Discharge Identification Method Based on Optical Emission Spectroscopy

The emission spectrum of plasma discharges encapsulates abundant physical information, reflecting critical plasma parameters such as reactive groups, gas temperature, and electron density. It also serves as a foundational element for identifying discharge types. This study designs two discharge platforms for executing microwave jet discharges and pulsed spark discharges. Through emission spectrum analysis, it elucidates the variations in plasma gas temperature and electron density within air, argon, and helium, correlating these changes with power supply output and gas flow rate. Additionally, to assess the efficacy of machine learning in datasets of limited scale, the research establishes a discharge identification model utilizing Deep Neural Networks (DNN) and another model based on the Kmeans clustering algorithm for recognizing four types of discharges. The findings demonstrate a direct relationship between discharge power and gas temperature across all discharge types, inversely correlated with gas flow rate, albeit with minimal impact on pulsed helium discharges. Except for an increase in electron density with power supply output in pulsed argon discharges, the other discharge types showed a decrease in electron density as power output increased. The influence of gas flow rate on electron density was marginal across all discharge types. The Deep Neural Network model achieved a 100% accuracy rate in identifying microwave and pulsed discharges, outperforming the K-means clustering algorithm-based model, which attained a 71.25% recognition rate, indicating a superior performance on small-scale datasets. This method, which necessitates only the extraction of light intensity data from the emission spectrum for feature derivation, is applicable across various discharge types, presenting significant practical relevance.