Best Practice Advice AC2-10: Difference between revisions
Line 20: | Line 20: | ||
In the following, we give some recommendations concerning the numerics and the physical models. Finally, some general considerations concerning the more practical aspects of IC engine simulations and the analysis of results are given. | In the following, we give some recommendations concerning the numerics and the physical models. Finally, some general considerations concerning the more practical aspects of IC engine simulations and the analysis of results are given. | ||
===Numerics=== | ===Numerics=== | ||
The use of second or higher order schemes minimizes numerical dissipation and this allows to accurately resolve the large-scale flow structures. Further, dissipation is exclusively controlled by the SGS model. This approach can be compromised by any upwind contribution that increases the amount of numerical dissipation. Formally, schemes with upwinding (QUICK, TVD, ENO, ...) can still achieve high orders of accuracy, but this order is mainly relevant for grid convergence - something that is (almost) never achieved with LES, where the underlying solution is typically poorly resolved, as the underlying solution will gain smaller scales with a smaller filter-scale or grid-refinement. Similarly, time-integration schemes that induce dissipation (e.g. non-central schemes with a bias to an implicit scheme) should be avoided. For engine cases where higher Ma numbers occur (e.g. detonation waves), robust schemes are required, and one option is Ma-number dependent blending as demonstrated by \cite{Janas2017}. | |||
At the present time, only very few engine LES have shown the ability of achieving robust momentum transport with central schemes without artificial stabilization (which in turn increases the amount of numerical dissipation) - but these tend to reduce accuracy near the walls instead \cite{Nguyen2017}. This means that there is still a clear need for further research on numerical schemes for engine LES! | |||
Quality estimators, for example by comparing the resolved to the modeled turbulent kinetic energy, cannot be trusted. Much work has shown, e.g. \cite{Klein2005,Klein2008} or \cite{Nguyen2017} that the criteria tend to be optimistic, and that they are usually misleading if used with (dissipative) non-central schemes: in these cases, numerical dissipation will reduce the turbulent kinetic energy on the smallest resolved scales, so that a reduced sub grid-energy is predicted - falsely implying that the unresolved part of the turbulent kinetic energy is low, whereas it is only underestimated due to poor numerics. | |||
Depending on the amount of numerical dissipation, combustion models may need to be adjusted, in particular the modelling of wrinkling factors, and effective grid filter-widths may need to be considered, which may be higher than the “ideal” (Schumann) case where the filter width is proportional to the cell-size \cite{Mercier2015}. | |||
===Physical models=== | ===Physical models=== | ||
===General considerations=== | ===General considerations=== |
Revision as of 14:43, 9 October 2018
Internal combustion engine flows for motored operation
Application Challenge AC2-10 © copyright ERCOFTAC 2024
Best Practice Advice
Key Fluid Physics
Application Uncertainties, Computational Domain and Boundary Conditions
Physical Modeling, Numerics and Recommendations
From experience and previous studies in the literature, it is quite obvious that the intrinsic unsteadiness and anisotropy of the flow field in IC engines renders scale-resolving approaches such as LES more suitable than RANS for the numerical simulation of such configurations, even though the high computational costs remain a real issue for routine design and optimization procedures.
Three simulations of the same test case were presented. However, a direct comparison is challenging because
- there are many differences between the simulations concerning physical models and numerical approaches and
- the in-cylinder flow processes are coupled non-linearly.
Numerics and physical modeling are closely intertwined with each other in scale-resolving simulations. In the view of the authors, dissipation should be prescribed by the subgrid scale model and not by the numerical scheme. This implies that scale separation through implicit filtering is defined by the turbulent viscosity rather than the numerical diffusivity. Other approaches and viewpoints consider a much closer coupling of numerics and physical modeling, which might make it even more difficult to quantify the individual contributions and draw general conclusions.
In the following, we give some recommendations concerning the numerics and the physical models. Finally, some general considerations concerning the more practical aspects of IC engine simulations and the analysis of results are given.
Numerics
The use of second or higher order schemes minimizes numerical dissipation and this allows to accurately resolve the large-scale flow structures. Further, dissipation is exclusively controlled by the SGS model. This approach can be compromised by any upwind contribution that increases the amount of numerical dissipation. Formally, schemes with upwinding (QUICK, TVD, ENO, ...) can still achieve high orders of accuracy, but this order is mainly relevant for grid convergence - something that is (almost) never achieved with LES, where the underlying solution is typically poorly resolved, as the underlying solution will gain smaller scales with a smaller filter-scale or grid-refinement. Similarly, time-integration schemes that induce dissipation (e.g. non-central schemes with a bias to an implicit scheme) should be avoided. For engine cases where higher Ma numbers occur (e.g. detonation waves), robust schemes are required, and one option is Ma-number dependent blending as demonstrated by \cite{Janas2017}.
At the present time, only very few engine LES have shown the ability of achieving robust momentum transport with central schemes without artificial stabilization (which in turn increases the amount of numerical dissipation) - but these tend to reduce accuracy near the walls instead \cite{Nguyen2017}. This means that there is still a clear need for further research on numerical schemes for engine LES!
Quality estimators, for example by comparing the resolved to the modeled turbulent kinetic energy, cannot be trusted. Much work has shown, e.g. \cite{Klein2005,Klein2008} or \cite{Nguyen2017} that the criteria tend to be optimistic, and that they are usually misleading if used with (dissipative) non-central schemes: in these cases, numerical dissipation will reduce the turbulent kinetic energy on the smallest resolved scales, so that a reduced sub grid-energy is predicted - falsely implying that the unresolved part of the turbulent kinetic energy is low, whereas it is only underestimated due to poor numerics.
Depending on the amount of numerical dissipation, combustion models may need to be adjusted, in particular the modelling of wrinkling factors, and effective grid filter-widths may need to be considered, which may be higher than the “ideal” (Schumann) case where the filter width is proportional to the cell-size \cite{Mercier2015}.
Physical models
General considerations
Recommendations for future work
The presented experimental and simulation work revealed several open issues within IC-engine simulations as discussed in subsections \ref{sec:uncertainties} and \ref{sec:recommendations} which can be summarized as follows:
- Further research is needed on robust and accurate numerical schemes for engine LES.
- Many models were originally developed for RANS applications and were subsequently adapted for LES. A thorough comparison of such physical models is needed to give clear recommendations which models or model combinations are best suited.
- There is still a lack of accurate boundary layer models for momentum and heat transfer.
References
Contributed by: Carl Philip Ding,Rene Honza, Elias Baum, Andreas Dreizler — Fachgebiet Reaktive Strömungen und Messtechnik (RSM),Technische Universität Darmstadt, Germany
Contributed by: Brian Peterson — School of Engineering, University of Edinburgh, Scotland UK
Contributed by: Chao He , Wibke Leudesdorff, Guido Kuenne, Benjamin Böhm, Amsini Sadiki, Johannes Janicka — Fachgebiet Energie und Kraftwerkstechnik (EKT), Technische Universität Darmstadt, Germany
Contributed by: Peter Janas, Andreas Kempf — Institut für Verbrennung und Gasdynamik (IVG), Lehrstuhl für Fluiddynamik, Universität Duisburg-Essen, Germany
Contributed by: Stefan Buhl, Christian Hasse — Fachgebiet Simulation reaktiver Thermo-Fluid Systeme (STFS), Technische Universität Darmstadt, Germany; former: Professur Numerische Thermofluiddynamik (NTFD), Technische Universität Bergakademie Freiberg, Germany
© copyright ERCOFTAC 2018