This action might not be possible to undo. Are you sure you want to continue?
DEVELOPMENT AND VALIDATION OF A C++ OBJECT ORIENTED CFD CODE FOR HEAT TRANSFER ANALYSIS
L. Mangani, C. Bianchini, A. Andreini, B. Facchini Department of Energy Engineering ”Sergio Stecco” University of Florence - Italy Via Santa Marta, 3 - 50139 Florence - Italy firstname.lastname@example.orgﬁ.it
INTRODUCTION One of the most demanding problem in gas turbine design is the proper evaluation of heat transfer phenomena which involve all hot components of the engine. Furthermore, all topical design criteria make heat transfer problems more and more difﬁcult. An improvement in gas turbine performance, for example, can be produced by increasing turbine inlet temperature, which is usually well above the metal critical temperature. In addition, new design concepts adopted for combustors, based on lean premixed ﬂames, reduce the amount of air available for wall cooling. These are only two typical examples that justify the increasing interest in developing more and more advanced cooling systems. The complexity of geometries usually adopted in such designs and the high costs required for accurate heat transfer measurements justify the increasing use of CFD analysis in each phase of the design process. Nevertheless, CFD simulations for evaluation of thermal loads and effectiveness of the cooling devices in gas turbine engines are demanding both in terms of physical modeling and geometrical mesh handling. Actual cooling geometries are characterized, for example, by intricate shapes, with non aerodynamic turbulators such as pins and ribs that must be properly discretized, or they involve complex ﬂows such as impinging jets that make turbulence modelling a key point. Such issues usually require CFD codes to satisfy some essential features: a quite large set of turbulence models, in order to have accurate predictions with all possible ﬂows and the capability in handling hybrid unstructured meshes. The consequence of such strict requirements is that a very reduced set of CFD codes is available worldwide, and the choice is usually limited to few 1 Copyright c 2007 - by ASME
This paper describes the development and validation steps of computational sub-models for gas turbine heat transfer applications, within an open source CFD code based on the Field Operation and Manipulation C++ class library for continuum mechanics (OpenFOAM, http://www.opencfd.co.uk). Open FOAM is based on a polyhedral ﬁnite-volume approach with a co-located variables arrangement. In order to set up OpenFOAM toolbox to analyze heat transfer problems with RANS approach, it was necessary to add and implement some additional submodels. First of all a SIMPLE like algorithm was speciﬁcally developed to solve the fully three dimensional, steady state form of compressible Navier Stokes equations. Moreover several eddy viscosity models such as the standard, the Two Layer version and the realizable k − ε model and the k − ω SST model have been implemented. The accuracy of the implementations was validated comparing results with experimental data available both from standard literature test cases and from in house performed experiments. The geometries considered as validation tests cover the typical heat transfer problems in gas turbine design . On the whole, during the tests, OpenFOAM code has shown a good accuracy and robustness. The purpose of this work is to show the ability of an innovative CFD tool as support for gas turbine designers and to verify its role as an effective substitute for standard commercial CFD packages.
well known commercial codes. Commercial software have dominated, in the last decade, CFD analysis of heat transfer problems for turbomachinery applications both in industrial and academic ﬁeld. Besides their numerous advantages, such as the simplicity of use via practical graphical interfaces, they present some common drawbacks: for example the waste of system resources with a large part of packages not used in standard simulations, which is one of the source of their poor performances in terms of calculation times. However, according to experts, we think that the main drawback of commercial CFD codes is their nature of “black box solution maker”. Advanced users in heat transfer applications need to understand the physics and sometimes the use of “ad hoc” models or modiﬁcations suitable for speciﬁc cases. User subroutine features provided by commercial packages become quickly inadequate as the complexity of modiﬁcations grows. Furthermore R&D department of big companies usually need to tune built-in models in order to feed calculations tools with their design practice frequently based on detailed and expensive experimental tests. The objective of the work presented in this paper is to show the capabilities of a new open-source software environment where it is possible to implement new models, renew the existing ones and experiment with model combinations. The OpenFOAM package (Field Operation And Manipulation) [1, 2] is an object-oriented numerical simulation toolkit written in C++ language [3, 4, 5]. Besides its advanced basic native CFD features, which will be described in the next parts, its essential characteristic is the opportunity to build new models and solvers with high simplicity and in less time than with standard Fortran based codes. Object-oriented programming of C++ drastically reduces the probability of bugs introduction with a consequent saving in debugging time. This paper describes the attempt to build a CFD package suitable for typical steady state heat transfer analysis which could be able to assist gas turbine design process. As will be described later, to reach such goal it was necessary to introduce speciﬁc modules to the standard release in order to overcome the limitations of built-in approaches: ﬁrst of all a compressible steady state solver capable at handling transonic ﬂows, then a set of turbulence two-equations closures with particular reference to a detailed near wall treatment. Additional features such as temperature dependent thermo-physical properties and generic grid interfacing have also been developed. It’s important t remark that such features are not available in the released version of the toolkit, as built models are mainly focused on unsteady weakly compressible ﬂows. As a consequence it’s not possible to draw out speciﬁc comparisons between default and developed models. As conﬁrmation of the work done a set of validation testcases were performed. In particular, in this paper, we will focus our attention on the validation of the code with some complex conﬁgurations typical of heat transfer problems such as ﬁlm cooling and impingement cooling. Both ﬁlm and impingement 2
cases were analyzed with single and multi-hole conﬁgurations. In particular the well known Sinha experiment was considered for single hole ﬁlm cooling test , while for multi-hole case an experiment performed in our Department was chosen . Film cooling geometries here considered belong to full-coverage ﬁlm cooling case also known as effusion cooling, a promising technique used in combustor wall and turbine end-wall cooling. For single hole impingement tests, we referred to the classical ERCOFTAC C.25 test  while for multi-hole case again an experiment performed in our Department was considered . Comparison with experimental data are reported in terms of adiabatic effectiveness for ﬁlm cooling tests and wall heat transfer coefﬁcient for impingement runs. Furthermore, in order to verify the accurate implementation of selected turbulence models, a simple ﬂat plate tests was considered, while to show the robustness of the steady state solver developed results of classical validation tests are brieﬂy commented.
