State-of-the art engine technologies are susceptible to high cycle-to-cycle variability. Researchers have successfully used Large Eddy Simulations (LES) to capture this cyclic variation with CFD. However, LES is computationally expensive. The current work demonstrates that using RANS turbulence models can also exhibit cyclic variation if the simulation approach minimizes numerical viscosity. This is accomplished by using fine mesh resolution, non-morphing mesh motion, higher-order accurate numerical schemes, and small timesteps.
RANS turbulence models act to destroy time-varying smaller eddies and replace the mixing effects of these eddies with enhanced viscosity. In an IC engine, larger-scale eddies can change from cycle to cycle, and may not be small enough to be dampened out by the RANS turbulence viscosity. By minimizing the numerical viscosity, the length scale at which eddies are destroyed is reduced and more structure is seen in the simulated flowfield. If the injection and combustion strategy in an engine is susceptible to cyclic changes in these large-scale eddies, then cyclic variation will be apparent in the simulation when using a RANS model.
This work will also demonstrate that perturbations in initial conditions, boundary conditions, or numerical settings can give run-to-run variability in simulation consistent with cycle-to-cycle variability in an actual engine.
For the current work, three studies are performed to show that the use of a RANS turbulence model does not always yield an ensemble average result. One of the studies is a basic cylinder-in-cross-flow case. The other two studies are for engines. One of the engine studies focuses on global mixing parameters and compares to TCC (Transparent Combustion Chamber) experimental data. The other engine study looks at cycle-to-cycle variation in combustion predictions.