Machine learning-based prediction of process conditions in atmospheric-pressure microwave plasma reactor from plasma images

Atmospheric-pressure plasma is thermal and has been used to decompose gases with high global warming potential and strong chemical bonds based on its high temperature and abundance of radicals. These process gases were in typical diluted with inert molecular nitrogen prior to the decomposition process, and oxygen-containing species such as molecular oxygen and water were added to prevent the decomposed species from recombination into the original gases. The concentration and flowrate of the process gas vary in time and it is necessary to monitor these process conditions to operate the system optimally. In this study, we proposed a method that predicts the process conditions of process gas concentration and flowrate from a plasma plume image using a convolutional neural network. The type of the plasma used is the microwave plasma and methane and molecular oxygen were selected as the process gas and reaction agent, respectively. Both the concentration of the process gas and the total flowrate were predictable to within ±2% of their full operating ranges with 95% accuracy (i.e., ±100 ppm and ±2 slpm, respectively).