Analysis of OH Emission Spectra Using Deep Learning

Plasmas in liquids have attractive features both of plasma and liquid, such as high reactivity and solubility. However, such reaction fields including phase transitions are more complicated than conventional gaseous plasma fields. Thus we need to improve diagnostics and analysis methods. Deep learning is a well-known method for analyzing and generalizing some complicated data and extensively studied in various areas. As for the plasma diagnostics, several studies have been reported, including analysis to extract internal parameters from optical emission spectra. In this study, we also apply deep learning to OH optical emission spectra with the purpose to analyze the plasma in liquid reaction fields more effectively. Using more than million theoretical spectra, we make a deep learning model to train the relationship between emission spectra and various parameters not only OH rotational temperatures, but also apparatus functions. The trained model were tested for estimating each parameter from the experimental spectra. The interesting point with this model is that we do not need to know the apparatus function for finding rotational energy distribution, which we assume here in bi-Maxwellian distribution. Further details and discussion about various errors will be presented.