You are on page 1of 12

COMPUTATION OF GAS TURBINE BLADE FILM COOLING

W Vickery and H Iacovides* Turbulence Mechanics Group School of Mechanical, Aerospace and Civil Engineering The University of Manchester Sackville Street Manchester, M60 1QD UK *Email h.iacovides@manchester.ac.uk

ABSTRACT This study focuses on the efficient computation of film-cooling flows. The main objective is to establish how reliably film cooling flows can be predicted, using commercial software. A simplified jet-in-a-cross-flow problem has been focused on and the commercial CFD code, FLUENT, has been used. A two-equation turbulence model (the realizable k- model) and a two-layer zonal near-wall treatment method (the enhanced wall treatment in FLUENT) have been used. Two cases at a boundary layer Reynolds number of 400,000, which involve three cooling holes 3 diameters apart, with velocity ratio of 0.28 and density ratios of 1 and 1.53 have been examined. The density ratio of 1.53 is achieved by using CO2 as the coolant and air as the main stream fluid. Preliminary computations helped produce a solution domain which achieved considerable grid savings. The resulting predictions are compared with available experimental data and are shown to be able to reproduce most of the key features of the mean and also of the fluctuating motion. The mean mixture fraction of the coolant has been used to trace the jet trajectory. The footprint of the mixture fraction along the wall surface has been compared to experimental data of film cooling efficiency along the wall. The computed mixture fraction is in good agreement with the experimentally determined film cooling efficiency, suggesting that the mixture fraction can be used as an indicator of the film cooling efficiency.

INTRODUCTION Film cooling has long been established as one of the main mechanisms used to maintain the temperature of gas-turbine blades down to safe operating levels. Pressures to improve engine efficiency continuously drive engine temperatures to higher levels, which leads to the need for more effective blade-cooling systems. In modern gas turbines, blades are cooled both internally and externally, Han et al. (2000). Internal cooling includes a number of heat-transfer-enhancing mechanisms, like artificial surface roughness through the use of ribs, impingement and pin-fins Iacovides and Launder (2007). External cooling involves the use of film cooling, in which the objective is, through the injection of coolant from the blade surface, to set up a thermal barrier between the blade surface and the hot gas stream. The focus of this investigation is external film cooling. .Film cooling has been studied extensively, both numerically and experimentally. Studies on the subject are generally divided into two subjects: flow characteristics and heat transfer characteristics. The flow of the engines working fluid (or the free stream) over the turbine blade surface has a Reynolds number, Re (based on the free stream velocity and the distance from the leading edge) of approximately 400 000, according to Mehendale et al. (1994), and forms a turbulent boundary layer on the surface of the blade. The coolant jet is injected into this turbulent boundary layer and the cooling film is then formed downstream of the injection hole. The boundary layer thickness at the point of injection, , is approximately twice the diameter of the injection hole, D. A jet in cross-flow problem is characterised by several parameters that relate the coolant jet flow to the free-stream flow. The density ratio, DR, and velocity ratio, VR, are two of the more important parameters. It had been suggested by authors such as Jabbari and Goldstein (1978) and Kruse (1985) that the maximum effectiveness occurs at the injection hole; however the correlations suggested by those authors lead to an overestimation of the cooling effectiveness near the injection hole. The maximum adiabatic effectiveness, , defined as the ratio between the difference of the free-stream and wall temperatures divided by the difference between the free-stream and coolant temperatures, occurs at some distance downstream of the hole and there is also zero-effectiveness mid-span between a pair of injection holes Bauldauf et al. (2001). Bauldauf et al. (2001) found that as the angle of the coolant injection is increased, and the velocity component of the coolant perpendicular to the blade surface increases, the coolant is driven away from the wall and coolant-free stream interaction is intensified. This mixing with the free stream has an adverse effect on the effectiveness. It was also found that elevated turbulence levels lead to an increase in the coolant and free stream interactions and therefore a reduction in the effectiveness A recirculation region is formed immediately downstream of the injection hole. Jessen et al. (2007), among others, showed that increasing the blowing ratio, BR, leads to a larger recirculation zone. A larger recirculation zone encourages mixing of the free stream and the coolant and therefore leads to a larger proportion of freestream gas being present in the cooling film. A high density ratio, DR, results in improved lateral spreading of the flow downstream of the injection hole, and is therefore beneficial to the cooling process. In a real engine, the density ratio is determined by the temperature difference between the coolant and the free stream. Use of computational fluid dynamics, can greatly facilitate the optimisation of such systems. Optimisation tasks can only be undertaken, however, with flow solvers which have already been validated for film cooling applications. This has led to many attempts to predict film cooling flows. Garg et al. (1997) carried out a computational heat transfer study of film cooling using a 3D Navier-Stokes code. The results of this study were compared to experimental results that were carried out on a rotating test rig. It was found that the results were acceptable for the leading edge and suction side of the blade; however the surface heat transfer was underestimated on the pressure side of the blade. The reason for the underestimation was put down to unsteady stator-rotor effects that were not accounted for in the computational model. Renze et al. (2008) have carried out studies of a jet in cross-flow, using large eddy simulations (LES). The simulations were verified using experimental data from Jessen et al. (2007). This study was based around a turbulent, multi-species, non-reacting flow. Air was used for the freestream and carbon dioxide was used for the coolant. This was done in order to produce a density ratio other than DR=1. Conservation equations for partial densities had to be used along with the more usual Navier-Stokes equations. The results of this LES study agreed well with the experimental data and showed that a LES is appropriate for the study of film cooling. This study aims to contribute to this validation process, by employing the widely used commercial flow simulation package FLUENT to predict the flow development that results from coolant injection into the flow over a flat surface, from a single row of inclined injection holes, over a range of momentum and density ratios. The recently produced detailed data by Jessen et al. (2007) are used for validation purposes.

