Leveraging Physics-informed Machine Learning to Develop a Rapid Surrogate Model for Plasma Sheaths

When plasma makes contact with a material, a plasma sheath forms where strong electric potentials moderate particle and heat fluxes from the plasma into the wall. The present study utilizes a physics-informed neural network (PINN) to evaluate a hierarchy of models of plasma sheaths. Unlike traditional deep learning methods, PINNs use the governing PDEs to constrain the predictions of a neural network, and thus do not require any experimental or simulation data to train. As a first application, we utilize a PINN to identify the parametric solution to a fluid model of the plasma sheath. While the offline training time of the PINN is far longer than a traditional solver, once trained, the PINN is able to predict the sheath profiles across a broad range of parameter regimes in a matter of milliseconds, thus yielding an effective surrogate of the plasma sheath. Ongoing work is focused on extending this sheath model to incorporate a fully kinetic ion distribution, using a recently developed Vlasov-Fokker-Planck PINN [1]. This kinetic extension of the sheath model will enable the rapid prediction of the ion distribution function at the plasma-material interface, and thus offer an efficient, but also high physics fidelity description of the plasma sheath.

[1] C.J. McDevitt and X.-Z. Tang, PoP 31, 062701 (2024)