Accelerated Neural Network Solvers of Navier Stokes Equations for Turbulent Flows

Physics-based data driven Neural Network solver is developed to solve the inertial and buoyancy-generated turbulent flows. A configurable U-Net architecture has been trained to solve the multi-scale Elliptical Partial Differential Equations. Building on the underlying concept of V-cycle multigrid methods, a Neural network framework using U-Net architecture is optimized to solve the Poisson Equation and Helmholtz equations - the characteristic form of the discretized Navier-Stokes Equations. The U-Net based Elliptical solver is coupled to the Neural network framework to accelerate the computational time measured using FLOPS by two orders of magnitude. The method is extremely promising for very high Reynolds number turbulent flows.