EQUATIONS OF FLUID MOTION Mean Flow Equations The starting point of any computational analysis is, of course, the set of equations that result from the laws of mass, momentum and heat transport. For RANS treatments the equations may be presented in the compact Cartesian notation form: Mass Conservation Equation

Momentum and Enthalpy Transport Equations

Concentration of Species Transport Equation

Turbulence Modelling The turbulent stresses, heat and mass fluxes that appear in the momentum enthalpy and concentration transport equations respectively, are modelled through the effective-viscosity and effective diffusivity approximations shown in equations (5) to (7) below.

The effective viscosity approximation has a number of predictive weaknesses, in flows with streamline curvature, flows subjected to body forces and generally in flows in which the anisotropy of turbulence influences the development of the mean flow. Since, however, effective-viscosity models are the most widely used turbulence models in RANS simulations, especially in industry, it is appropriate, indeed essential, that any validation exercise starts by first focussing on this group of models. Here the variant of the well known k- model available within FLUENT, known as the realisable k- of Shih et al. (1995), has been used to provide the distribution of the turbulent viscosity, t, which is needed in the approximations of equations (5), (6) and (5). The turbulent Prandtl number, , is assigned the constant value of 0.9 and the turbulent Schmidt number C the value of 1. Since one of the main objectives here is the prediction of wall heat transfer, one issue that requires special consideration, is the modelling of the effects of near-wall turbulence on the transport of momentum and thermal energy across the wall boundary layers. Here the enhanced wall treatment option, available within FLUENT, has been employed, which for finer near-wall meshes, switches to a two-layer treatment with a low-Re 1-equation model of turbulence, Wolfstein (1969). Two-layer approaches have long been employed within the authors group in the computation of complex 3-dimensional flows, Iacovides and Launder (1984, 1987). They provide an often acceptable compromise between predictive accuracy and computational efficiency, since the 1-equation model one the one hand involves less restrictive approximations than log-law-based wall functions and on the other hand is not as expensive as two-equation low-Reynolds-number models.

CASES EXAMINED

Figure 1- Experimental setup of Jessen et al. (2007). Only one symmetric half shown. The flow has been specified as steady and incompressible and the second-order upwind scheme has been used for the discretization of convective transport. The experimental flow domain (one symmetric half) is presented in Figure 1. The hole spacing ratio, s/D, is 3 and the length of flow development upstream of the injection holes is 71.3 diameters, which results in a boundary layer thickness of 2D just before injection. The coolant injection tubes are 24 diameters long. The local Reynolds number, based on free stream velocity and distance from the leading edge is 400,000. Two isothermal cases, for which Jessen et al. (2007) provide experimental data, are presented. They are all for a free stream Reynolds number of 400,000, but for different velocity (VR), density (DR), blowing (BR) and momentum (IR) ratios. A density ratio higher than one was achieved by replacing air with CO2 as the coolant fluid, in order to reproduce the temperature effects on density under engine conditions. These are included in Table 1 below. Table 1. Details of cases computed Parameters VR [=VC/U] DR [=C/] BR [=(C VC)/( U)] IR [=(C VC2)/( U2)] Air Injection Case 1 0.28 1 0.28 0.08 CO2 Injection Case 2 0.28 1.53 0.43 0.12

