Evaluation of scale resolving turbulence generation methods for Large Eddy Simulation of turbulent flows

•Artificial turbulence generation is examined for homogeneous isotropic turbulence.•Prescribed integral length scale and root-mean-square velocity are reproduced.•Spectral content of HIT is fully represented for one algorithm.•Energy spectra represent only large scale structures for two other algorithms.•These spectra relaxe to physical state during decay of turbulence. Large Eddy Simulation (LES) has become an attractive simulation method even for technical processes and it usually provides space and time resolved fluctuations of a significant portion of the spectrum. However, in contrast to a RANS simulation an accurate LES requires the definition of suitable initial and boundary conditions, which includes turbulent structures with physically sound spatial and temporal correlations. Such turbulent structures are usually generated artificially at the boundary. Three different algorithms for generating turbulent fluctuations are evaluated in the present work. The investigated methods are Filtered noise [1], Diffused noise [2] and an Inverse Fourier approach [3,4]. These techniques were developed for generating inflow data for LES and have already been used in published research [5–15], e.g. for investigating turbulent combustion processes. In the present work the turbulent statistics i.e. energy spectra and velocity correlations as well as derived quantities such as turbulent kinetic energy and subgrid scale viscosity are investigated in more detail in a comparative fashion for the generated turbulent velocity fields. As a simple test case, the decay of turbulence in a cubical box, is considered here to provide information on the initially generated turbulence as well as its temporal evolution. The results are analyzed in detail and are compared to experimental data. Turbulence fluctuations generated by Filtered noise and Diffused noise lead to similar results. The resulting energy spectra and velocity correlations agree generally well with experimental data despite some discrepancies at very early times after initialization. The Inverse Fourier approach yielded good agreement at all times, but at increased computational cost. In addition, the implementation of Filtered noise and Diffused noise might be easier for most cases of practical interest. In particular, the Diffused noise approach can be used for the generation of inhomogeneous turbulence on arbitrary grids.