Search Results

Tightening emission regulations and accelerating production cycles force engine developers to shift their attention towards virtual engineering tools. When simulating in-cylinder processes in commercial LDD DI engine development, the trade-off between run time and accuracy is typically tipped towards the former. High-fidelity simulation approaches which require little tuning would be desirable but require excessive computing resources. For this reason, industry still favors low-fidelity simulation approaches and bridges remaining uncertainties with prototyping and testing. The problem with low-fidelity simulations is that simplifications in the form of sub models introduce multi variable tuning parameter dependencies which, if not understood, impair the predictive nature of CFD simulations. In previous work, the authors have successfully developed a boundary condition dependent input parameter table.

The Direct Numerical Simulation (DNS) approach to solving the fundamental transport equations down to the smallest scales of motion is favorable should the requirement be a truly predictive solution of fluid dynamic problems, but the simulation run times are unacceptable for most practical industrial applications. Despite the steadily increasing computational capabilities, Reynolds Averaged Navier-Stokes (RANS) based frameworks remain the most commercially viable option for high volume sectors, like automotive. The sub models within RANS simplify the description of key physical phenomena and include several numerical constants. These so-called “tuning constants” introduce multivariable dependencies that are almost impossible to untangle with local sensitivity studies.