NOMENCLATURE U Vector velocity D Hole diameter h Heat transfer coefﬁcient k Turbulent kinetic energy L Height of the jet p Pressure p Pressure corrector q ˙ Heat ﬂux Re Reynolds number T Temperature Ts Turbulent time scale Pk Turbulence production term “ = µt S X Y Cρ H Greeks α η
[m s−1 ] [m] [W m−2 K −1 ] [m2 s−2 ] [m] [N m−2 ] [N m−2 ] [W m−2 ] [K] [s] [kg m−1 s−3 ] [s−1 ] [m] [m] [kg J −1 ] [kg −1 s m3 ]
” ∂U ∂U ∂U − 2 δij ∂x k ∂x i − 2 ρk ∂x j j k “ 3 ”j 3 ∂Uj ∂Ui Tensor strain = 0.5 ∂x + ∂x +
∂Uj ∂xi j i
Streamwise direction Spanwise direction 1 Compressibility term RT Diffusive discretization term
Angle between hole and crossﬂow (T∞ −Taw ) Adiabatic effectiveness (T −T ) ∞ c P <η> Spanwise averaged effectiveness n ω Turbulence frequency ε Turbulence dissipation µt,ef f Eddy viscosity ρ Density Subscripts 0 Uncooled plate aw Adiabatic wall ∞ Crossﬂow c Coolant w Wall
(T∞ −Taw ) (T∞ −Tc )
[s−1 ] [m2 s−3 ] [kg m−1 s−1 ] [kg m−3 ]
OpenFOAM The OpenFOAM (Field Operation And Manipulation) code [10, 11, 12] is an object-oriented numerical simulation toolkit written in C++ language. The toolkit implements operator-based implicit and explicit second and fourth-order Finite Volume (FV) discretization in three dimensional space. Efﬁciency of execution is achieved by the use of preconditioned Conjugate Gradient  Copyright c 2007 - by ASME
and Algebraic Multigrid solvers and the use of massively parallel computers in the domain decomposition mode. Being primarily a C++ library ready to create executables, OpenFOAM uses object based programming language. It means programmers can use OpenFOAM native classes both to deﬁne their own classes or to build new applications, such as solvers or utilities, with ease of development. Object oriented programming allows data abstraction, object orientation, operator overloading and generic programming. It means that it enables the construction of new types of data speciﬁc for the problem to be solved, the bundling of data and operations into hierarchical classes preventing accidental corruptions, a natural syntax for user deﬁned classes and it easily permits the code re-use for equivalent operations on different types. OpenFOAM native grid engine can handle meshes of arbitrary polyhedra bounded by arbitrary polygons, giving a large ﬂexibility in mesh generation, see Fig. 1. Switching to OpenFOAM way of thinking, programmers must approach a ﬁeld based philosophy more than a cell or face based one, each physical quantity (no matter what dimension, rank or size) is represented by a single object and treated as a ﬁeld.
ter of combining in a different way the same set of basic differential operator. Just to give an example of the capability of such a top-level code, let’s consider a standard equation like momentum conservation:
∂ρU + ∂t
· (ρU U ) −
(µ U ) = − p .
It can be implemented in an astonishingly almost natural language which is ready to compile source C++ code: solve ( fvm::ddt(rho, U) + fvm::div(phi, U) - fvm::laplacian(mu, U) == - fvc::grad(p) ); letting programmers concentrate their efforts more on the physics than on programming. Another important feature allowed by object programming is the dimensional check, physical quantities objects are in fact constructed with a reference to their dimensions and so only valid dimensional operations can be performed avoiding errors and permitting once again an easier understanding. Even if OpenFOAM can be used as a standard simulation package, its tools are in general too rough to well predict cases of industrial interests. Its strength in fact is not really to be a ready-to-use code but stands in being open not only in terms of source code but, what’s more, in its inner structure and hierarchical design, giving the user the opportunity to fully extend its capability. The subject of this work is the preparation of a set of models capable of transforming OpenFOAM in a complete calculation suite for heat transfer turbomachinery simulations. In these initial steps we have focused our efforts in standard steady state RANS approach considering that it still represents the most common for CFD design process. Nevertheless, OpenFOAM code is fully able to handle unsteady calculations and it is already equipped with a LES module: the relevant computational and post-processing costs followed by an inevitable simpliﬁcation of near wall treatment usually prevent the use of LES approach on actual geometries. To extend LES to cover industrial ﬂows at high Reynolds numbers, new approaches (hybrid LES-RANS, DES, URANS) must be used: they are all based on a mix of LES and RANS and they require further development which will be the matter of future work. In the next part we are going to discuss the development of the solver algorithm considered for our application and, nonethe3 Copyright c 2007 - by ASME
Figure 1. Polyhedral mesh example.
