ENGINEERING AND CFD

During the last decades, emissions regulations for internal combustion (IC) engines have become very strict. To reduce emissions, the current generation of IC engines is based on a combination of direct injection and downsizing. Both technologies require a significant charge motion, especially a reproducible tumble during the intake and compression stroke. The large-scale tumble acts like a storage system for the kinetic energy, introduced into the combustion chamber during the intake stroke. Ideally, this large-scale kinetic energy is released repeatable shortly before ignition due to the tumble breakdown. The resulting small-scale turbulent fluctuations accelerate the flame front propagation and therefore the combustion process.

The intake jet is the main contributor to the large-scale tumble motion and includes several interacting phenomena (e.g. flow separation, vortex shedding, wall reattachment), which have to be captured correctly by computational fluid dynamics (CFD). The intake jet is addressed by numerical and experimental studies, normally based on a steady-state flow configuration called “flow bench”. The complexity of such simulations was demonstrated in previous studies, which showed discrepancies between PIV (particle image velocimetry) and CFD results concerning jet orientation and penetration depth.

This numerical project focuses on the grid in the vicinity of the intake valve and its effect on the simulation results. Three different grids (M1, M2, M3) are analyzed using a state-of-the-art large eddy simulation (LES) turbulence model. The simulations are compared to highly-resolved 2D-2C PIV measurements, taken at the University of Duisburg-Essen. The overall goal is to develop a methodology for a quantitative comparison of different results in terms of the intake jet as well as the identification of crucial mesh regions. All investigations are performed with ANSYS CFX Release 16.0. The finite volume and node centered code solves the compressible transport equations for mass and momentum. For an increased accuracy, all simulations are performed by a second order scheme in space and time. Scaling a spatial resolution of 0.125 mm close to the valve gap is realized. For the finest grid (M3), there is a total number of 150 million grid points. The time step width is set to 1.5µs, leading to a CFL < 1 for the entire domain. Based on M3, 3000 CPUs per Job are required, summing up to 2 million CPUh for this case.

Figure 1 illustrates the velocity magnitude and the vorticity magnitude for two different numerical meshes (about 6 million grid points in M1). Both M1 and M3 exhibit a similar flow field topology as well as similar values for the velocity magnitude. A significant larger amount of resolved turbulent structures (illustrated by vorticity) can be identified for M3. This has a large impact on the jet-orientation, which is captured by a two-dimensional jet-centerline at mid-valve plane presented in Figure 2. M3 predicts the jet orientation and curvature much more accurate compared to the PIV data than grid M1. That demonstrates the sensitivity of the jet-orientation concerning the grid resolution.

Scientific Contact:
Prof. Dr.-Ing. Christian Hasse
Numerical Thermo-Fluid Dynamic
University of Freiberg
Fuchsmühlenweg 9, D-09599 Freiberg (Germany)
e-mail: Christian.Hasse [at] iec.tu-freiberg.de