A turbulent eddy-viscosity surrogate modeling framework for Reynolds-averaged Navier-Stokes simulations

The Reynolds-averaged Navier-Stokes (RANS) equations for steady-state assessment of incompressible turbulent flows remain the workhorse practical computational fluid dynamics (CFD) applications. Consequently, improvements in speed or accuracy have potential to affect a diverse range We introduce machine learning framework surrogate modeling eddy viscosities RANS simulations, given initial conditions. This strategy is assessed parametric interpolation, while numerically solving pressure and velocity steady state, thus representing that hybridized with learning. achieve competitive results significant reduction solution time when compared those obtained by Spalart–Allmaras one-equation model. because proposed methodology allows considerably larger relaxation factors solvers. Our assessments are made backward-facing step considerable mesh anisotropy separation represent CFD application. For test experiments either varying inlet conditions heights we see time-to-solution reductions around factor 5. an opportunity rapid exploration parameter spaces prove prohibitive utilizing turbulence closure models multiple coupled partial differential equations.