Richard Jefferson-Loveday

LES of Impingement Heat Transfer on a Concave Surface

Wall-resolved and zonal numerical large eddy type simulations are performed for a round jet impinging on a concave hemisphere at Re = 23,000. The zonal method uses a near-wall k–l model and a Hamilton-Jacobi equation to match this to the large eddy simulation zone. To minimize numerical dissipation, a self-adaptive discretization (SDS) scheme is examined. Both second- (n = 2) and sixth- (n = 6) order-based central discretization schemes are tested. The characteristics of the schemes is assessed using two test cases: the development of a subcritical Tollmien-Schlichting (T-S) stability wave in a plane channel and the decay of homogenous, isotropic turbulence (DHIT). It is found, that Smagorinsky LES simulations tend to be too dissipative in the high wave-number region, even with the SDS scheme; hence, the SGS model is omitted. Significant flow feedback is observed for the hemisphere case. Both shear-layer excitation and stabilization is observed. Computed wall pressure coefficients for the zonal NLES method are encouraging; for the wall-resolved case the stagnation region value is overpredicted. Heat transfer for the wall-resolved and zonal large eddy simulations are encouraging. For both quantities the difference between the n = 2 and n = 6 schemes is small, and the modeling approach used appears to be more influential. It is concluded that the presence of feedback mechanisms should be considered when designing experiments and/or numerical simulations for this case, and that the importance of boundary conditions for LES should not be neglected.

Wall-Resolved LES and Zonal LES of Round Jet Impingement Heat Transfer on a Flat Plate

Numerical large-eddy simulation (NLES) is performed for a round jet impinging on a flat surface at a Reynolds number of Re = 23,000 for nozzle-to-plate spacings of H/D = 6 and 2, where H is the distance from the nozzle to the plate and D is the jet diameter. The Reynolds number has been set to match the experiments of Cooper et al. (Int J. Heat Mass Transfer, vol. 36, pp. 2675–2684, 1993). Two numerical large-eddy simulation approaches are examined. The first quasi-direct numerical simulation (DNS) approach resolves streaklike structures using fine near-wall grids; the second is the zonal approach of Tucker (Int J. Heat Fluid Flow, vol. 25, pp. 625–635, 2004), which uses the Wolfshtein k–l (Int J. Heat Mass Transfer, vol. 12, pp. 301–318, 1969) Reynolds-averaged Navier-Stokes (RANS) model near the walls and NLES elsewhere. A Hamilton-Jacobi equation is used to match the RANS region to the NLES zone. The use of a Spalart-Allmaras model leads to low levels of turbulent viscosity in the near-wall region. This is also observed when using detached-eddy (DES) when using a volume-based filter. The use of the standard DES filter based on maximum grid spacing prevents jet shear-layer transition. The k–l near-wall model maintains RANS levels of turbulent viscosity in the boundary layer. The results of both the near-wall quasi-DNS and hybrid RANS-NLES methods are generally encouraging.

Large Eddy Simulation for Turbines: Methodologies, Cost and Future Outlooks

Flows throughout different zones of turbines have been investigated using large eddy simulation (LES) and hybrid Reynolds-averaged Navier–Stokes-LES (RANS-LES) methods and contrasted with RANS modeling, which is more typically used in the design environment. The studied cases include low and high-pressure turbine cascades, real surface roughness effects, internal cooling ducts, trailing edge cut-backs, and labyrinth and rim seals. Evidence is presented that shows that LES and hybrid RANS-LES produces higher quality data than RANS/URANS for a wide range of flows. The higher level of physics that is resolved allows for greater flow physics insight, which is valuable for improving designs and refining lower order models. Turbine zones are categorized by flow type to assist in choosing the appropriate eddy resolving method and to estimate the computational cost.

