Repository logo

Towards robust unstructured turbomachinery large eddy simulation

Change log


Watson, RA 
Tucker, PG 
Wang, ZN 
Yuan, X 


Industrial legacy codes usually have had long pedigrees within companies, and are deeply embedded into design processes. As the affordability and availability of computing power has increased, these codes have found themselves pushed into service as large eddy simulation solvers. The approximate Riemann solver of Roe, which is frequently used as the core method in such legacy codes, is shown to need much user care when adopted as the discretisation scheme for large eddy simulation. A kinetic energy preserving (KEP) scheme—which retains the same advantageous stencil and communications halo as the original Roe scheme—is instead implemented and tested. The adaptations of code required to switch between the two schemes were found to be extremely straightforward. As the KEP scheme intrinsically bounds the growth of the kinetic energy, it is significantly more stable than the classical non-dissipative schemes. This means that the expensive smoothing terms of the Roe scheme are not always necessary. Instead, an explicit subgrid scale turbulence model can be sensibly applied. As such, a range of mixed linear–non-linear turbulence models are tested. The performance of the KEP scheme is then tested against that of the Roe for canonical flows and engine-realistic turbine blade cutback trailing edge cases. The new KEP scheme is found to perform better than the original in all cases. A range of mesh topologies: hexahedral; prismatic; and tetrahedral; are also tested with both schemes, and the KEP scheme is again found to perform significantly better on all mesh types for these flows.


This is the final version of the article. It first appeared from Elsevier via


Large eddy simulation, Kinetic energy preserving, Roe, Industrial, Unstructured, Low Mach

Journal Title

Computers and Fluids

Conference Name

Journal ISSN


Volume Title



Elsevier BV
Engineering and Physical Sciences Research Council (EP/H001395/1)
This work was supported by an iCASE studentship from the Engineering and Physical Sciences Research Council, via Rolls-Royce plc. The funding from both organisations is gratefully acknowledged.