NUMERICAL ASPECTS As mentioned earlier the commercial finite-volume code FLUENT was used in this investigation. Here it is used in incompressible mode with the second-order upwind scheme employed for the discretization of convective transport. A key numerical issue in the simulation of film cooling flows, is grid resolution of the flow domain. The challenge is to accurately resolve the flow, with the minimum number of grid nodes. In order to achieve this thee steps have been taken. First even though the experimental setup involved three injection holes, which are 3 diameters apart, here, as shown in Figure 2, only the domain between a plane through the centre of a middle hole and one halfway between the middle and a adjacent hole is considered with symmetry boundary conditions on either side. This effectively assumes that there are enough injection holes on either side of the middle one, for the flow to repeat itself in the spanwise (z) direction.

Figure 2 - Spanwise extent of solution domain. A further reduction in the size of the solution domain and hence in grid requirements, is accomplished through the selection of the location of the inlet boundary. As already noted, the experimental setup involved a long upstream section, necessary for the boundary layer thickness to grow to the required size. As shown in Figure 3, instead of including this long upstream domain in the film-cooling computations, a preliminary computation of boundary-layer flow is carried out which generates a boundary layer of appropriate thickness, which is subsequently used to provide inlet conditions to the film cooling domain, which now starts two diameters upstream of the injection hole. This leads to a further 34% saving in grid requirements. As can be seen from the comparison of Figure 4, this approach provides the correct velocity distribution at the start of the flow calculation domain.

Figure 3 Generation method for flow conditions upstream of injection hole.

Figure 4 Comparison between the computed boundary layer profile at the centre of the injection hole and that measured by Jessen et al. (2007)

Finally the mesh size was further reduced by halving the height of the flow domain (to 60D) and inserting a symmetry condition at the new upper surface instead of wall boundary conditions at 120D, as shown in Figure 5. It was not clear whether this step would be detrimental to the results, as it effectively introduces a second coolant injection jet opposite the main coolant injection jet in the y-direction. Comparative simulations were run to test what effect this modification had on the results and no noticeable difference was found. A half-height flow domain was therefore accepted for use in the main simulations At the inlet to the injection hole, a uniform velocity profile was prescribed, while the outlet was modelled using the outflow boundary condition. In this condition, FLUENT extrapolates pressure and velocity data from the interior. This was a suitable boundary condition since the details of the pressure distribution at the outlet were unknown prior to the solution of the problem. After the above modifications had been carried out, it was felt that the mesh was reasonably efficient and suitable for use. The final preliminary work that was carried out was to determine the number of cells at which the solution became grid-independent. Simulations were run using meshes with 700000, 1400000, and 2100000

Figure 5 - Alternative locations of flow domain boundary in Figure 6 Unstructured grid around junction wall normal direction. between coolant tube and flat plate. cells. A considerable difference was noticed between the results of the meshes with 700000 and 1400000 cells, while the meshes with 1400000 and 2100000 cells showed similar results. The solution was, therefore, deemed to be practically grid-independent at 1400000 cells. A close-up of the junction where the coolant injection tube meets the flat-plate, for the final mesh, is shown in Figure 6. RESULTS Starting with Case 1, for which the velocity and density ratios are 0.28 and 1 respectively, Figure 7, compares the predicted and measured mean flow fields at the vicinity of the injection hole. This qualitative comparison shows that the current computations are able to reproduce the main flow features, namely the upward deflection of the mainstream fluid the formation of a separation bubble at the downstream corner of the injection hole and also presence of slow moving fluid, over the plate, downstream of the injection.

Figure 7 - Streamwise velocity contour plots within the symmetry plane (z/D=0). VR=0.28 and DR=1.

