Hybrid Turbulence Simulation to Predict Cyclic Variations (PDF)

Since the first passenger car with internal combustion (IC) engine was developed over 120 years ago, the device has been significantly improved regarding efficiency, emissions, smoothness and ease of use. Today IC-engines are used in roughly 850 million passenger cars worldwide. Even though many other concepts as e.g. fuel cells are investigated, it seems that no system can replace IC-engines in the near and intermediate future.

Two different combustion concepts are considered to have the potential to full fill future requirements with respect to fuel consumption and emission standards: turbo-charged diesel and stratified spark ignition (SI) engines with high pressure direct injection (DI) systems. Both systems can operate with overall lean air/fuel mixtures.

The first DISI-engine in a passenger car used a homogeneous air/fuel mixture. It was implemented in 1951 in the models Gutbrod Superior and Goliath GP 700 leading to a significant reduction in fuel consumption. The first application in mass production of direct injection systems in SI-engines was in 1997 in the Mitsubishi Carisma GDI (gasoline direct injection).

The greatest issues of stratified DISI-engines today, which give a much higher potential in fuel consumption economy compared to the homogeneous combustion concept, are combustion stability and emissions. Cycle-to-cycle variations of the gas motion have been identified to play a key role in the further optimization of the device since they have a great impact on the combustion process. Engine parameters are set according to the behavior of the mean cycle. However, the extreme engine cycles, cycles of greatest and slowest burning rates, determine the operating range of the engine. Consequently, the optimal spark timing, equivalence ratio and compression ratio are a compromise. A critical issue in stratified DISI-engines is that cyclic variations are substantial to the combustibility of the air/fuel mixture at the time of the discharge of the spark plug leading to partial burning or even misfire, which is undesirable in terms of engine roughness, efficiency and unburned hydrocarbon emissions.

Computational fluid dynamics (CFD) with common Reynolds averaged Naviers-Stokes (RANS) turbulence modeling has been established to be a very efficient and reliable tool within the design process of IC-engines. I. e. optimization of engine geometries can be accomplished with a short turnaround time. Additionally, insights into various physical processes can be gained that are difficult to study experimentally. However, this approach is limited by definition if unsteady features such as cycle-to-cycle variations are investigated and cannot capture this kind of phenomenon.

On the other hand, large eddy simulation (LES) provides the ability to predict cyclic variations because smaller spatial scales and temporal fluctuations are resolved. Since in LES a significantly smaller range of turbulent length scales needs to be modeled compared to the RANS approach, the accuracy of LES is superior to RANS. However, resolving smaller temporal and spatial scales requires higher order numerical schemes, smaller time steps and higher resolutions of the computational grids. This can lead to a significant increase of CPU time compared to RANS.

For wall-bounded turbulent flows at high Reynolds number and in complex geometries hybrid RANS/LES approaches have become more and more popular in the recent years. They combine attractive features of both methods. These methods provide the opportunity to use LES in regions, where its performance is known to be essentially superior to RANS. In other regions, where the accuracy and the averaged information on turbulent properties is sufficient, RANS can be used in order to save CPU-time. In contrast to pure RANS temporal fluctuations can be resolved in the LES regions in hybrid methods giving these approaches the potential to predict cycle-to-cycle variations or other turbulent flows of highly unsteady nature.

The present work focuses on unsteady turbulent flow phenomena in IC-engines such as cyclic variations of the gas motion and investigates the ability of subgrid turbulence modeling to predict those. In Chapter 2 the basic physical principles of fluid dynamics and turbulent flows are described both phenomenologically and based on the underlying governing equations. Furthermore, a review of filtering operations applied to the Navier Stokes equations and state of the art turbulence modeling is given. The different methods as well as the corresponding specific treatment of the boundary conditions of conventional RANS simulation and LES are presented and the hybrid RANS/LES method is introduced.

The numerical requirements for the hybrid approach in terms of spatial and temporal schemes as well as the meshing method that is needed for the computation of flows in complex geometries with moving boundaries as in IC-engines are described in Chapter 3. Different numerical schemes of the CFD code CFX, which is used in this work, are evaluated and tested against the numerics of other commercial and academic codes.

In Chapter 4 the hybrid method is tested against measurements and data of direct numerical
simulation (DNS) for simple flow cases. For a fundamental evaluation of the approach classic turbulence test cases such as the decay of homogeneous isotropic turbulence and the flow past a backward-facing step are used. The most relevant flow configurations in engine development are the steady flow through an intake port/valve assembly and the transient flow in a reciprocating engine. However, before the hybrid method is applied to these complex turbulent flows in IC engines at high Reynolds number, simplified configurations of theses cases are investigated. The hybrid RANS/LES method is compared to RANS and LES computations in terms of accuracy and level of information on turbulence properties.

Chapter 5 is dedicated to flows in IC-engines. The specific flow characteristics are described and quantified and key issues in engine design are discussed. The hybrid RANS/LES method is used for the computation of the steady flow through an intake port and the multi-cycle simulation of the flow in a series production BMW engine. Optical measurements are used to evaluate the quality of the averaged flow field of the simulation as well as the ability to predict cyclic variations of the gas motion in IC-engines.