Differential operators can be treated like ﬁnite volume calculus (fvc) or ﬁnite volume method (fvm) operators. The ﬁrst approach performs explicit derivatives returning a ﬁeld, the second one is an implicit derivation converting the expression into matrix coefﬁcients. The idea standing behind is to think about partial differential equations in terms of a sum of single differential operators that can be discretized separately with different discretization schemes. Differential operators such as gradient, divergence, laplacian and curl have been overloaded for the different types of ﬁeld, giving each of them the most suitable meaning. Implementing different types of equations is now only a mat-
less, the implementation of speciﬁc turbulence models for LowReynolds and High-Reynolds simulations.
smaller than a prescribed small number.
SOLVER In turbomachinery and heat transfer applications, involved ﬂuid ﬂows may usually cover a wide range of Mach regimes. In particular, it usually happens that different Mach conditions simultaneously arise in the same domain. Such situation makes the accurate solution of viscous ﬂows governing equations a complex task. Most widely used algorithms for compressible ﬂows calculation use density as one of the main independent variables and pressure is determined via an equation of state. As there is very little or no change in density for low subsonic or nearly incompressible ﬂows, these density-based methods fail in such regimes. Their application in cases of incompressible or low Mach number ﬂows is questionable, since in that situation the density changes are so small that the pressure-density coupling becomes very weak. To avoid this weakness another class of methods, proposed originally for viscous incompressible ﬂows [14, 15, 16, 17] and later extended to compressible ﬂows [18, 19, 20, 21, 22, 23, 24] use pressure as the main independent variable also with the concept of the ‘retarded density’ [25, 26, 27]. Such pressure-based approach is founded on the SIMPLE algorithm (Semi-IMplicit Pressure Linked Equations) . In this method, continuity equation is converted into an equation for pressure corrector overturning the linkage between pressure and density to extend applicability range up to zero Mach number. The SIMPLE algorithm uses a segregated approach where the equations are solved in sequential steps letting to the iterative process the care of the non-linearity as well as the coupling between equations. To better visualize the cycle of SIMPLE algorithm a ﬂow-chart of the pseudo code is reported in Fig. 2. For each transport equation, a system of linear algebraic equations is obtained. These are solved cyclically applying the preconditioned conjugate gradient. Keeping the coefﬁcients in the algebraic equations ﬁxed, generally one to ten iterations are performed in the inner cycle. Typically for velocity, temperature and pressure correction equation, iterations are stopped when the sum of the absolute residuals over the whole solution domain has fallen about three orders of magnitude, or the prescribed maximum number of inner iterations has been reached. The equation of state is used to update density after new solutions for temperature and pressure are obtained. After one cycle of inner iterations has been performed for each variable, the coefﬁcients of the algebraic equations are updated using the newest values of all variables, outer iterations. In this way the non-linearity and coupling of equations is accounted for. Outer iterations are stopped when the sum of the absolute residuals for each variable decreases of prescribed orders of magnitude or when the normalized sums are 4
Figure 2. Flow chart of SIMPLE algorithm.
In this work we considered a pressure-based ﬁnite volume solver using a co-located variable approach suitable for calculating steady-state ﬂows at all speeds. The development of this class of methods in contrast with the standard SIMPLE technique lies in a more precise derivation of the equation for the pressure corrector, allowing the possibility of treating at the same time low subsonic, almost incompressible, and high compressible ﬂows. Such methods have been validated in many heat transfer problem conﬁgurations. The pressure correction equation, details can be found in [28, 29, 30, 31], in the compressible form says: · (Cρ U p ) − · (ρH( p )) = − · (ρU ). (2)
The role of Eq.(2) in the SIMPLE cycle is to enforce mass conCopyright c 2007 - by ASME
servation, it is in fact derived from a combination of momentum conservation and continuity equation. In order to solve Eq.(2), attention should be posed on the fact that the pressure correction equation now assumes a convective-diffusive form instead of a purely diffusive behavior like the original incompressible formulation. While the other steady-state form transport equations have to be relaxed in order to characterize the inertial physics lost by the elimination of the time derivative, for the pressure correction equation this cannot be done. Usage of usual implicit relaxation techniques on pressure corrector, in fact, corrupt mass conservation on single iteration steps breaking the concept standing behind SIMPLE algorithm. In subsonic cases, standard Neumann conditions at inlet velocity boundary, like in incompressible tests, determine ill-deﬁned problems for Eq.(2). Care must be taken in handling pressure correction boundary condition in order to solve in a well-posed manner such an equation . A combination of Dirichlet and Neumann type condition for the inlet has been tested.
