Development and application of synthetic turbulence methods for computational fluid dynamics

Abstract

Synthetic turbulence methods are an important tool for the study of turbulent flows. They allow to reduce the computational effort of numerical simulations of fluid flows and thereby, improve the quality of simulations of complex flow problems. Contributing to the field of turbulence research, this thesis proposes two new methods of simulating turbulent flows using synthetic turbulence.

The methods developed in this work were tested for two scenarios of turbulent flow simulations. The first scenario was the numerical simulation of turbulent flow around a wing. For this simulation a synthetic turbulence method was developed, which generated an initial 3D turbulent wind field to initialise the simulation. Using a complex numerical setup it was possible to simulate the interaction of the synthetic turbulence field, representing atmospheric boundary layer (ABL) turbulence, with a wing on a relatively large range of scales. This method allows to simulate the influence of ABL turbulence on the aerodynamics of the wing, for example, at large angles of attack. In the second scenario a new method was developed to generate synthetic turbulence as inflow boundary condition for Large-Eddy Simulation (LES). A new method to generate anisotropy in the turbulence field was introduced, which allowed to prescribe 1D statistics of the turbulent flow independently. This method can be used, for example, for feeding synthetic turbulence into the interface between the Reynolds-Averaged Navier Stokes (RANS) and LES part of a hybrid RANS/LES.

For the first scenario, the generated turbulence was tested in a simple LES of decaying turbulence where it was found that the input statistics for the turbulence generator were reproduced very well. It was also shown that the statistical properties were maintained reasonably well during the simulation with the exception of fluctuations observed in the cross-correlations.

In order to investigate the quality of the turbulence generator further, the generated turbulence field was compared to data from an LES of the ABL. It was found that the synthetic turbulence was not able to represent the coherent structures present in a convective boundary layer, but apart from that the turbulence statistics from the synthetic turbulence and LES of the ABL agreed very well.

After studying the properties of the synthetic turbulence generator in detail, a synthetic turbulence field was generated for the initialisation of the simulation of the flow around a wing. In a complex setup of two different grid types (Cartesian and unstructured) and two different turbulence model types (LES and RANS), the development of the turbulence in the different numerical environments was studied. It was found that the change in grid characteristics led to a stronger dissipation of turbulence on the unstructured grid. No significant effect on the turbulence could be found when the turbulence model switched from LES to RANS mode, most likely due to the short time the turbulence was exposed to the RANS model.

For the second scenario, a new approach for generating anisotropic turbulence was developed. An extensive analysis of the statistics of the generated turbulence was carried out and the results showed very good agreement with the reference data from a Direct Numerical Simulation (DNS). The generated turbulence then served as inflow boundary condition in an LES of a channel flow. A strong influence of the statistical properties of the synthetic turbulence on the behaviour of the turbulence in the channel was found. Comparison to two established synthetic turbulence methods showed a similar performance of the new approach, which at the same time caused much less computational costs and allowed better control of the statistical parameters of the synthetic turbulence.