Machine-Learning Optimization of Energy Efficiency in Microwave Applicators with Plasma

With a widening range of applications of microwave plasma in processing technologies, there is a demand of computer-aided design of efficient and controllable applicators containing microwave plasma. The use of modern computational tools here is limited due to a principal uncertainty in data on the plasma medium for electromagnetic simulations. In this paper, based on simplified characterization of microwave plasma, we propose to embed a model of an applicator with plasma in a machine-learning optimization procedure aiming to maximize energy efficiency of the system. To demonstrate functionality of the procedure featuring the dynamic training of the decomposed RBF network, we use a model of a conventional cylindrical applicator with a coaxial quartz vessel containing plasma. The model is built in the FDTD simulator QuickWave. For a neutral gas chosen, the procedure finds the geometry of the applicator and the plasma frequency satisfying the optimization goal.