Real-time monitoring system for detection and characterization of thermal events on WEST Tokamak: Implementation and first results

Abstract

Analyzing thermal events in fusion devices is essential for both scientific understanding and machine protection. The infrared diagnostics of long-pulse machines, such as W Environment Steady-state Tokamak (WEST), generate a large amount of video data, presenting a challenge for human operators to conduct real-time thermal event analysis. Therefore, there is a need for a process able to detect and analyze thermal events automatically and in real-time, for feedback control and investment protection. Up to now, multiple types of intelligent methods have been applied between experiments during previous WEST campaigns (detection of thermal events, characterization of strike lines on the tungsten divertor, and detection of electric arcs). This paper introduces a modular framework enabling their use on infrared data in real-time. With a NVIDIA A40 Graphical Processing Unit (GPU) added to the Wall Monitoring System (WMS) of WEST, the detection of thermal events and characterization of strike lines, which rely on Artificial Intelligence (AI) approaches, run at 30 fps. The electric arc detection system, based on a compact Convolutional Neural Network (CNN), operates at over 100 frames per second. For the first time in the fusion domain, an AI-based image analysis method is being utilized to send instructions to the heating antennae through integration with the Plasma Control System (PCS). Therefore, this framework unlocks the possibility of using the produced results for real-time heating feedback control. Furthermore, these results are stored in a dedicated local database for later studies. It was first tested during the C9 campaign of WEST and performed well on the large quantity of infrared videos captured, paving the way toward intelligent real-time control of large fusion machines such as ITER.