This thesis presents a numerical study of turbulent flows in internal combustion engines (ICEs) with focus on selected modeling and physical aspects. All studies base on a substantial number of consecutive cycles (up to 100) generated for simplified as well as state-of-the-art engine setups. Throughout the work, the results are compared and validated to existing experimental data and results obtained by direct numerical simulation (DNS). One major aspect is to study cycle-to-cycle variations (CCVs).

Appropriate modeling strategies for ICEs are intensively discussed. One example is the most suitable treatment of the intake and the exhaust ports. Here, three different port modeling strategies are applied on a well-known experimental engine setup. Integral quantities are evaluated and the velocity components as well as their fluctuations are compared to existing experimental data. Furthermore, the accuracy of selected scale-resolving turbulence models and their capability to capture large-scale and small-scale fluctuations are analyzed. For this, three LES models (Smagorinsky, WALE and Sigma), one hybrid model (DES-SST) and one second-generation URANS model (SAS-SST) are applied to a simplified engine setup. The predicted averaged velocities and the resolved fluctuations are compared to each other and to reference data from DNS and experiment. The investigated key aspect is the models' capability to resolve CCVs.

A quasi steady state flow bench configuration is used to analyze the effect of the applied turbulence model and the numerical grid on the flow in the vicinity of the intake valve. For a detailed investigation with regard to the intake jet, the velocity field is transformed into a local jet-adapted coordinate system. Based on this transformation, three characteristic zones within the intake jet are identified. A simplified engine setup is used to quantify the cyclic variability of large-scale structures within the combustion chamber in a next step. For that purpose, a novel ad-hoc methodology is presented. This methodology (combining proper orthogonal decomposition and conditional averaging) groups the instantaneous flow fields into different subsets and allows a quantification of a large and a small-scale contribution to the total fluctuations. The generation of the large-scale tumble structure and its interaction to the piston boundary layer during the intake stroke is studied based on an experimental gasoline engine setup with a state-of-the-art cylinder head. The instantaneous and phase-averaged tumble structures within the 3D flow field are visualized. Based on specific values of the dimensionless wall normal distance, the thickness of the piston boundary layer is computed and its interaction with the large-scale tumble structure is studied. Finally, the tumble development during the compression stroke is considered based on two established experimental engine setups, for which benchmark data was made available. After a general evaluation of the phase-averaged and instantaneous tumble structures, the CCV is quantified. To quantify the kinetic energy stored by the in-cylinder charged motion, the phase-averaged tumble intensity is evaluated. The tumble development during the compression stroke is subdivided into four consecutive phases.

In summary, this thesis offers a significant advance in the evaluation of several modeling strategies. Furthermore, it contributes to a deeper understanding of the in-cylinder flow processes, especially during the intake and compression stroke.