Machine learning-based simple and fast approach for the real-time characterization of laser-induced plasma

Abstract

We report a simple approach to estimate the fast and real-time pulse-to-pulse prediction of plasma parameters, specifically electron temperature and electron number density, using an Artificial Neural network (ANN) in combination with Laser-induced breakdown spectroscopy (LIBS). In a variety of spectroscopic applications, it is essential to have real-time observation of plasma parameters. However, direct measurement of these parameters is challenging and requires complex and time-consuming calculations. Artificial Neural Network (ANN) can be used to model the relation between spectral features from recorded LIBS emission spectra and plasma parameters. The ANN is trained on a suitable preprocessed spectroscopic dataset with corresponding plasma parameters to predict electron temperature and electron number density. The accuracy of Artificial Neural Network (ANN) in predicting the plasma parameters is evaluated, and results are validated with existing conventional methods of calculating plasma parameters, namely the Boltzmann Plot Method for plasma temperature and the Stark Broadening Method for electron number density. The present results show that ANN is an effective method in accurately predicting the plasma parameters directly from the spectral features. The ability to fine-tune plasma in real time enhances control and accuracy in Pulsed Laser Deposition (PLD) process and other plasma coating techniques.