During the last decades, the improvements of both computational ability and numerical schemes have stimulated increasing industrial interest in the use of Large Eddy Simulations (LES) for practical engineering flow problems. However, almost all current approches cannot treat complex geometries at affordable cost to enable LES of industrial problems. A robust, parallel and efficient solver using a general unstructured grid & based on high order flux reconstruction formulation, which uses local reconstruction, is compact and written in differential form without a mass matrix, was developed and has proved the ability to get accurate LES results but using RANS scale meshes. This work is aimed at using flux reconstruction method to perform Large Eddy Simulations for complex geometries in more robust and highly efficient way. Both explicit Runge-Kutta method and implicit LU-SGS method are implemented with improvements as solvers for better performance on boundary layer meshes including large aspect ratio cells. The current solver is ported to GPU architectures and speed up ratios of different order accuracy are presented in this work. A local reconstruction method is introduced to generate high order curved boundary from readily available first order meshes. The large eddy simulations for low pressure turbine blade and low pressure turbine blade with endwall are presented in this work, resolved with total number of degree of freedoms up to 34 million to chieve fourth order accuracy using limited computational resource. The results show that this approach has the potential to obtain LES results of real-geometry problems with affordable computational costs.
Performance of high-order implicit large eddy simulations