, Lien , Abe et al , Chien , Chen et al , Hwang and Lin  and Lam and Bremhorst  have been implemented. It is known from literature that in high strain rate regions eddy viscosity models overpredict turbulent kinetic energy: this problem is sometimes referred to as “stagnation point anomaly” . These higher values of k are due to an overestimate of production term Pk . To avoid such overprediction linear dependence between Pk and |S|2 should be bounded in regions where |S| grows. This is achieved with a time scale bound, derived from a “realizability” constraint for Reynolds stress tensor to be deﬁnite positive: µt = min Cµ ρk k α ,√ ε 6Cµ |S|
TURBULENCE MODELS The correct modeling of turbulent quantities is fundamental in conducting heat transfer simulations, because of the simultaneous importance of well predicting both the near wall behavior and the complex structures of the main ﬂow [33, 34]. Correct predictions of thermal quantities and gradients inside boundary layers are necessary to establish whether or not the cooling system is efﬁcient. At the same time wall properties are very dependent on the development of the free stream ﬂow. Usage of wall function approach has to be avoided because of the unpredictability of boundary thermal gradient and the failure in predicting transitional, Low Reynolds as well as adverse pressure gradient ﬂows. First step in modeling ﬂows close to solid walls has been the implementation of several so called Low Reynolds k − ε models . The idea standing behind such models is to damp turbulent viscosity near the wall through a damping function fµ going towards zero as the distance from the wall is reducing. Constants multiplying source terms in the turbulent dissipation equation are in some cases also damped. The basic structure of the models is the same for all of them differing in the tuning of the damping functions and some extra sources in dissipation equation, as shown below: ∂ρk + ∂t ∂ρε + ∂t · (ρU k) − · (ρU ε) − · (µef f k) = Pk − ρε , · (µef f ε) = f1 Pk (3)
This limiter proposed by Durbin  has been inserted in all Low Reynolds models above presented as an option to be switched on or off by the user. Then, in order to match good near wall predictions with suitable modeling of ﬂow structures far from the wall, Two Layer k − ε models have been implemented. Such methods consist in patching together a one equation model in the near wall layer and a two equation High Reynolds model in the outer layer . Both Wolfestein and Norris&Reynolds closure formulas  have been tried without signiﬁcant discrepancies in the results. Last model to be mentioned is the k − ω SST: it includes the modiﬁcation of the standard k − ω to avoid sensitivity to quite arbitrary freestream values of ω [45, 46]. The basic idea is similar to Two Layer models: two different approaches are merged together to model the two different ﬂow regions. The sublayer and logaritmic model is the standard k − ω, chosen because of its robustness, the absence of damping function and Dirichlet type boundary conditions. From the wake region and outside the boundary layer the standard k −ε, written in terms of ω, has been preferred due to its good compromise in predicting different kind of ﬂows.
RESULTS Generalities All the cases to be presented, apart from the ﬂat plate one, have been chosen because already tested and analyzed with commercial solvers by the authors, with some results already published, see for example [9, 47]. Grid sensitivity analysis have been performed when those runs were set up and it’s not repeated in this case, the various meshes however guarantee a ﬁrst node 5 Copyright c 2007 - by ASME
ε ε2 − f2 ρ . (4) k k
Of the many Low Reynolds k − ε models proposed in literature in the course of years, the models by Lien and Leschziner
y + ≤ 1. All ﬂuid domains are discretized via hexahedral elements except Goldman test, total number of elements for each test is reported in Tab. 1.
Table 1. Grid sizes (thousands of elements)
GAMM test 1 GAMM test 2 Goldman test Flat plate ERCOFTAC C25 Axial-symmetric Impingement Jet 1-Hole Impingement Jet 5-Holes Impingement Jet Sinha test 6-Holes Effusion 10.0 13.5 7.2 17.6 69 387.0 1705.0 177.0 2000.0
If this condition was satisﬁed for all scalar but pressure corrector calculations were stopped. Due to its nature, pressure correction residual is of no interest and, moreover, its initial value is set to zero at every iteration. To verify whether or not pressure ﬁeld is still varying, the maximum module of the pressure corrector, at convergence exactly null, is imposed to be less than 10 [P a], remember an averaged relaxation factor for the pressure corrector is of the order of 10−2 .
Solver Validation Tests GAMM tests The developed calculation procedure has been used to solve a variety of problems in heat transfer applications. Here the emphasis is on the high compressible ﬂows. The capability of the present method is demonstrated by computing inviscid ﬂow in a channel with a bump on the lower wall named GAMM test. This test case has been used by various researchers to test their algorithms [18, 48]. Application of the method to two different types of inviscid ﬂow, transonic and supersonic, are presented below. The width of the channel is equal to the length of the bump, and the channel length is equal to three lengths of the jump. For transonic calculation, the thickness-to-chord ratio is 10% while for supersonic ﬂow calculations it is 4%. In transonic and supersonic regime at inlet is assumed that ﬂow has uniform properties and the upstream far ﬁeld variable values (except pressure in transonic case) are speciﬁed while at the outlet all variable (except pressure in transonic case) are extrapolated. At the upper and the lower boundaries wall slip condition is prescribed. First case with imposed inlet Mach number M ain = 0.675, gives the Mach number distributions along the walls and density gradient magnitude contour plot shown in Fig. 3(a) and Fig. 4(a). In the supersonic case, M ain = 1.65, the ﬂow results supersonic all along the bump: Mach number distributions and density gradient magnitude contour plots are shown in Fig. 3(b) and Fig. 4(b). These results correspond to reference solutions from literature [18, 48]. Fig. 3(c) and Fig. 4(c) show the Mach number distribution and density gradient magnitude contour plot under the same condition of supersonic case but with two bumps. As can be seen by comparing Fig. 3(b), Fig. 3(c) and Fig. 4 the second bump does not inﬂuence the ﬂow upstream indicating that the solution algorithm correctly reproduces the hyperbolic behavior of the ﬂow. Goldman test As example of highly compressible subsonic, we have reported the simulation of a test based on the work of Goldman et al.. It is a 2-D turbulent analysis of a stator blade at the mid-span; the details of the geometry and the mesh are shown in Fig. 5(a). The Reynolds number, based on the chord length of the 6 Copyright c 2007 - by ASME
Due to the great number of implemented turbulence models, a shortcut has been used to name most of them: the acronyms presented in Tab. 2 will be widely used in substitution of authors’ full name.