Contrasting Code Performances for Computational Aeroacoustics of Jets

The numerical propagation of subcritical Tollmein-Schlichting (T-S), inviscid vortical and cut-on acoustic waves is explored. For the former case, the performances of the very different NEAT, NTS, HYDRA, FLUXp and OSMIS3D codes is studied. A modest/coarse hexahedral computational grid that starkly shows differences between the different codes and schemes used in them is employed. For the same order of discretization the five codes show similar results. The unstructured codes are found to propagate vortical and acoustic waves well on triangular cell meshes but not the T-S wave. The above code contrasting exercise is then carried out using implicit LES or Smagorinsky LES for and Ma = 0.9 plane jet on modest 0.5 million cell grids moving to circa 5 million cell grids. For this case, even on the coarse grid, for all codes results were generally encouraging. In general, the spread in computational results is less than the spread of the measurements. Interestingly, the finer grid turbulence intensity levels are slightly more under-predicted than those of the coarse grid. This difference is attributed to the numerical dispersion error having a favourable coarse grid influence. For a non-isothermal jet, HYDRA and NTS also give encouraging results. Peak turbulence values along the jet centreline are in better agreement with measurements than for the isothermal jets. Copyright

LES for Turbines: Methodologies, Cost and Future Outlooks

Flows throughout different zones of turbines have been investigated using Large Eddy Simulation (LES) and hybrid Reynolds-Averaged Navier-Stokes-LES (RANS-LES) methods and contrasted with RANS modelling, more typically used in the design environment. Cases studied include low and high-pressure turbine cascades, real surface roughness effects, internal cooling ducts, trailing edge cut-backs and labyrinth and rim seals. Evidence is presented that shows that LES and hybrid RANS-LES produces higher quality data than RANS/URANS for a wide range of flows. The higher level of physics that is resolved allows greater flow physics insight which is valuable for improving designs and refining lower order models. Turbine zones are categorised by flow type, to assist in choosing the appropriate eddy resolving method and to estimate computational cost.© 2013 ASME

Investigation of Wake Induced Transition in Low-Pressure Turbines Using Large Eddy Simulation

Modern ‘high-lift’ blade designs incorporated into the low pressure turbine (LPT) of aero-engines typically exhibit a separation bubble on the suction surface of the airfoil. The size of the bubble and the loss it generates is governed by the transition process in the separated shear layer. However, the wakes shed by the upstream blade rows, the turbulent fluctuations in the free-stream and the roughness over the blade complicates the transition process.The current paper numerically investigates the transition of a separated shear layer over a flat plate with an elliptic leading edge using large eddy simulations (LES). The upper wall of the test section is inviscid and specifically contoured to impose a streamwise pressure distribution over the flat plate to simulate the suction surface of a LPT blade. The influences of free-stream turbulence (FST), periodic wake passing and streamwise pressure distribution (blade loading) are considered.The simulations were carried out at a Reynolds number of 83,000 based on the length of the flat plate (S0 = 0.5m) and the velocity at the nominal trailing edge (UTE ∼ 2.55 m/s). A high turbulence intensity of 4% and a dimensionless wake passing frequency (fr = fwakeS0/UTE, where fwake is the dimensional wake frequency) of 0.84 is chosen for the study. Two different distributions representative of a ‘high-lift’ and an ‘ultra-high-lift’ turbine blade are examined. An in-house, high order, flow solver is used for the Large Eddy Simulations (LES). The Variational Multi-scale approach is used to account for the sub-grid scale stresses.Results obtained from the current LES compare favorably with the extensive experimental data previously obtained for the test cases considered. The LES results are then used to further explore the flow physics involved in the transition process, in particular the role of Klebanoff streaks and their influence on performance. The additional effect of surface roughness of the blade has also been studied for one of the blade loadings. The benefit that roughness can offer for highly loaded turbine blades is demonstrated.Copyright

Computational modeling of jets with co-flow

An integrated solution approach using Hamilton-Jacobi and Eikonal equations for providing turbulence length scales, sponge, numerical order and also blending control is shown to be promising. For all these aspects essentially distance functions from surfaces can be needed. Although, use of these integrated features focuses on jet noise problems and structured grids perhaps the greater potential is for more complex geometry problems with dense unstructured solution adaptive grids suited for say aircraft landing gear noise modeling. Studies are made exploring optimal scheme options in the high order Large Eddy Simulation (LES) code. Since, for jets, being able to model the initial shear layer instability is important, the propagation of a Tollmien-Schlichting (TS) wave is considered. Comparison with an exact analytical solution of the Orr-Sommerfeld equation for the TS wave is made. On relatively coarse grids (intended to most strongly test the discretization) the code gave excellent agreement with the analytical TS data. The code is also found to propagate well the TS wave through distorted embedded, overset grid sections. A new hybrid Implicit LES (ILES) – Reynolds Averaged Navier Stokes (RANS) method making use of the HamiltonJacobi equation is discussed and applied to jets with co-flow. Using the hybrid ILES-RANS method, the influence of swirl, synthetic chevrons, jet eccentricity, jet external surface taper and width of the co-flow region is explored. Jet centre line velocity decay predictions show encouraging agreement with established data. Results show the co-flow region thickness has a more startling influence on potential core region length than jet eccentricity. As expected, predictions show that with co-flow the acoustic waves are directed more in the downstream direction. Predictions suggest a thick co-flow region has a strong acoustic influence and for a thin co-flow jet eccentricity has a diminished effect on the acoustic field. Mild co-flow angle (<30 and produced by nozzle external surface taper) has little effect on centerline velocity decay.

