LES of Flow Processes in an SI Engine Using Two Approaches: OpenFoam and PsiPhi

In this study two different simulation approaches to large eddy simulation of spark-ignition engines are compared. Additionally, some of the simulation results are compared to experimentally obtained in-cylinder velocity measurements. The first approach applies unstructured grids with an automated meshing procedure, using OpenFoam and Lib-ICE with a mapping approach. The second approach applies the efficient in-house code PsiPhi on equidistant, Cartesian grids, representing walls by immersed boundaries, where the moving piston and valves are described as topologically connected groups of Lagrangian particles. In the experiments, two-dimensional two-component particle image velocimetry is applied in the central tumble plane of the cylinder of an optically accessible engine. Good agreement between numerical results and experiment are obtained by both approaches. Introduction Direct injection, downsizing and advanced combustion modes are key fuel-saving technologies in gasoline engines. To further decrease the fuel consumption and pollutant emissions and to increase the power output, a better understanding of the in-cylinder processes is crucial. Currently, advanced combustion modes cannot be used over the full operating range, often due to turbulence-induced flame quenching or as a result of poor fuel-air mixing near the spark. In-cylinder phenomena are commonly studied in single-cylinder research engines with optical access for laser diagnostics. On the other hand, engines are investigated by numerical techniques like CFD, which is often less expensive and more flexible than an experiment. As the state of the art, U-RANS simulations are successfully applied by industry to gain an understanding of the engine, but U-RANS will normally fail to predict cyclic variations. A promising alternative are large eddy simulations (LES) that resolve smaller flow structures, enabling them to capture cyclic variations. However, LES is computationally more expensive and requires high-quality meshes on which high-order numerical schemes must be applied. In the context of LES, several CFD codes like AVBP [5,6], KIVA [7], FLUENT [8], or Star-CD [9] have shown at least partial ability to predict some relevant phenomena in internal combustion engines. A critical problem with the application of LES is that any discretization of less than second order accuracy and CFL numbers greater than one lead to artificial dissipation – causing slow mixing, insufficient flame wrinkling, and hence slow flame propagation. Unfortunately, it is very hard to satisfy these accuracy requirements with CFD codes that have been optimized for RANS on unstructured grids. In this study, two different approaches are compared, which satisfy the stringent requirements for mesh quality and numerical accuracy for the LES of internal combustion engines. Both methods require very limited effort for the grid generation (less than one personnel hour for meshing). The first approach (OpenFOAM) [1] uses unstructured grids with deformable meshes. The second approach (PsiPhi) [2] is based on a structured grid with a combination of Lagrangian particles [3] and Page 2 of 13 immersed boundaries [4] to represent the moving parts of the engine. Both codes, OpenFOAM and PsiPhi (in-house, developed at the chair of Fluid Dynamics, University DuisburgEssen) were available without excessive cost for licensing and have demonstrated good parallel scaling, in the case of PsiPhi beyond 4000 cores. So far, the simulations have concentrated on a motored case, for which the velocity fields obtained in both approaches will be compared to each other and to measurements in an optically accessible engine. At the end of the paper, preliminary results are presented for a fired case.