Figure 8 shows a more quantitative comparison of streamwise velocity profiles at various locations along the symmetry plane (z/D=0), also for Case 1 (VR=0.28 and DR=1). Velocity profiles have been considered at x/D = -1, 0, 1, 1.5 and 2. At x/D=-1, the upstream edge of the injection hole, injection has little effect on the freestream flow and the close agreement between computation and measurements is not surprising. At x/D=0, over the centre of the injection hole, the effects of injection are vey striking. The wall velocity is no longer zero and the velocity gradient in the wall normal direction becomes zero. At x/D=1, half a diameter after the hole, the coolant penetrates the freestream flow and a boundary layer is re-established very close to the surface. At x/D=1.5 there is evidence of flow separation very close to the wall, which by the z/D=2 location disappears. The computations are in close agreement with the measurements, reproducing most flow features present. One notable exception is at x/D=1, where the computations return a slower re-establishment of the boundary layer, than the measurements.

Figure 8 - Streamwise velocity profiles over the symmetry plane (z/D=0). VR=0.28 and DR=1. : Computations, : Measurements Jessen et al. (2007) Turning the attention now to the effects of density ratio, case 2, Figure 9 shows the streamwise velocity profiles within the symmetry plane (z/D=0), for a velocity ratio, VR, of 0.28 and a density ratio, DR, of 1.53. The measurements suggest that the flow development within the symmetry plane is broadly similar to that of Figure 8, for a density ratio of 1, perhaps with the only detectable difference a thinner separation bubble, downstream of the injection hole, at z/D=1.5. The computations also return broadly the same flow field as for case 1, with a similar deviation from the measurements, just after the injection hole, at x/D=1. The computations now also fail to predict the presence of a separation bubble at x/D=1.5, which for this higher density ratio has reduced in size. Overall agreement between the predicted and measured streamwise flow around the injection hole is again reasonable. For this case, VR=0.28 and DR=1.53, profiles of the velocity component normal to the wall, within the symmetry plane (z/D=0), are presented in Figure 10. Just upstream of the injection hole, at x/D=-1, the flow is largely unaffected by the coolant injection, and consequently the wall-normal velocity is low. The wall-normal velocity component increases over the hole, x/D=0, and its effects extend to about two diameters above. The wall-normal velocity reaches its maximum value just after the hole, x/D=1, probably because the separation bubble further downstream deflects the injected fluid upwards. At x/D=2, by which location the flow has reattached, the wall-normal velocity has decreased again. The FLUENT realizable k- predictions of the wallnormal velocity are again impressively close to the data. While over and after the injection hole, the near-wall levels of this velocity component are over-estimated a little, the level, the variation of the wall-normal velocity with wall distance and also its recovery after the injection slot are faithfully reproduced.

Figure 9 - Streamwise velocity profiles over the symmetry plane (z/D=0). VR=0.28 and DR=1.53 : Computations, : Measurements Jessen et al. (2007)

Figure 10 Wall-normal velocity profiles over the symmetry plane (z/D=0). VR=0.28 and DR=1.53 : Computations, : Measurements Jessen et al. (2007) The final quantitative flow comparison, also for case 2, focuses on profiles of the streamwise turbulence intensity in Figure 11. The experimental data show that just upstream of the injection hole, x/D=-1, possibly because the approaching boundary layer flow is subjected to an adverse pressure gradient, there is a strong increase in the near-wall intensity levels. As the flow moves over the injection slot, x/D of 1, 1.5 and 2, this peak in turbulence intensity which before the slot is near the wall, is now displaced further into the flow, by the injected fluid. The minimum in turbulence intensity near the wall just after the slot, at x/D=1, suggests that the injected fluid has low intensity. At the next location, x/D=1.5 a second peak appears in the measured intensity profile, closer to the wall. This second peak is probably caused by the flow separation. These two peaks in turbulence intensity merge further downstream, at x/D=3 and at a distance of about half D from the wall. Beyond this streamwise location, the intensity peak starts to diminish and a near-wall peak, associated with fully developed boundary layer flow starts to develop. A comparison with the mean velocity profiles of Figure 9, shows that the turbulence field evolves more slowly than the mean flow field downstream of the injection hole. The intensity profiles predicted with the FLUENT realizable k- and enhanced wall treatment, while not in compete agreement with the measurements, manage to capture most of the important features. Failure to predict the high near-wall levels at x/D=1, is partly due to the insensitivity of the one-equation model, embedded in the

