FR1.5 – Novel software for plasma-surface interactions, multi-phase plasmas, and biomatter.
Goal: Integrate artificial neural networks and machine learning algorithms into plasma simulations.
Background: The effectiveness and applications of artificial intelligence and machine learning in the physical sciences remains poorly understood. Physics Informed Neural Networks (PINNs) can solve high-dimensional partial differential equations (PDEs) efficiently but their full potential for fundamental science is not yet elucidated. The deep training abilities of PINNs could help identify appropriate solvers for hybrid kinetic-fluid simulations of classical and quantum transport processes. Inverse problems insoluble by existing methods are ideally suited to PINNs methods.
Challenge: To use artificial neural networks and machine learning algorithms for plasma simulations. Background: The effectiveness and range of applications of artificial intelligence and machine learning in the physical sciences remains poorly understood. Physics Informed Neural Networks (PINNs) can solve high-dimensional partial differential equations (PDEs) efficiently but their full potential for fundamental science is not yet elucidated. The deep training abilities of PINNs could help identify appropriate solvers for hybrid kinetic-fluid simulations of classical and quantum transport processes. Inverse problems insoluble by existing methods are ideally suited to PINNs methods.
Proposed Research: We explore the potential of artificial neural networks for solving plasma kinetic equations, developing smart software using multi-scale kinetic-fluid models, and solving inverse problems for plasma technologies. Using NVIDIA Modulus, we develop PINN-based tools for direct and inverse plasma problems.
Impacts: PINNs will accelerate solutions of the kinetic equations and facilitate the development of kinetic fluid models compared to traditional methods, utilizing smart software, and enable solving inverse problems that are currently insoluble.