Novel hybrid LES-RANS schemes for simulating physically and geometrically complex turbulent flows

Lead Research Organisation: University of Manchester
Department Name: Mechanical Aerospace and Civil Eng


Large Eddy Simulation (LES) is gradually replacing traditional Reynolds-averaged-Navier-Stokes (RANS) modelling as the method of choice for predicting complex turbulent flows in research as well as industrial practice. This is especially so when unsteady phenomena are to be resolved (vibrations, acoustics, thermal striping, pressure peaks). However, the exploitation of LES for predicting practical wall-confined flows, particularly those involving separation from curved surfaces, is seriously inhibited by practically untenable resource requirements at high Reynolds numbers. Hybrid LES-RANS schemes, employing some form of RANS-like solution in the near-wall region, are generally regarded as a compromise strategy circumventing the resource obstacle. Existing schemes are based on the use of RANS models that operate in unsteady mode, as they are subjected to high amplitude, high-frequency fluctuations imposed on the layer by the outer LES solution. These models thus operate far outside their intended range of applicability. Moreover, in most approaches, the small-scale motions not resolved explicitly by the LES are represented by an ill-defined blend of subgrid-scale and RANS turbulence models - i.e. there is no clear dividing line between the LES and RANS components. Not surprisingly, such models display a whole range of disconcerting defects.This submission proposes a collaboration between two groups who have been at the forefront of developing RANS-LES schemes in the UK. Indeed, the two groups are the only UK academic partners who have participated in the four-year EU FP6 project DESider, specifically devoted to RANS-LES modelling for industrial applications, and in the follow-up 22-partner FP7 project ATAAC (Advanced Turbulence Simulation for Aerodynamic Application Challenges). Electricite de France (EDF) will support the programme to the level of one man-year of PDRA.The proposed project aims specifically at LES-RANS hybrids that distinguish carefully between the LES and RANS elements, each applied subject to appropriate, well-established constrains and coupled rationally. The project involves two major strands: (i) the development of a novel zonal (two-layer) scheme, which entails the solution of steady, parabolized RANS equations, subject to on-the-fly time-averaged constraints derived from the LES solution, and the use of an anisotropy-resolving turbulence model over a thin near-wall layer superimposed onto the LES domain; (ii) the integration and validation of (i), as well as an extended version of a newly-developed RANS-LES hybrid (Uribe et al [2007]), which shares some basic concepts with proposed model under (i), into a state-of-the-art numerical framework (Saturne), which is promoted by EPSRC's CCP12 as a general prediction tool for computing turbulent flows in very complex geometries on HPCx and HECTOR. A key characteristic of Uribe et al's model is that it respects the need to separate the RANS-derived Reynolds stresses from the inherently unsteady LES, and to desensitize the resolved perturbations and the subgrids-scale stresses from the RANS model. To that extent, the model is based on the same philosophy underpinning the zonal scheme to be developed, although the two models differ radically in respect of their design.


10 25 50
Description In turbulent Computational Fluid Dynamics (CFD) Direct Numerical Simulation (DNS) is a deterministic approach resolving the entire range of turbulent flow structures down the tiniest ones that molecular viscosity will allow. This requires 100 million-cells meshes and precise numerical methods, but no modelling assumption is made. DNS, a "flawless numerical experiment", is however limited to plain and small-extent geometries (only 10 to 100 times the largest eddy) with simplistic inlet conditions (often periodic: inflow = outflow).

Inversely, the industrial CFD workhorse is Reynolds Averaged Navier Stokes (RANS) tentatively predicting only statistical values - mean velocity, temperatures and their variance - which is fast and, with fully unstructured meshes, applicable to very complex geometries.

To study less trivial and larger geometries UoM and EDF have jointly developed over 10 years coupling of RANS and Large Eddy Simulation (LES is like DNS but with still some statistical model for the smaller eddies). The method is based on correcting the stresses in the LES using those given by the RANS on coarse mesh regions, and vice-versa, has successfully predicted e.g. thermal mixing in a T pipe junction, separations in a diffuser or on a blade at incidence
Exploitation Route This has contributed to a series of specialised conferences,
"Symposium on Hybrid RANS-LES Methods "
first in the Eu and now world-wide, for applications in any fluid flow simulation related topical area (US) (China)
Sectors Aerospace, Defence and Marine,Agriculture, Food and Drink,Chemicals,Digital/Communication/Information Technologies (including Software),Energy,Environment,Manufacturing, including Industrial Biotechology,Transport

Description Combining both, statistical (RANS) & deterministic (LES) representations of turbulence, enables localised refined studies embedded in global industrial simulations. Developed via EU projects with major aerospace industries as EADS, Dassault ... (DESIDER, ATAAC, continuing with Go-Hybrid). Extended to heat transfer, it is used with EDF for thermal stresses impact on power plant lifetime. CD-Adapco, 2nd worldwide CFD software vendor, is implementing it for local-global effect of e.g. car mirrors acoustics and drag. Following, Academics now recognised the need to use this "two fields" approach since combining RANS and LES in a single field fails for physical reasons.
First Year Of Impact 2012
Sector Aerospace, Defence and Marine,Digital/Communication/Information Technologies (including Software),Energy,Transport
Impact Types Economic,Policy & public services