Development of an n = 1 optical mode tracking feedback control system on HBT-EP using a deep learning neural network

Active feedback control is critical to tokamak operation for mitigating plasma instabilities. Feedback latency is a key challenge for effective control systems and must be significantly lower than the time scale of instability growth. In this work we describe the development of an optical based mode control system on HBT-EP as well as the development and training of the mode tracking machine learning algorithm. Using a fast camera diagnostic and convolutional neural network (CNN) deployed on a field programmable gate array (FPGA), we predict sine and cosine components of n=1 modes [1]. We analyze the feedback outputs and predictions from the CNN compared to GPU driven magnetic control signals [2]. The fast camera feedback system achieves a trigger to output latency of 17.6us making it sufficient for MHD mode control and competitive to the GPU system. We also discuss the implementation of the feedback system on HBT-EP.

[1] Y. Wei et al 2023 Plasma Phys. Control. Fusion 65 074002

[2] Q Peng et al 2016 Plasma Phys. Control. Fusion 58 045001