Some Issues On Dynamic Unstructured Grid Generation For Parallel CFD Simulations

Moving boundary problems have commonly occurred in many important aerospace industry applications, such as store separation, stage separation, cluster munition dispersion and wing flutter etc. The accurate and efficient simulation of these problems is one of the most challenging issues of computational fluid dynamic (CFD). In particular, in order to achieve a higher level of fidelity, CFD simulations nowadays involve more and more complicated geometric and physical models. As a result, the magnitudes of simulations grow up rapidly. To overcome the performance bottleneck of large-scale simulations, parallelization has become very common in the routine CFD simulations. Nevertheless, most of existing parallel CFD systems adopt a quasi-parallel workflow, i.e., the simulation procedure is executed in parallel, while the run of the meshing procedure remains sequential. Apparently, when the simulated problem involves tens of millions, or even hundreds of millions freedoms, and the CFD solver could run in parallel on thousands or more computer cores efficiently, the sequential grid generation procedure and the data transfer procedure between the sequential mesher and the parallel solver could easily dominate the entire simulation cycle in terms of computing time.This study is mainly concerned with the simulations for typical moving boundary problems that is indispensable for flight vehicle design, such as the simulations of multi-body separation problems. Different with those quasi-parallel workflows, the simulation system developed in this study follows a completely parallel workflow. In other words, both the grid generation part and the solution part of this system are executed in parallel. To this end, our research covers different issues occurring in the development of such a system, including numerical approaches of CFD, automatic grid generation approaches and the setup of a parallel CFD system. Basically, this research work and its contributions can be grouped into the following three parts.Firstly, based on the governing equations descripted by the Arbitrary Lagrange-Euler (ALE) method, a finite volume method is proposed to simulate flows using unstructured moving grids. A CFD system is thus developed by combining the CFD solver with the modules treating unstructured grid deformation, unstructured grid remeshing and rigid body motion (governed by six degrees-of-freedom (6DOF) equations), respectively. Benchmark simulations and simulations for real engineering problems have been conducted to verify the efficiency and effectiveness of the developed CFD system.Secondly, a new local reconnection technique named shell transformation is proposed for unstructured grids. This technique could remarkably improve the reliability of boundary recovery and thus the reliability of Delaunay mesh generation (which requires a boundary recovery procedure to ensure the boundary integrity of the resulting grid), in particular when the input surface grid contains a large number of stretched elements. Besides, this new local reconnection technique could be used to improve grid quality further. Consequently, supported by this new technique, a high-quality unstructured grid generator has been developed and successfully applied in the initial grid generation and local remeshing procedures of the developed CFD system.Thirdly, to enable the simulations of large-scale unsteady problems, a domain decomposition based approach is presented for the parallelization of the CFD solver, and the implementation is accomplished by using the Message Passing Interface (MPI) standard. Meanwhile, a parallel preprocessing technique for the CFD solver is investigated to bridge the parallel grid generator and the parallel solver. In this fashion, a complete parallel numerical simulation system is set up to compute the initial solution of the unsteady flow. And a novel parallel local remeshing technique is suggested using the domain decomposition approach. Finally, a complete parallel workflow is built up for unsteady flows with moving boundaries.