| Summary: | To date no generally accepted prediction method is available for a low pressure turbine designer. Detailed information about the flow field that can be used to improve turbulence models is difficult to obtain from experiments. Direct numerical simulations (DNS), capable of delivering that information, have been published and showed the ability of this method to accurately predict the flow field. However, in these previous studies detailed turbulence information was not published and to date detailed information about turbulence in LPT flows is missing. In this work compressible DNS were conducted and turbulence was investigated in detail. First a method was developed that enables large scale numerical simulations of low pressure turbines on state of the art high performance computers. Particular emphasis was put on boundary conditions with respect to reflections and an efficient way of generating inlet turbulence. To allow for efficient computation the computational performance was examined and optimized and showed good scaling up to large numbers of cores and GPUs, allowing the exploitation of large parallel computing systems. A grid convergence study is presented showing that while large scales are grid converged at similar resolutions to previously published work, convergence of quantities related to ne scales required a substantially higher resolution than published work. Budgets of the transport equations of Favre averaged momentum, total energy and TKE are presented in the vortex shedding region, the developed wake and the blade boundary layer. These are useful for turbulence modellers to understand the accuracy of a particular turbulence model. Energy transport mechanisms were discussed to provide an understanding of the important mechanisms and their variation with Reynolds number. Finally a local assessment of linear eddy-viscosity models was presented where a turbulence model based on an optimized turbulence viscosity was compared to the standard k- model. It was found that in the vortex shedding region there was scope for improvement of the k- model. However, the remaining modelling error of the optimized model was large highlighting the limitation of linear eddy viscosity models. Further, regions with large errors in the turbulence model were identified close to leading and trailing edge.
|
|---|
