Implementation of an Efficient Synthetic Inflow Turbulence-Generator in the Open-Source Code OpenFOAM for 3D LES/DNS Applications

A new inflow boundary condition (BC) has been implemented into the open-source CFD code OpenFOAM, which generates synthetic turbulent fluctuations at the inlet boundary for 3D transient simulations. The method is based on convolution of digital random data series. The filter coefficients of the convolution process prescribe a two-point correlation function that possesses the basic properties of real turbulent flow. In this way, spatially and temporally correlated flow fields with specified bulk flow rate, turbulence intensity, turbulent length and time scales can be generated. Compared to previous implementations, the new turbulence generator is computationally more efficient by using coarse virtual grids and can be used for arbitrarily shaped inlets. Compared to OpenFOAM’s native turbulence generator, which shows some anomalies during parallel runs, the new implementation gives consistent results even for large-scale parallel simulations. The inlet BC has been applied to two turbulent combustion cases with Large Eddy Simulation (LES) and Direct Numerical Simulation (DNS), using up to 8192 CPU cores on Hazel Hen at HLRS. The results reveal the significance of the inflow turbulence for reproducing the correct flame structure. A performance analysis of intra and inter-node performance on the Vulcan and Hawk clusters shows that the OpenFOAM solver is memory bound. Therefore, higher performance is reached when only half of the AMD CPU cores per node are utilized on Hawk because the L3 cache is shared by a core complex (CCX) and each core has a relatively low bandwidth. The simulation scales super-linearly on Hawk and reaches ideal speedup down to 8 000 computational cells per MPI rank, which is consistent with scaling results on the previous system Hazel Hen. The implementation of the BC is described in full detail.