LES Engines

LES in 3D engine simulation

The 3D approach used at present for combustion simulations in piston engines is the RANS method (Reynolds Averaged Navier Stokes). The underlying hypothesis is that the operation of the engine can be statistically described by a single average engine cycle. Conceptually, the latter is the result of phases average of a large number of consecutive cycles. In 3D simulation, the resolution of statistically averaged Navier-Stokes equations then makes it possible to obtain, via just one engine cycle, a direct assessment of the main characteristics of the case considered (work supplied, pollutants generated). With current computation capacities, RANS simulations can be produced in a few hours, enabling their industrial use. They can then be used to compare various configurations and help to identify those with the best potential.
 
Despite their value as a design and advanced optimization tool, RANS-type approaches have limitations inherent in their basic hypotheses. In particular, they can only be applied to cases presenting low variations around the mean. In the presence of marked cyclic variations, such as those that are the subject of the current project, this type of hypothesis is not verified, one average cycle therefore no longer being sufficient to be able to correctly describe the characteristics of the case studied.
 
For the last few years, continuous improvements in computation capacities (and, in particular, in parallel machines) have led to the development of a new 3D simulation approach based on the Large Eddy Simulation (LES) technique. This is not based on a statistical approach, but consists in distinguishing flow eddies resolved on the computation grid from those that are not, and that must therefore be modeled. Formally, this is performed via application of a spatial filter to Navier-Stokes equations, with the resulting equations being solved by digital methods presenting a sufficient quality, combined with “subgrid” models for describing the effects of non-resolved phenomena.
This approach has been tried and tested for the prediction of non-stationary phenomena inaccessible to RANS methods, such as acoustic instabilities in gas turbine burners, in particular.
 
LES can be considered to be a "digital test" that further refines the results obtained using RANS, by providing a detailed view on the filtered completions of one or more engine cycles in order to optimize, among other things, the cyclic or transient variabilities between stable engine points.
 

The AVBP LES engine code

The AVBP LES code is being co-developed by IFP Energies nouvelles and CERFACS and is co-owned by them. Its main characteristics are:
• non-stationary, compressible and reactive NS solver;
• order 2 and 3 convective patterns;
• any fixed and mobile non-structured grids;
• parallel solver highly efficient on massively parallel machines, based on domain decomposition;
• choice of turbulence model: Smagorinsky, Wale, ksgs;
• choice of turbulent combustion model: TFLES, CFM-LES;
• pre- and post-processing interfacing to the majority of commercial software packages.

Recent results: the LESSCO2 project

In the context of the EC LESSCO2 project coordinated by IFP Energies nouvelles (http://project.ifp.fr/lessco2); ; a first LES simulation of 9 consecutive cycles of a controlled ignition engine was performed using the AVBP code, with IFP Energies nouvelles and CERFACS working in close collaboration.
The calculation time for a complete cycle (720°DV) was 120 h on 32 Xeon processors of a Linux cluster.

View the animation for a complete cycle (Flash format)

 
Results of simulations in a PSA IIE XU10 4-valve engine performed by IFP Energies nouvelles in LESSCO2:

Left: averaged flame front obtained by RANS; no wrinkling of the front by turbulence.
Right: LES result at the same angle: the front is wrinkled by the bigger structures and ~25% of the total flame surface is resolved.

Results of a LES simulation of 9 consecutive cycles in a II PSA IIE XU10 4-valve engine performed by IFP Energies nouvelles in LESSCO2.

Left: the simulated cylinder pressure curves fit within the experimental range.
Right: the simulated work variations qualitatively reproduce the sequences measured.
The XU10 experimental database does not contain any detailed study on cyclic variations.