Table 2. Acronyms for the various turbulence models.
k − ε Low Reynolds by Abe et al. k − ε Low Reynolds by Yoder and Georgiadis k − ε Low Reynolds by Lien et al. k − ε Low Reynolds by Hwang and Lin k − ε Low Reynolds by Lam and Bremhorst k − ε Low Reynolds by Lien and Leschziner k − ε Low Reynolds by Lien Realizability constraint correction Two Layer k − ω SST AKN CH CLL HW LB LW LNR Real TL SST
The convective spatial discretization used is based on the Normal Variable Approach (NVA), and named in literature as Self Filtered Central Differencing (SFCD) scheme . To check convergence the arrest criterium has been deﬁned as single scalar normalized residual lower than 10−6 . Normalization factor, N orm, was not changed from the released version and is deﬁned as:
Φref = Φ , S = A·Φ, Sref = A · Φref , Norm = (|S − Sref | + |Q − Sref |) .
2.25 2 1.75 1.5 Mach Number 1.25 1 0.75 0.5 0.25 0 -1.5 -1 -0.5 0 x 0.5 1 1.5 Lower-Wall Upper-Wall
(a) Transonic ﬂow.
(a) Mach proﬁle in transonic ﬂow.
2.5 2.25 2 1.75 1.5 1.25 1 0.75 0.5 -1.5 -1 -0.5 0 x 0.5 1 1.5 Lower-Wall Upper-Wall
(b) Supersonic ﬂow.
(b) Mach proﬁle in supersonic ﬂow.
2.5 2.25 2 1.75 1.5 1.25 1 0.75 0.5 0 0.5 1 1.5 2 x 2.5 3 3.5 4 4.5 Lower-Wall Upper-Wall
(c) Supersonic ﬂow with two-bump geometry. Figure 4. Density gradient contour plots.
(c) Mach proﬁle in supersonic ﬂow with twobump geometry. Figure 3. Proﬁle Mach number in upper and lower wall.
blade and the free-stream velocity, is 500000 and the inlet Mach number is approximately 0.2. A comparison of the predictions for blade loading (deﬁned as the ratio of static pressure to the inlet total pressure) with the experimental data is shown in Fig. 5(b).
that reported by Wieghardt  and later included in the 1968 AFOSR-IFP Stanford Conference . Details about ﬂow conditions are listed in Tab. 3.
Table 3. Flow conditions for ﬂat plate test
Inlet temperature Inlet Mach number Pressure Turbulence kinetic energy - k 294.4 0.2 101400 23.6 3365 4.6870 Pa m2 s−2 m2 s−3 m K
Turbulence Models Validation Test Flat Plate A simple test of a ﬂow over a ﬂat plate has been considered the best choice to start the validation and selection process of the turbulence models. Because of the large amount of accessible data, the ease and velocity of the test, corrections and tuning could be carried out quickly and accurately. It has been properly checked the near wall behavior of both turbulent kinetic energy and turbulent kinetic energy dissipation. Due to the relative simplicity of the case, runs have been performed with a very ﬁne grid: ﬁrst node y + ≈ 0.1. The ﬂow ﬁeld being modeled is 7
Dissipation - ε Comparison axial loc. - x
Results are reported in terms of non dimensional k and ε plotted versus non dimensional wall distance at an axial location where ﬂow is fully developed, Fig. 6. Apart from CH  and HW , all models result to be in good agreement with turbulent kinetic energy experimental data for y + ≥ 40. The tendency, excluded the above mentioned and the LB  model, is to totally miss the peak registered for Copyright c 2007 - by ASME
8 7 6 5 4 3 2
exp Patel CH HW LW AKN CLL LB LNR
(a) Geometry mesh.
1.05 1 0.95 0.9 0.85 FOAM exp
1 0 0 20 40
0.8 0.75 0.7 0.65 0.6 0.55 0 0.005 0.01 0.01 0.02 x(m) 0.03 0.03 0.04 0.04
(a) k+ proﬁle
1/xy+ AKN CLL LB
exp Patel CH HW LW
(b) Pressure ratio over experimental data.
Figure 5. Stator Blade analysis.
y+ 10. LB estimation viceversa is higher than experimental registration. Level of approximation results in being quite uniform for the different models with CLL and AKN  slightly better in overlapping free stream values for k + . Also for turbulent dissipation, models well predict outer layer behavior: all models apart from LNR  basically coincide for y + ≥ 40. The major disagreements are registered inside the viscous layer. There is no agreement in fact in predicting the peak both in terms of positioning and values. Due to the different boundary conditions imposed to the dissipation, near wall behavior is quite different for each model: LB fails in predicting the peak, AKN and HW results in having pretty high wall values for ε. The models that best suit experimental data reported by Patel  are LNR and CLL. Of all the tested Low Reynolds models the CLL has been chosen as the most reliable one and used as the example for Low Reynolds models in the following cases.
0,00 0 20 40 60 80 100
(b) ε+ proﬁle Figure 6. Turbulence quantities proﬁles.
Heat Transfer tests Impingement Cooling Among various possible techniques to enhance heat transfer rate, impingement cooling certainly presents very high cooling efﬁcency, thus it’s commonly found both in typical blade and 8
combustor cooling systems, operating in a wide range of design conditions. From a numerical point of view, impinging jet ﬂows present several interesting aspects allowing deep evaluation of turbulence models. The problem of a 2-D normal impinging jet of air has been performed following a test case by ERCOFTAC. Then, the ﬁrst row of an array of holes was simulated, with a comparison of the experimental results obtained during the European project LOPOCOTEP, with different turbulence models. Further simulations of the complete array were done with the selected models. To obtain the desired heat transfer results, runs simulation with an imposed heat ﬂux on the impact wall has been performed. Copyright c 2007 - by ASME
ERCOFTAC C25 Axial-symmetric Impingement An incompressible ﬂow of a turbulent air jet impinging onto a ﬂat plate was modeled . The impact surface is heated and kept at constant heat ﬂux. From the experiment, the Nusselt number distribution for various jet Reynolds numbers is known. Fig. 7 shows the geometry of the test case. The diameter of the pipe is D = 0.004 m. The inﬂow velocity and turbulence conditions were obtained from the development of a 50 D upstream extruded inlet hole. For validation purpose, a Reynolds number of 23.000 and a distance of L/D = 2 were chosen. The heated surface was modeled as a wall at constant heat ﬂux q = 200 W/m2 . All ˙ other walls were treated as adiabatic walls. The far-ﬁeld boundaries are modeled as mixed inﬂow/outﬂow pressure boundaries.
