(1-14) CFD and Combustion in Engines with an Unstructured Parallel Solver Based on KIVA((DE-4)Diesel Engine Combustion 4-Modeling)

Abstract

KIFP, an hexahedral unstructured version of KMB (our present CFD code for engines), has been developed. Based on KIVA's algorithms (finite volume on staggered grids, time-splitting, SIMPLE loop, sub-cycled advection...), the new solver has been built step by step with a strong control on the numerical results. This paper shows the different phases of this work. The numerical approaches and developments are discussed. Some academic examples are shown and compared with KMB or analytical results, like scalar advection or multi-species diffusion. Better precision and convergence in the physical fields are observed. Iterative loops and advective sub-cycles are also reduced thanks to the unstructured formalism. Super-scalar machines being widely used and developed, KIFP is targeted to them. OPEN-MP paradigm is used to parallelize the code, and the paper gives results on performance and speed-up (more than 3 for 4 processors in the best case). Results on a compression of a swirl motion in Direct Injection Diesel engine are given. For a given mesh density, physical results and numerical quality (driving CPU time) seems better for KIFP than for KMB. This is due to a better mesh configuration in the KIFP case (no corner cells). At last, a pancake spark ignited engine is used to compute homogeneous combustion with the extended coherent flame model (ECFM) implemented in the new solver. We find a good agreement between the results of the two codes and experiment.