Integration of parallel computation and dynamic mesh refinement for transient spray simulation

Abstract

This study developed parallel computing schemes to enhance the computational efficiency of engine spray simulations when adaptive mesh refinement was used. Spray simulations have been shown to be grid dependent and thus fine mesh is often used to improve solution accuracy. In this study, dynamic mesh refinement adaptive to the spray region was developed and parallelized. The change of element numbers in dynamic mesh refinement posed difficulties in developing parallel computing schemes for the purpose of achieving good load balance and low communication overhead. In this work, the cost of communication re-initialization was minimized by performing only necessary data gathering, scattering, mapping and property update that involve local mesh refinement at each adaptation. The implementation was first validated by comparing the predicted spray penetrations and structures under different conditions. Results using local mesh refinement are the same as those using the uniformly fine mesh. This approach was also shown to effectively remove the artifacts related to the algorithms of spray modeling. Furthermore, the parallel implementation was validated by comparing the liquid penetrations using both the fine mesh and the coarse mesh with local refinement using different numbers of processors. The present computational schemes were then used to simulate transient sprays in three different engine geometries including a cylindrical chamber, a 2-valve engine, and a direct-injection gasoline engine. Various spray conditions and grid resolutions were tested in the three geometries to assess the computational efficiency. Reasonable speed-ups were obtained for the test cases with both single-jet and six-jet injections. The six-jet injection has mixed influences on the parallel performance in different geometries compared to the single-jet injection. The reason can be related to the domain partitioning along with local mesh refinement in different geometries. It was also found that the parallel performance for the 2-valve engine geometry suffered from the extensive snapping activities required during valve motion as compared to the other two cases without valve motion.