Fig. 8 compares the predicted Nusselt number distributions of 5 turbulence models, namely AKN, SST, CLL, CLLReal and Two Layer , with the experimental proﬁle. The predictions of all two-equation models used in this validation case are in good agreement with the experimental data far from the stagnation point. As known in literature  in the area around the stagnation point Low Reynolds models without realizability constraint fail and dramatically overpredict the peak in the heat transfer coefﬁcient (error of almost 200%). At the same time Two Layer SST and realizable models show about the same peak value. The local maximum at r/D ≈ 2 is not seen by the Two Layer and is slightly predicted by the SST. On the contrary, the CLLReal is well predicting such peak only shifting a bit towards higher values of the radius. 1-Hole Impingement cooling Both this and the following case, simulating typical design conditions for impingement cooling of a gas turbine, has been performed on the set up of an experiment done at the Energy Engineering Department of the University of Florence for the European project LOPOCOTEP (LOw POllutant COmbustor TEchnical Programme). Coolant is injected from a plenum through a perforated plate and impacts over a ﬂat plate at uniform heat ﬂux. The holes on the plenum compose an array of 10 − 11 spanwise rows per 9 streamwise holes. This test is simulating the behavior of the ﬁrst row while in the following one 3 − 2 rows (the array of holes is staggered) for a total of 5 jets are impinging. For further details refer to . Main ﬂow parameters and grid are reported in Tab. 4 and Fig. 9.
Figure 7. Entire geometry and particular of the grid around the stagnation point.
Table 4. Flow conditions for 1-hole impingement test
Inlet Temperature Outlet Pressure
450 400 350 300 250
Exp AKN SST CLL CLLReal TL
308.2 85101 ≤ 0.5% 7600 0.28956 3000
K Pa % m/s W/m2
Inlet Turbulence level - Tu Rej Inlet Velocity Wall Heat ﬂux
200 150 100 50 0 0 1 2
Figure 8. Nusselt number distribution along radius.
Simulations have been validated in terms of heat transfer coefﬁcient calculated with respect to inlet static temperature almost coincident for such low Mach number with inlet total temperature. Adiabatic simulations have been done too, in order to check whether this approximation could be done or not. First thing to notice from Fig. 10 is that, contrarily to ERCOFTAC test, CLLReal model fails in well predicting heat transfer coefﬁcient around the stagnation point. Moreover, due to the potential core that is not extinguished at the wall, two unphysical spurious peaks are predicted at X/D ≈ 1. 9 Copyright c 2007 - by ASME
experimental TL path1
TL path2 SST path1 SST path2
HTC [Wm K ]
X / D
Figure 9. Impingement single hole grid. Figure 11. Heat transfer coefﬁcient along center lines.
kOmega SST ChenLienReal
HTC [Wm K ]
X / D
(b) Two Layer
Figure 10. Heat transfer coefﬁcient on impinged wall along symmetry line.
Two Layer and SST result in being almost equivalent both for the peak level and the far from the stagnation point values, with the Two Layer predictions slightly lower everywhere on the impinged surface. 5-Holes Impingement cooling This case refer to the same set of experiments of the previous test. For this multi-hole simulation the plenum as been schematized with a big plenum where the inlet mass ﬂow is imposed. Computational boundary conditions follow exactly the previous 1-hole test. For this geometry only the Two Layer and k−ω SST models have been tested against experimental results in terms of heat transfer coefﬁcient, see Fig. 11 and Fig. 12. Both experimental and numerical data are sampled onto the two different lines connecting symmetry planes and then merged together in the zone where a relative minimum is localized. 10
(c) k-ω SST Figure 12. Heat transfer coefﬁcient [W pinged wall.
