arXiv subjects

Subrata Roy

Publications and source records attributed to Subrata Roy.

2 recordsLinked to original sources

arXiv2026

A Two-Dimensional Physics-Informed Neural Network Model of a Hall Effect Thruster Plasma: Self-Consistent Electrostatics and Physical-Consistency Diagnostics

We present a two-dimensional axisymmetric physics-informed neural network (PINN) of the discharge channel of the P5 Hall thruster at 300 V and 5.4 A, which solves eleven coupled plasma fields and assimilates 145 internal probe measurements per field reconstructed from the data of Haas and Gallimore. The electric potential is obtained from Poisson's equation in charge-separation form, n_e = n_i(1+eta), with eta bounded by construction, which keeps the stiff electrostatic constraint active and trainable. The reference solution agrees with the training profiles to relative L2 errors of 0.017, 0.073 and 0.205 for the potential, electron temperature and ion density, and held-out tests show the potential to be far more robust than the ion density to the withdrawal of measurement lines. We then test the solution against requirements the data fit does not reveal: global particle and current balances, the electron current implied by the momentum equation, the field-aligned electron force balance and the axial ion momentum budget. These expose a neutral inflow six times the measured flow, a bound in the potential representation, and a momentum-implied electron current, more than 99% field-aligned, exceeding the discharge current by orders of magnitude. Correcting these, with a divergence-free magnetic field fitted to measured data, uncertainty-weighted data, the electron pressure gradient and progressively imposed field-aligned and current constraints, yields a solution that conserves particles to within 3%, holds the current between 5.1 and 5.4 A and places 97% of the potential and 88% of the temperature measurements within their uncertainties. Within the tested single-velocity fluid formulation, however, the ion density, potential and ion velocity could not be matched together, and the field-aligned electron balance remains the main obstacle to electron self-consistency.

physics.plasm-ph↗

arXiv2022

Physics-Constrained Deep Learning of Incompressible Cavity Flows

High resolution simulations of incompressible flows have become routine across a range of engineering applications. Despite their routine use, due to the high dimensional parameter space present for most practical applications, a comprehensive exploration of the available parameter space is often impractical. In this work, we demonstrate the ability of physics-constrained deep learning methods to provide an efficient means of exploring high-dimensional parameter spaces with minimal amounts of data from high resolution computational fluid dynamic simulations. As a specific application, we choose the well established problem of a 2D lid driven cavity flow. While giving an extensive treatment of the classic case of a square cavity, we extend the analysis to treat an isosceles trapezoid. In so doing, the number of parameters determining the solution includes not just the Reynolds number, but also two additional parameters characterizing the geometry of the cavity. Thus, together with the $\left( x, y \right)$ variation of the flow and pressure in configuration space, the presence of these three parameters results in the solution varying in a 5D space. It is shown that in the absence of data, physics-constrained methods are able to provide an accurate description of the cavity flow in this 5D space up to intermediate values of the Reynolds number, but fails to train for sufficiently high Reynolds numbers. In contrast, using a small quantity of flow data, a single neural network is able to provide an accurate description for a broad range of Reynolds numbers and cavity geometries. Once trained, such a model provides a rapid surrogate that can be used to efficiently explore the 5D space. This 5D surrogate model is subsequently used to identify critical parameter values for the merger and splitting of vortices as the Reynolds number and cavity geometry are varied.

physics.flu-dyn↗