Differential Equation Specification of Integral Turbulence Length Scales

A Hamilton–Jacobi differential equation is used to naturally and smoothly (via Dirichlet boundary conditions) set turbulence length scales in separated flow regions based on traditional expected length scales. Such zones occur for example in rim-seals. The approach is investigated using two test cases, flow over a cylinder at a Reynolds number of 140,000 and flow over a rectangular cavity at a Reynolds number of 50,000. The Nee–Kovasznay turbulence model is investigated using this approach. Predicted drag coefficients for the cylinder test-case show significant (15%) improvement over standard steady RANS and are comparable with URANS results. The mean flow-field also shows a significant improvement over URANS. The error in re-attachment length is improved by 180% compared with the steady RANS k-ω model. The wake velocity profile at a location downstream shows improvement and the URANS profile is inaccurate in comparison. For the cavity case, the HJ–NK approach is generally comparable with the other RANS models for measured velocity profiles. Predicted drag coefficients are compared with large eddy simulation. The new approach shows a 20–30% improvement in predicted drag coefficients compared with standard one and two equation RANS models. The shape of the recirculation region within the cavity is also much improved.

Evaluation of the SST-SAS Model for Prediction of Separated Flow Inside Turbine Internal Cooling Passages

The flow and heat transfer over a three-dimensional axisym-metric hill and rectangular ribbed duct is computed in order to evaluate the Shear Stress Transport - Scale Adaptive Simulation (SST-SAS) turbulence model. The study presented here is rele¬vant to turbine blade internal cooling passages and the aim is to establish whether SAS-SST is a viable alternative to other turbulence models for computations of such flows. The model investigated is based on Menter‘s modification to Rotta‘s k-kL model and comparison is made against experimental data as well as other models including some with scale resolving capability, such as LES, DES & hybrid LES-RANS. For the hill case the SAS model dramatically overpredicts the size of the separation bubble. The LES on the other hand proved to be more accurate even though the mesh is courser by LES standards. There is little improvement of SST-SAS compared with RANS. Broadly speaking all models predict streamwise ve¬locity profiles for the ribbed channel with reasonable accuracy. The cross-stream velocity is underpredicted by all models. Heat transfer prediction is more accurately predicted by LES than RANS, DES & SST-SAS on a mesh that is slightly coarser than required by LES standard, however it still exhibits significant er¬ror. It is concluded that more investigation of the SST-SAS model is required to more broadly assess its viability for industrial com¬putation.

On LES Methods Applied to Seal Geometries

Nine Large Eddy Simulation (LES) methods are used to simulate flow through two labyrinth seal geometries and are compared with a wide range of Reynolds-Averaged Navier-Stokes (RANS) solutions. These involve one-equation, two-equation and Reynolds Stress RANS models. Also applied are linear and nonlinear pure LES models, hybrid RANS-Numerical-LES (RANS-NLES) and Numerical-LES (NLES). RANS is found to have a maximum error and a scatter of 20%. A similar level of scatter is also found among the same turbulence model implemented in different codes. In a design context, this makes RANS unusable as a final solution. Results show that LES and RANS-NLES is capable of accurately predicting flow behaviour of two seals with a scatter of less than 5%. The complex flow physics gives rise to both laminar and turbulent zones making most LES models inappropriate. Nonetheless, this is found to have minimal tangible results impact. In accord with experimental observations, the ability of LES to find multiple solutions due to solution non-uniqueness is also observed.Copyright