LARGE EDDY SIMULATION OF FUEL-AIR-MIXING IN A DIRECT INJECTION SI ENGINE

This work extends the investigation by Goryntsev et al. (2007) of cycle-to-cycle variations of flow and mixing field using Large Eddy Simulation (LES) method to fuel spray injection driven flows using the CFD KIVA-3V code (Amsden et al., 1989) as extended to LES. Especially the effect of the cycle-to-cycle variations on the fuel-air-mixing close to the ignition point will be investigated. The configuration investigated represents the “BMBF” generic four-stroke direct fuel injection engine with variable charge motion system. Some experimental data for single phase flow and spray are available. They will be used for model validation. INTRODUCTION The call for environmentally compatible and economical vehicles, still satisfying demands for high performance, necessitates immense efforts to develop innovative engine concepts. However, such engines involve liquid fuel along with multiphase flow phenomena such as droplet evaporation and spray combustion. A good knowledge of the spray evolution properties, the heating and evaporation as well as the interaction with the gas-phase phenomena such as turbulence, mixing and chemical reactions is important for the design and flow control of such engineering devices. While numerous experimental and RANS-based numerical investigations concentrated on the way to gain insight into the behavior of the spray in IC-Engines, LES may help in delivering detailed unsteady information needed to better understand the strongly transient phenomena going on the combustion chamber. A recent review of LES in IC-Engines was provided by Celik et al. (2001) focused on single-phase flows while Sadiki et al. (2006) deal with turbulent two-phase flows. It turns out that modern internal combustion engine concepts like the Gasoline Direct Injection (GDI) offer a great chance to meet current and future emission standards. Especially air-guided direct injection systems used to instantiate stratified charge at part load allow for an optimized fuel consumption and a low level of emissions. During this crucial process, the engine is very sensitive to cycle-to-cycle variations of the flow and mixing field. Therefore, this work extends the investigation by Goryntsev et al. (2007) of cycle-to-cycle variations of flow and mixing field using Large Eddy Simulation (LES) method to fuel spray injection driven flows using the CFD KIVA-3V code (Amsden et al., 1989). Especially the effect of the cycle-to-cycle variations on the fuel-air-mixing close to the ignition point will be investigated. The paper proceeds as follows. The configuration, numerical method and models are briefly described in the next section. The presentation and discussion of the results for singleand two-phase flow are given further. The main findings are summarized in the final section of the paper. CONFIGURATION AND NUMERICAL MODELS The KIVA-3V code allows for the solution of the 3dimensional, unsteady, compressible equations of fluid motion. The conservation equations are discretised using the Finite Volume Method (FVM) on an arbitrary hexahedral mesh applying the Arbitrary Langrangian Eulerian (ALE) method. For details see (Amsden et al., 1989, Amsden, 1993, Amsden, 1997) and references therein. KIVA offers two different RANS models ( ε − k and RNG) to account for turbulence effects and is widely used for the simulation of ICE fluid dynamics, especially for in-cylinder flows. The current study is based on a LES approach using the classical Smagorinsky model (Smagorinsky, 1963), which has been implemented in the code (Amsden et al., 1989). In the present implementation, the model constant was taken to be 0.1, following typical literature values. A square duct configuration was used to validate the new KIVA-3VLES code and results were found to be in good agreement with available DNS data (Goryntsev at al., 2005). Simulations of spray were carried out using the standard Smagorinsky SGS model for the SGS stress tensor implemented in KIVA-3V. The so-called DDM (discrete droplet model of Dukowicz) (Amsden et al., 1989) with Lagrangian, computational particles that represent parcels of spray droplets with uniform properties was applied for the spray description. The spray and fluid interactions are