Research Projects

Low-temperature ExB plasma (LTP) based devices have a wide range of applications such as negative ion sources for nuclear fusion, Hall thrusters, magnetrons for plasma processing, ECR sources in semiconductor industry and linear machines for basic research. Most of the fundamental principles involved in such LTPs are well established, except anomalous transport and instabilities. Current research is primarily focused on development/ optimization of plasma devices for specific technological applications either through experimental investigations or computational modeling using PIC-MCC (Particle-In-Cell-Monte-Carlo-Collisions) simulations which provides spatio-temporal evolution of charged-particle velocity distribution functions under effect of self-consistent electromagnetic fields and collisions. LTPs are primarily characterized by measuring plasma density, temperature and potential. However, a detailed investigation of Energy Distribution Functions (EDFs) helps in accurately interpreting the complex physics. Electron-EDF controls the plasma properties and rates of electron impact reactions that generate reactive species, whereas ion-EDF is associated with surface effects. Accurate determination of EDFs, which can be non-Maxwellian in nature, is important because it controls rate coefficients (excitation, ionization etc.) and plasma transport. In experiments, probe measurements and optical diagnostics are used for plasma density and temperature determination, but EDF measurement is challenging due to presence of magnetic field, instabilities, probe-induced perturbations and difficulties in obtaining spatially resolved data. However, in a PIC simulation, plasma can be accurately characterized using the macroscopic quantities (such as density, temperature, potential) as well as the phase-space quantities which facilitate study of the EDF. In this project, firstly, we aim to perform a comprehensive investigation of spatio-temporal evolution of EEDF/ IEDFs in ExB plasmas using our in-house parallel 2D-3V PIC-MCC code. Subsequently, we propose to build Machine Learning (ML) based approaches which can overcome some of the major limitations of the rule based classical approaches for plasma characterization. Training data for ML will be generated using multiple PIC simulations under different experimental conditions. Feedback from experimentalists will be employed to prepare a diverse synthetic training dataset for performance evaluation of ML models. The final goal is to obtain a mapping function between macroscopic quantities (easily obtained in experiments) and EDFs. We also propose to perform an uncertainty quantification of predicted solutions (EDFs) obtained from ML to establish the acceptability of this proof of concept study. The project outcome will aid in providing feedback on how experimental diagnostics and data collection can be performed for generating training data for future deployment of ML based EDF diagnostics in real systems.