Learning control coil currents from heat-flux images using convolutional neural networks at Wendelstein 7-X

An important goal of Wendelstein 7-X, the most advanced operating fusion experiment of the stellarator line, is to demonstrate the ability of stellarators to perform steady-state discharges. In this respect, the monitoring and control of the heat loads on the plasma facing components, especially of the strike-lines in the ten island divertors, will be critical during next operation phase OP2. In this paper, it is shown that deep convolutional neural networks are able to learn the relationship between the heat-flux images, obtained by the analysis of thermographic data, and the applied control coil currents in standard magnetic configuration experiments. This study is carried out in view of understanding and modeling the relationship between the heat-flux distribution in the divertor strike-lines and the actuators influencing them.