Based On Parallel Computing Three-dimensional Compressible Mixing Layer Large-eddy Simulation

The present thesis is supported by the director foundation of The National Key Laboratory of Transient Physics (NKLTP), and its research contents can be classified to the interdisciplinary field of computational fluid dynamics(CFD) and computer science and its application.An important physical phenomenon concerning the mixing of two compressible fluid streams moving with different velocities in the same direction had been investigated numerically by means of LES. The evolution of a large vortex structure during mixing process had been discussed under the conditions of different density and velocity ratio of two mixing layers, the influences on the mixing for different fluctuating initial velocity hadalso been studied based on LES and parallel computing.The mathematical model was developed following the practical three-dimensionalcompressible mixing layer flow, after the Favre filtering, the sub-grid-scale stress (SGS) model dependent on the eddy-viscosity hypothesis was used close the governing equations. The two-order Runge-Kutta scheme is adapted to handle the computations in time, and the inviscid part of fluxes in governing equations was evaluated using monotone piecewise-parabolic approximation, and the viscous fluxes were computed from the second order in space central differences. One computational example was calculated with the Mach number 0.65, the linear growth of momentum thickness indicating the pairing and amalgamating large vortex structures was obtained. The self-similar property for mixing layer has also been exposed. Moreover, the effects of the initial fluctuating velocity, density and velocity ratio of mixing layers on the mixing layer had also been calculated numerically. Our results showed that the larger absolute values of fluctuating velocity, density and velocity differences, the earlier for the occurrence of large vortex structures.Parallel computing was also one of our research contents. A PC clusters consists of six personal computers and administrated by the Linux Enterprise 4.0 has been developed, MPICH1.2.7 and MPICH 2 had been applied for the parallel computing respectively. The original code had been rewritten to run in a MPI parallel environment with the use of domain spliting method. Our test results showed that the computing performance is significantly improved, the parallel speed up ratio was obtained at its highest value of about 4.0 with use of 5 PCs as the parallel computing processors. Therefore, the use of simple and cheap PC cluster for substituting the complicated and expensive super computer in CFD field is achieved.