Optical and machine learning based low latency plasma feedback control of n=1 resistive wall modes on HBT-EP

We report on the progress and performance of the machine learning based optical feedback control system on HBT-EP which is an application of an ultra-low latency optical based mode tracking algorithm [1]. We discuss performance differences between the fast camera and machine learning control system on HBT-EP with the magnetic feedback control system [2]. Using a convolutional neural network (CNN) the feedback system achieves an actuation latency of 18.2us when deployed on a field programmable gate array (FPGA). CNN hyper parameter and architecture optimization was done using a genetic algorithm with an expanded training data set compared to prior algorithms [1]. The control algorithm orchestrates 40 actuating coils through a preset mode structure based on 5 unique coil requests attempting to reduce targeted MHD mode amplitudes. Machine learning based control algorithms can fill a needed role for coupling novel diagnostics to real time plasma control systems.

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

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