TP 5 – Flame-propagation in inhomogeneous mixtures resulting from direct injection – large-eddy-simulation, model validation and analysis

The project aims to analyze cycle-to-cycle variations in internal combustion engines with inhomogeneous fuel-air mixtures. This requires simulating, varying parameters, and evaluating fired cycles in internal combustion engines with direct fuel injection and partially premixed (stratified) combustion. These simulations will be based on highly resolved large-eddy simulations (LES), which do not only resolve fluctuations of the bulk flow, but also the relevant turbulent eddies, so that their influence on the essentially stochastic process can be considered. The simulations are run over several consecutive cycles, in both motored and fired operation, relying on state-of-the-art combustion models, spray models, and massively parallel numerical techniques. The simulations reproduce the experiments from other sub-projects, they will be cross validated against them, and they help interpreting the measurements. For this purpose, experiment and simulation will vary injection, valve timing, and the location and time of ignition to determine the effect of mixture inhomogeneity on the stability and efficiency of the cycles. Innovative methods will be developed to trace phenomena and effects backwards in time and to find their causes. In the LES, this is enabled by Lagrangian particle tracking and the deterministic repeatability of simulation runs. Finally, causal connections will be described in a simplified form and combined with the findings of the other projects into new, global insights. Eventually, this sub-project will improve our understanding of cyclic variations, contribute to the modelling of such variations, and help in deriving measures to mitigate such variations.

The engine simulation with LES is based on previous works by Nguyen, using our in-house code PsiPhi. The goal of this development is the transfer of the high computational efficiency, accuracy, parallelization and robustness of the PsiPhi code to motor simulations. For this purpose, work from Wysocki on moving immersed boundaries was transferred from Nguyen to the motor and supplemented. The code for compressible flows was also extended, including suitable boundary conditions. With this approach it was possible to perform engine simulations computationally efficient, massively parallel (on more than 8000 cores) and high resolution (resolution 0.2mm in the entire combustion chamber). Thus, results were obtained which corresponded very well with the measurements in both towed and fired operation. Of particular relevance is the fact that the amount of work required to generate the computational grids using Immersed Boundary methods was negligible. It was also shown that a high resolution is required in the entire combustion chamber and not only in the valve gap in order to enable the maintenance of realistic turbulent kinetic energy levels and to be able to depict realistic flame propagation. Nguyen was also able to show that numerical methods, which are often used with commercial or freely available codes for engine simulation, can lead to enormously high numerical dissipation. Thus, unrealistically low turbulent viscosities are calculated for such LES, so that the quality of the LES is massively overestimated and the turbulent flame speed and engine efficiency are greatly underestimated.

The PsiPhi code describes moving geometries by means of immersed boundaries, i.e. ultimately by activating and deactivating computational cells, forcing an adhesive condition on the moving surface. All moving objects (pistons, valves) are represented by a point cloud, whose movement is specified. This approach avoids the time-consuming generation of different calculation grids and the (numerically dissipative) interpolation between these grids. Simple cubic-cartesian gratings are used for the computational grids, on which central fourth-order methods for the convection of the momentum, TVD methods for the convection of scalars, and an explicit Runge-Kutta method for time integration are used. The fully compressible, density-based solver allows an accurate mapping of the gas dynamics and an efficient parallelization due to turbulent boundary conditions are generated by an efficient implementation of Klein’s turbulence generator and superimposed on the gas dynamic boundary conditions. A PFGM model (Premixed Flamelet Generated Manifolds) is used for combustion simulation. The fuel spray is described by punctiform Lagrangian particles with mass, which are initialized after primary decay in such a way that they reflect the droplet statistics in detail, i.e. have the same compound probability density function for size, location and velocity vectors. Various strategies exist for the exact description of the ignition. For stable operating conditions, it can be assumed that the ignition is mapped sufficiently accurately on the already fine grids. In unstable and critical cases, however, an embedded DNA should be used to determine the initial development of the flame core immediately after ignition and to map it to the LES – respectively (phase 2) also by the hybrid flame-let / transported-FDF description.

The validation for stable operation will essentially be carried out by classical comparisons, firstly of pressure curves from experiment and simulation, then based on velocity and existing scalar profiles across the combustion chamber diameter. For unstable operation with cyclic fluctuations, validation is more difficult and requires more computing time. Comparisons must then be made using statistics conditioned to “strong” or “weak” cycles. Subsequently, relevant data of the flow and scalar fields (mixture, reaction progress, enthalpy defect) shall be investigated. At first, investigations of the tumble vortex and tumble decay are planned. For this purpose, the vortex trajectories will be followed and integral length measurements in the combustion chamber will be evaluated to better compare individual cycles. In the same way, the speed, shear rate, turbulence level and equivalence ratio at the location of (laser) ignition will be recorded and evaluated to obtain information about their correlation with strong or weak cycles. During the simulations interesting events are observed which lead to particularly high or low peak pressures – e.g. misfiring in lean mixtures, high flow velocities at the ignition site, or knocking in hotspots in gases which were previously in contact with the hot exhaust valve. The cause of such events is difficult to trace even in simulations, which is why an innovative backward analysis is to be developed. This is based on the hypothesis that the causes of interesting events are transported convectively – in other words, events must then be traced “convectively”.

Prof. Dr. Ing. Andreas Kempf andreas.kempf(at)
Linus Engelmann