Cycle-to-cycle variations (CCVs) in engines limit the ability of engine designers to reach the theoretical limits of engine efficiency. This study investigates CCVs in Direct-injection Spark-ignition (DISI) engines through Large-eddy Simulations (LES). Multi-cycle simulations of motored engine flow and spray simulations with variable boundary conditions were performed. The Dynamic Structure turbulence model, which is an advanced 1-equation non-viscosity turbulence model, was used to enable coarser, engineering type meshes and reasonable computational requirements.
Multi-cycle motored engine simulations were run for an optical engine at several different engine operating conditions. Comparisons included both pressure and velocity measurements using Particle Image Velocimetry (PIV). Simulations were run using computational domains that either did or did not include intake and exhaust mixing plenums. Results from runners-only and full-domain simulations were overall similar, but there were differences in specific flow structures at certain times. Changes to engine speed or manifold pressure increased flow magnitudes, even after adjusting for different mean piston speeds, but had relatively minor effects on the flow structure.
A spray model adapted from diesel spray simulations is presented. The adapted models were unable to match experimental trends with changing ambient density in both liquid and vapor phases simultaneously. Two spray break-up model parameters were changed to vary as functions of ambient density, which greatly improved the vapor predictions but worsened liquid predictions.
Two methods from Uncertainty Quantification (UQ) were used to test the response of the spray models to prescribed uncertainty in spray boundary conditions. The effects of having uncertainty in two numerical model parameters and two physical boundary conditions was examined. Overall simulation uncertainty was much larger than the experimental uncertainty. Further tests showed the uncertainty in the simulation response variables was due primarily to uncertainty in the numerical modeling parameters. When examining the effects of uncertainty in the physical boundary conditions alone, the resulting variability in the response variables was approximately equal to the variability in the spray measurements.