m−2 K −1 ] distribution on im-
Even if obtained results are in good agreement with experimental data far from the stagnation point, it should be noticed that predictions for the peak value are quite different from measured data. Higher discrepancies on the even peaks are probably due to errors in the experimental measurements . Comparing the two models, Two Layer predicts peak values a 10% better of the SST giving basically identical results outside the stagnation points area. In any case, it should be considered that temperature gradients are quite small. A better agreement is expected for higher values of wall heat ﬂux. Copyright c 2007 - by ASME
Film and effusion cooling Among the different techniques used in the cooling of hot parts in a gas turbine engine, the injection of cooling air in the main ﬂow, producing a thin ﬁlm of air that isolates the walls from the hot gases, is one of the most used. Because of the complex interaction between air and hot gases during mixing, many different injection hole shapes and distribution have been studied and a great amount of research work is still on going . In particular, most recent developments in drilling capabilities allow the manufacturing of wide arrays of micro-holes (diameters below 1 mm), currently referred to as effusion cooling. Even if this technique does not produce a ﬁlm wall protection as in standard ﬁlm cooling, its most important feature is the heat removed by the passage of coolant inside the holes (heat sink effect): the great number of holes and their high length/diameter ratio (with angles below 30◦ ) allows to heavily increase the overall cooling effectiveness . Effusion cooling represents the base in the thermal design of modern aero-engine combustors and its use in the cooling of turbine endwalls is also investigated . Even if ﬁlm wall protection may not represent the main cooling effect in effusion technique, the prediction of mixing between coolant and cross ﬂow and the corresponding assessment of adiabatic effectiveness, still represent some of the most difﬁcult task in CFD analysis . Despite the well known deﬁciency of standard eddy viscosity turbulence models in the accurate prediction of jet mixing in cross ﬂow, essentially due to the isotropy assumption for turbulent stresses [54, 55, 56], both k − ε and k − ω models are still widely used in industrial CFD computations. Therefore we will analyze in this part the accuracy of OpenFOAM code in the prediction of adiabatic effectiveness in effusion cooling geometries, using the set of turbulence models selected for heat transfer analysis of this work. As introduced above, two test-cases will be studied: the well known single hole experiment by Sinha  and an experimental multi-hole geometry aimed at turbine endwalls cooling . Sinha test Experimental data and geometries are based on tests made by Sinha et al.; local and spanwise averaged effectiveness is compared with calculated values. The geometry is a ﬂat plate with a single row of holes, while ﬂow conditions are listed in table 5. In the chosen geometry, hot gas ﬂows over a ﬂat plate, while coolant is injected through one row of holes; upstream of the injection channel there is a plenum. Fig. 13 shows the ﬂuid domain and the different boundary conditions imposed; in particular, symmetry planes pass through hole axis and half spanwise pitch. On all inlet surfaces mass ﬂow rate and static temperature are imposed, symmetry conditions ensure zero gradient over boundaries in span direction. Single row conﬁguration was mainly considered in order to have a reference geometry to compare results for different turbu11
Table 5. Flow conditions for Sinha test
Cross ﬂow temperature Coolant temperature Pressure Density ratio - DR Blowing-rate - M Momentum ratio - I Turbulence level - Tu Cross ﬂow velocity Rec 300 153 105 2.0 0.5 0.125 ≤ 0.2 20 15700 % m/s K K Pa
Figure 13. Calculation domain and boundary conditions (Sinha al.).
lence models and calculation meshes. The performances of the same ﬁve turbulence models as single hole impingement case were analyzed, namely AKN, CLL, CLLReal, SST, TL. The grid used is a structured grid, see Fig. 14 in which it is also reported a magniﬁcation of the zone around the hole.
Figure 14. Mesh details near walls.
In Fig. 16, laterally averaged effectiveness downstream of the hole is shown. Local lateral effectiveness at 1, 10 and 15 diameters downstream, is also shown in Fig. 15. A map of wall Copyright c 2007 - by ASME
between numerical and experimental results for all models used.
exp CLL AKN
0,0 0,0 0,2 0,4 0,6 0,8 1,0 1,2 1,4
Y / D
0,00 0 5 10 15 20 25
X / D
(a) Laterally averaged
CLL AKN TL
Exp CLL AKN
0,0 0,0 0,2 0,4 0,6 0,8 1,0 1,2 1,4
Y / D
0,0 0 5 10 15 20 25
X / D
(b) Center line
exp CLL AKN
TL SST CLLReal
Figure 16. Comparison between laterally averaged and local center line ﬁlm cooling effectiveness.
0,0 0,0 0,2 0,4 0,6 0,8 1,0 1,2 1,4
Y / D
(c) 15D Figure 15. Spanwise distribution of ﬁlm cooling effectiveness at various sections.
effectiveness as well as distribution over the symmetry plane is reported in Fig. 17 and Fig. 18. There is a fairly good agreement 12
Attention must be paid on the results at one diameter test section for the CLLReal model. The bafﬂe underlined at the end of the spanwise direction is a consequence of a poor developing of the boundary layer predicted by the model. Low values of turbulent viscosity do not dissipate the horse-shoe vortex upstream of the hole. A low pressure zone drives coolant gas coming from the plenum around the jet. As a consequence there is more spreading of the ﬁlm in spanwise direction and the evidence is a local maximum for the effectiveness at X/D ≈ 1.2. On the other hand, centreline values show that SST model has a deeper penetration of the jet, see also Fig. 18, showing a local minimum in the effectiveness proﬁle just downstream the hole. It’s clear that numerical simulations predict a coherent jet Fig. 17, thus severally underestimating coolant lateral diffusion. Copyright c 2007 - by ASME
Table 6. Flow conditions for 6-holes effusion test
Cross ﬂow temperature 323 298 7.0 · 104 1.103 0.2 K K Pa
Coolant temperature Pressure Density ratio - DR Blowing-rate - M
(d) Two Layer
Figure 19. 6-holr effusion cooling case mesh. (e) map Color
Figure 17. Effectiveness distribution over the wall.
Results are reported in terms of spanwise averaged adiabatic effectiveness, see Fig. 20. Together with experimental data, correlative approach predictions using L’Ecuyer and Soechting correlation with Sellers superposition criterion have been reported . First of all it can be noticed that Two Layer model strongly
Experimental OF-TL OF-SST Ecuyer-Soechting
Figure 18. Sinha - Temperature distribution on symmetry plane [K].
