Computation of an unsteady complex geometry flow using novel non-linear turbulence models

Non-linear zonal turbulence models are applied to an unsteady complex geometry flow. These are generally found to marginally improve predicted turbulence intensities. However, relative to linear models, convergence is mostly difficult to achieve. Clipping of some non-linear Reynolds stress components is required along with velocity field smoothing or alternative measures. Smoothing is naturally achieved through multilevel convergence restriction operators. As a result of convergence difficulties, generally, non-linear model computational costs detract from accuracy gains. For standard Reynolds stress model results, again computational costs are prohibitive. Also, mean velocity profile data accuracies are found worse than for a simple mixing length model. Of the non-linear models considered, the explicit algebraic stress showed greatest promise with respect to accuracy and stability. However, even this shows around a 30% error in total (the sum of turbulence and unsteadiness) intensity. In strong contradiction to measurements the non-linear and Reynolds models predict quasi-steady flows. This is probably a key reason for the total intensity under-predictions. Use of LES in a non-linear model context might help remedy this modelling aspect.