A massively parallel solution of the three-dimensional Navier-Stokes equations on unstructured adaptive grids

This thesis describes a three-dimensional Navier-Stokes code developed for the solution of the external flow over complex geometries on massively parallel computers. The motivation for the present work comes from the introduction of the massively parallel computers into the market.;The code uses a finite volume approach and an unstructured discretization of the computational domain. The integral form of the governing equations is used with the computational cell replacing the control volume. The flow variables are stored in the cell centers and the fluxes through the cell faces are calculated based on a piecewise constant distribution of the flow variables inside the cell. The flow variable jump across the each face is taken as an initial condition for the Riemann problem and Roe's approximate Riemann solver is used to upwind the fluxes.;A four-stage Runge-Kutta scheme is used to integrate the equations to a convergence. Local time stepping is used for steady-state cases to enhance the convergence. Time accurate integration is optional, for time dependent cases. The discretized equations are very flexible in terms of the cell geometry. The code can be run on hexagonal, tetrahedra or prismatic grids.;The computational grid can be adapted to the evolving solution. The grid adaptation uses a face-based momentum difference, weighted by the local cell size as an error indicator.;Turbulent flows are simulated using the Baldwin-Barth model. The model was implemented in a massively parallel fashion by constructing a connectivity array between the solid surface faces and the interior cells. The results for the turbulent cases are in a good agreement with theory, as well as experimental data.