermaVIP software platform uses these standards to read infrared movies in the standardized format, read and write annotations to the proposed JSON standard, and read from and write to a MYSQL database. As a demonstration, a cross-machine database was deployed on the EUROfusion gateway. This database serves as an initial repository hosting infrared videos and annotations from both WEST and W7-X, laying the foundation for a comprehensive and diverse dataset essential for training intelligent models. This paper presents the proposed open source standards and the first deployment of a cross-machine database. This collaborative and standardized approach is vital to ensure the robustness and generalization of future protection models.