Figure 11 Profiles of the streamwise turbulence intensity over the symmetry plane (z/D=0). VR=0.28 and DR=1.53 : Computations, : Measurements Jessen et al. (2007) enhanced wall treatment, to the effects of adverse pressure gradient and partly due to the use of the effective viscosity approximation, which does not reproduce the wall effects on the anisotropy of turbulence. The latter is also responsible for the failure of the computations to predict the near-wall peaks that develop at the recovery region (x/D=3 and 6). The failure to predict the strong reduction in turbulence over the injection hole, x/D=1, suggests that in the computations, the inlet intensity levels in the coolant were too high. Nevertheless, overall, given the limitations of the turbulence model employed, agreement with the measurements is surprisingly good.

Figure 12 - Mean mixture fraction of CO2 in planes normal to the main flow (y-z) VR=0.28 and DR=1.53. The rate at which the coolant spreads, and hence provides a thermal barrier for the surface, can be studied following how the injected CO2 mixes with the mainstream flow, through the contours of CO2 concentration of Figure 12. Close to the injection hole, x/D=0.75, while already starting to spread, most of the coolant core, in contact with the plate, retains its original high concentration and the jet has an elliptical shape. Further downstream (x/D=1.5), the jet continues to expand and the coolant concentration at its core starts reduce. A more critical, from the thermal point of view, development, is that the jet starts to lift away from the surface. This lifting is due to freestream fluid becoming entrained between the wall and coolant jet and is largely driven by the counter-rotating vortex pair observed in earlier studies. By x/D=6, the density in the jet has reduced significantly. The jet has spread laterally, as well as further into the free stream.

Figure 13 - Contour plot of the mean mixture fraction of CO2 along the wall surface It was suggested by Renze et al. (2008) that the distribution of the mean mixture fraction f of CO2 along the wall, downstream of the coolant injection hole, should resemble the film cooling efficiency. Figure 13 is a contour plot of the mean mixture fraction of CO2 along the wall. The dilution of the dense coolant jet with freestream fluid can be seen downstream of the injection hole. Moreover, immediately downstream of the injection hole, x/D between 1 and 2, a narrowing of the jet footprint on the plate surface is evident. This is consistent with the lifting of the jet observed in the concentration contours of Figure 12 (at x/D=1.5) attributed to the presence of counter-rotating vortices. As the effects of these vortices start to die down further downstream, the jet footprint starts to spread again. Sinha et al. (1991) carried out experiments to measure the film cooling efficiency for a jet in cross-flow, with a velocity ratio of 0.25 and a density ratio of 2. The measured film cooling efficiency from these experiments is compared to the simulated mean mixture fraction of CO2, along the symmetry line (with VR=0.28 and DR=1.53) in Figure 14. Despite the differences in velocity and density ratios, the computed data agrees well with the measurements. The mean mixture fraction of CO2 along the wall therefore gives a reliable indication of what the film cooling efficiency is going to be. A consequence of this is that the contour plot of the mean mixture fraction of CO2 along the wall, presented in Figure 13, provides a reliable two-dimensional mapping of the film cooling efficiency CONCLUDING REMARKS Figure 14 - Mean mixture fraction of CO2 compared to The current work has looked at the flow physics the film cooling efficiency along the symmetry plane of jets-in-a-cross-flow using CFD and a RANS (z/D=0). : Computed mixture fraction model. FLUENTs realizable k- model was used, : Cooling efficiency, Sinha et al. (1991) together with the enhanced wall treatment, which is effectively a two-layer zonal method. Different density ratios were looked at density ratios higher 1 being obtained by using CO2 as the coolant fluid and a non-reacting multi-species flow in the simulation. One important finding, highly relevant to these three-dimensional flow computations, is that careful selection of the location of boundaries, which sometimes may require preliminary computations and comparisons, can reduce grid requirements by an order of magnitude. The computational results were in reasonable agreement with the experimental results of Jessen et al. (2007). The mean flow development is well reproduced at both density ratios with discrepancies between predictions and measurements confined to he near-wall region immediately (the first diameter) downstream of the injection hole. The computations also managed to reproduce most of the features present in the turbulence intensity measurements. The discrepancies identified can be attributed to the use of what is effectively a zonal near-wall

