In this paper, we compare the capabilities of two open source near-wall-modeled large eddy simulation (NWM-LES) approaches regarding prediction accuracy, computational costs and ease of use to predict complex turbulent flows relevant to internal combustion (IC) engines. The applied open source tools are the commonly used OpenFOAM, based on the finite volume method (FVM), and OpenLB, an implementation of the lattice Boltzmann method (LBM). The near-wall region is modeled by the Musker equation coupled to a van Driest damped Smagorinsky-Lilly sub-grid scale model to decrease the required mesh resolution. The results of both frameworks are compared to a stationary engine flow bench experiment by means of particle image velocimetry (PIV). The validation covers a detailed error analysis using time-averaged and root mean square (RMS) velocity fields. Grid studies are performed to examine the performance of the two solvers. In addition, the differences in the processes of grid generation are highlighted. The performance results show that the OpenLB approach is on average 32 times faster than the OpenFOAM implementation for the tested configurations. This indicates the potential of LBM for the simulation of IC engine-relevant complex turbulent flows using NWM-LES with computationally economic costs.

An investigation of the interaction between the in-cylinder flow and the spray topology in two spray-guided direct injection optical engines is reported. The bulk flow field in the combustion chamber is characterized using particle image velocimetry. Geometrical parameters such as the axial penetration and the spray angle of the liquid spray are measured using Mie scatter imaging and/or diffuse back-illumination. The measured parameters are compared with data from a constant volume chamber available in the literature. For a late injection strategy, the so-called ECN Spray G standard condition, the mean values of the spray penetration do not seem to be significantly perturbed by the in-cylinder flow motion until the plumes approach the piston surface. However, spray probability maps reveal that cycle-to-cycle fluctuations of the spatial distribution of the liquid spray are affected by the magnitude of the in-cylinder flow. Particle image velocimetry during injection shows that the flow field in the vicinity of the spray plumes is heavily influenced by air entrainment, and that an upward flow in-between spray plumes develops. Consistent with previous research that demonstrated the importance of the latter flow structure for the prevention of spray collapse, it is found that increased in-cylinder flow magnitudes due to increased intake valve lifts or engine speeds enhance the spray-shape stability. Compared with cases without injection, the influence of the spray on the in-cylinder flow field is still noticeable approximately 2.5 ms after the start of injection.

Near-wall processes in internal combustion engines strongly affect heat transfer and pollutant emissions. With continuously improving capabilities to model near-wall processes, the demand for corresponding measurements increases. To obtain an in-depth understanding of the near-wall processes within spark-ignition engines, flame distributions and flow fields were measured simultaneously near the piston surface of an optically accessible engine operating with homogeneous, stoichiometric isooctane–air mixtures. The engine was operated at two engine speeds (800 rpm and 1500 rpm) and two different intake pressures (0.95 bar and 0.4 bar). Flame distributions were obtained at high spatial resolution using high-speed planar laser induced fluorescence of sulfur dioxide ( $$\hbox {SO}_{{2}}$$SO 2 ). Particle tracking velocimetry was utilized to measure the flow field above the piston at high spatial resolution, which enabled the determination of hydrodynamic boundary layer profiles. Flame contours were extracted and statistical distributions of the burnt gas area determined. The burnt gas distributions were compared with the simultaneously recorded high-speed flow field measurements in the unburnt gas. A direct comparison with motored engine operation showed comparable boundary layer profiles until the flame approaches the wall. Flow acceleration due to flame expansion rapidly increases velocity gradients and the boundary layer development becomes highly transient. The interaction of flame and flow depends on the operating conditions, which results in a different evolution of burnt gas positions within the field-of-view. This has additional implications on the development of the velocity boundary layer. Depending on the operating conditions, the flame strongly affects the velocity boundary layer profiles resulting in boundary layer thicknesses (defined by 50% maximum velocity) in the order of $$80{-}180\, \upmu \hbox {m}$$80 - 180 μ m .

Machine learning (ML) models based on a large data set of in-cylinder flow fields of an IC engine obtained by high-speed particle image velocimetry allow the identification of relevant flow structures underlying cycle-to-cycle variations of engine performance. To this end, deep feature learning is employed to train ML models that predict cycles of high and low in-cylinder maximum pressure. Deep convolutional autoencoders are self-supervised-trained to encode flow field features in low dimensional latent space. Without the limitations ascribable to manual feature engineering, ML models based on these learned features are able to classify high energy cycles already from the flow field during late intake and the compression stroke as early as 290 crank angle degrees before top dead center ( − 290 ° CA ) with a mean accuracy above chance level. The prediction accuracy from − 290 ° CA to − 10 ° CA is comparable to baseline ML approaches utilizing an extensive set of engineered features. Relevant flow structures in the compression stroke are revealed by feature analysis of ML models and are interpreted using conditional averaged flow quantities. This analysis unveils the importance of the horizontal velocity component of in-cylinder flows in predicting engine performance. Combining deep learning and conventional flow analysis techniques promises to be a powerful tool for ultimately revealing high-level flow features relevant to the prediction of cycle-to-cycle variations and further engine optimization.