Accelerated real-time plasma diagnostics: Integrating argon collisional-radiative model with machine learning methods

The present work employs two advanced machine learning (ML) techniques: the Random Forest (RF) model and Deep Neural Network (DNN) for the non-invasive spectroscopic diagnostic of a non-thermal atmospheric pressure Argon plasma jet. By integrating the ML techniques with OES and the collisional radiative (CR) model, realtime prediction of electron temperature (Te) was achieved. Both ML models were meticulously optimized by tuning the hyperparameters, employing a random search strategy. An extensive data set was used to train and test both ML models. The DNN showed an impressive R-square value of 0.9964, while the RF model achieved an R-square of 0.9869. These high accuracy levels in predicting Te, underscore the effectiveness and precision of the combined ML approach. This innovative integration of the RF and DNN models paves the way for an alternate approach to enhance the speed and accuracy of plasma parameter estimation using traditional spectroscopic plasma diagnostic approaches.