M.Sc Thesis | |

M.Sc Student | Kidron Ya'eer |
---|---|

Subject | Parallel lmplicit Algorithms for Cartesian Grids |

Department | Department of Aerospace Engineering |

Supervisors | PROFESSOR EMERITUS David Durban |

DR. Yuval Levy |

Computational Fluid Dynamics (CFD) Simulations are composed of two main stages which may be coupled: discretization of the flow control volume using a grid generator and simulating the flow field using a flow solver. In the current work, the entire process of CFD simulation is covered. Starting from the initial definition of the geometry, developing a Cartesian grid generator, and an accompanying flow solver, continuing with parallelizing the flow solver to enable large scale simulations, and finalizing with dynamic load balancing to achieve maximum efficiency. Algorithms for Cartesian grid generation were developed. Among the algorithms, a new inside-out approach where the grid is generated using the local convex hull (in a grid cell that is cut by the geometry), followed by the projection of the hull onto the cell boundaries. A complementing algorithm for local refinement is also implemented. In addition, an algorithm for immersing a fine grid in a coarse grid, in a fashion that keeps the Cartesian nature of both grids, is also developed. The Cartesian grid generator is accompanied by a finite volume Navier-Stokes flow solver. The flow solver uses the HLLC scheme, an advanced upwind scheme for the evaluation of the convective fluxes. The flow solver is validated using several test cases for inviscid and viscous flows, and is compared to experimental data. The versatility of the whole process is demonstrated by simulating the flows about a multi-element airfoil and a staggered bi-plane configuration. The HLLC scheme is inherently imbalanced in the sense that the computational load is