Development of a three dimensional, unstructured, GPUaccelerated, parallel computational fluid dynamics solverfor compressible reacting flows

In this thesis a three dimensional, compressible solver is developed. It solves three dimensional compressible chemically reacting flows with multiple species. The solver can operate on parallel CPU's with the help of GPU hardware. It is a pressure based methodology that can operate on unstructured co-located grids. It solves transient Euler and Navier-Stokes Equations. Solver uses Finite Volume Method and it uses a first order upwind discretization for convective terms. With the help of iterative sparse matrix solvers supported by AMG smoother, it can solve problems implicitly. Arrhenius type chemical reaction rates are used with multiple species. Solver uses ideal gas equation as equation of state. Variables like specific heat is a function of temperature in the solver. With the third party matrix solver ViennaCL this solver can perform parallel computations on GPU hardware. In an effort to validate the solver, laminar flow over flat plate problem, subsonic, transonic and supersonic flow over circular arc bump problem, Sod's shock tube problem and Flame D test case from Sandia laboratories are studied. Computing performance of the solver is also studied. In the GPU performance study, it is concluded that the solver can solve same matrix three and half times faster in similarly priced GPU than CPU, including data transfer times between main memory and GPU memory.