model of turbulence and also to the effective viscosity approximation. Overall, this FLUENT model produced reliable film-cooling predictions of the test cases considered. For simulations using non-reacting multi-species flow, the mean mixture fraction of coolant throughout the flow was also analysed. The coolant jet was found to be lifted away from the wall surface downstream of the injection hole, and the coolant density was reduced due to mixing with the free stream. It was found that the footprint of the mean mixture fraction of CO2 along the wall surface was a good indicator of the film cooling efficiency. NOMENCLATURE BR Ci ci D DR f H IR P Pr Re Sc s T TC TW T t U Ui U ui V VC VR x y z ij Blowing ratio [=(C VC)/( U)] Mean mass concentration of species i. Fluctuating mass concentration of species i. Cooling hole diameter Density ratio [=C/] Mixture fraction of CO2 Eind tunnel height 2 2 Momentum ratio [=(C VC )/( U )] Mean Pressure Fluid Prandtl number Local Reynolds number based on the distance from the leading edge Schmidt number Injection hole spacing Mean temperature Coolant temperature Wall temperature Free stream temperature Fluctuating temperature Mean streamwise velocity Mean velocity vector Free stream velocity Fluctuating velocity vector Mean wall-normal velocity Coolant velocity Velocity ratio [=VC/U] Streamwise direction Wall-normal direction Spanwise direction Boundary layer thickness Kronecker dhelta Adiabatic temperature [(T-TW)/( T-TC] Fluid viscosity Fluid density

Coolant density Density of free stream fluid.


Turbulent Prandtl/Schmidt number

REFERENCES S. Baldauf, A. Schulz and S. Wittig (2001), High-resolution measurements of local effectiveness from discrete hole film cooling, ASME J. Turbomachinery, 123, 758-765 V.K. Garg, and R.S. Abhari (1997), Comparison of predicted and experimental Nusselt number for a filmcooled rotating blade, Int. J. Heat Fluid Flow, 18, 452-460. J.-C. Han, S. Dutta, and S. Ekkad (2000), Gas turbine heat transfer and cooling technology, Taylor & Francis, London H. Iacovides and B.E. Launder (1984), The computation of momentum and heat transport in turbulent flow around pipe bends, 1st UK National Heat Transfer Conference, Leeds.

H. Iacovides and B.E. Launder (1987), Turbulent momentum and heat transport in square-sectioned ducts rotating in orthogonal mode, Numerical Heat Transfer, 12, p 475, 1987. M.Y. Jabbari, and R.J. Goldstein (1978), Adiabatic wall temperature and heat transfer downstream of injection, through two rows of holes, ASME J. Eng. Power, 100, 303-307. W. Jessen, W. Schroder, and M. Klaas (2007), Evolution of jets effusing from inclined holes into cross-flow, Int. J. Heat Fluid Flow, 28, 1312-1326 H. Kruse (1985), Effects of Hole Geometry, Wall Curvature and Pressure Gradient on Film Cooling Downstream of a Single Row. Heat Transfer and Cooling in gas Turbines, AGARD CP-390, Paper 8. A.B. Mehendale, J.-C. Han, S. Ou, and C.P. Lee (1994), Unsteady wake over a linear turbine blade cascade with air and CO2 film injection: Part II Effect on film effectiveness and heat transfer distributions. ASME J. Turbomachinery, 116, 730-737. P. Renze, W. Schroder, and M. Meike (2008), Large-eddy simulation of film cooling flows at density gradients. Int. J. Heat Fluid Flow, 29, 18-34. T.-H. Shih, W. W. Liou, A. Shabbir, Z. Yang, and J. Zhu. (1995), A New Eddy-Viscosity Model for High Reynolds Number Turbulent Flows-Model Development and Validation., Computers Fluids, 24(3):227-238 A.K. Sinha, D.G. Bogard, and M.E. Crawford (1991), Film-cooling effectiveness downstream of a single row of holes with variable density ratio, ASME J. Turbomachinery, 113, 442-449. M. Wolfstein (1969), The Velocity and Temperature Distribution of One-Dimensional Flow with Turbulence Augmentation and Pressure Gradient, Int. J. Heat Mass Transfer, 12, 301-318.

You might also like