This is not evident at 1 D downstream, but the effect grows proceeding in cross ﬂow direction. This behavior is mostly due to an isotropic modeling of turbulence near the wall, see Simon, Jubran, Azzi and Lakehal [57, 54, 55, 56]. Similar results can be found also in Andreini et al.  with commercial solvers. 6-Holes effusion cooling The geometry of this case is a six holes ﬂat plate interposed in between a plenum and a channel at lower pressure. To enhance numerical stability, the plenum has been gridded as six different smaller plena each one with the same inlet mass ﬂow imposed to respect total experimental cooling air mass ﬂow Fig. 19. A summary of ﬂow conditions can be found in Tab. 6. 13
0.00 0 10 20 30 40 50 60
X / D
Figure 20. Spanwise averaged adiabatic effectiveness.
improves matching of both experimental and correlative data in comparison to k − ω SST. Two Layer is still in slightly over prediction for the peak values especially for even peaks, for the odd ones in fact peak values result in being on the same level of the Copyright c 2007 - by ASME
previous hole, meaning that rows interaction is very weak. This lack is due to the assumption of isotropic behavior of the turbulent viscosity. Such effect is even stronger for the k − ω SST. This can be seen also on the map of adiabatic effectiveness in Fig. 21.
(a) Two Layer
The combination of the new built-in OpenFOAM libraries is able to reproduce the ﬂow conditions with good accuracy for all the geometries studied. Good agreement with experimental data and with the common commercial software has been reached for impingement and effusion cooling conﬁgurations. Further investigations have to be made especially for effusion cooling simulations. First of all, implementation of anisotropic turbulence models is needed in order to correct the lack of spanwise diffusion. The used object oriented language give us a very ﬂexible way for implementing new turbulence models, solver algorithms, boundary condition types and physical models. Future work will be concentrated on expanding the capability of the code to simulate ﬂuid-structure interaction, with main focus in conjugate heat transfer analysis.
ACKNOWLEDGMENT Many thanks to Dr. Hrvoje Jasak of Assistant professor (docent), Faculty of Mechanical Engineering and Naval Architecture (FSB), University of Zagreb, Croatia.and to OpenCFD Limited 2004-2007, Reading UK
(c) experimental Figure 21. Comparison between laterally averaged and local center line ﬁlm cooling effectiveness.
Both models qualitative well predict the correlative decay of the spanwise effectiveness downstream the holes. By looking at the two-dimensional effectiveness map in Fig. 21 it’s evident the reason why SST turbulence model over predict previously plotted data. In fact the jet predicted by SST model presents a different shape a signiﬁcant lateral diffusion as well as a more coherent structure along the ﬂow. Both phenomena could be attributed to a higher turbulent viscosity. Moreover, the SST model predicts a thinner and larger ﬁlm which is able to keep its cooling potential being less affected by the interaction with the main ﬂow in the shear layer. CONCLUSIONS A numerical investigation was set up to validate an open source CFD code based on object oriented programming language. Many different tests were performed representing the state of the art for the cooling systems in turbomachinery applications. Validation of a pressure correction algorithm and various turbulence models have been made by comparison with experimental data on typical heat transfer geometries. Massive parallel calculation have also been tested for the multirow conﬁguration simulations both for impingement and effusion cases by the use of LAM/MPI library http://www.lam-mpi.org. 14
References  Openfoam user guide. Technical report, OpenCFD Limited.  Openfoam programmers guide. Technical report, OpenCFD Limited.  Stephen R. Davis. C++ for Dummies. 5th Edition.  Bjarne Stourstroup. The C++ programming language. Addison Wesley, 1997.  Stanley and Lippmann. C++ Primer. Addison Wesley 3rd ed.  A. K. Sinha, D. G. Bogard, and M. E. Crawford. Filmcooling effectiveness downstream of a single row of holes with variable density ratio. ASME Journal of Turbomachinery, 113:442–449, 1991.  L. Arcangeli, M. Surace, L. Tarchi, D. Coutandin, and S. Zecchi. Correlative analysis of effusion cooling systems. ASME Turbo Expo, (GT2006-90405), 2006.  D. Cooper, D. C. Jackson, B. E. Launder, and G. X. Liao. Impinging jet studies for turbulence model assessment. part i: Flow-ﬁeld experiments. Int. J. Heat Mass Transfer, (36): 2675–2684, 1993.  A. Andreini, E. Di Carmine, B. Facchini, and M. Surace. Combustor liner cooling: numerical analysis of impingment geometries. 2005.  H. Jasak. Error Analysis and Estimation for the Finite Volume Method with Applications to Fluid Flows. PhD thesis, Imperial College of Science, Technology and Medicine, 1996. Copyright c 2007 - by ASME
 H. Jasak, H.G. Weller, and N. Nordin. In cylinder cfd simulation using a c++ object-oriented toolkit. SAE Technical papers, (2004-01-0110).  F. Juretic. Error Analysis in Finite Volume. PhD thesis, Imperial College of Science, Technology and Medicine, 2004.  J. R. Shewchuk. An introduction to the conjugate gradient method without the agonizing pain. Technical report, School of Computer Science, Carnegie Mellon University Pittsburgh, PA 15213, 1994.  S. V. Patankar. Numerical Heat Transfer and Fluid Flow. Taylor & Francis, USA, 1980.  W. Malasakera and H. K. Versteeg. Computational Fluid Dynamics. Longman Scientiﬁc, England, 1995.  M. Peric. Numerical methods for computing turbulent ﬂows. Technical Report VKI LS 2004-06.  J. H. Ferziger and M. Peric. Computational Methods for Fluid Dynamics. Springer, Germany, 2002.  M. Peric, Z. Lilek, and L. Demirdzic. A collocated ﬁnite volume method for predicting ﬂows at all speed. Journal of Numerical Methods in Fluids, 16:1029–1050, 1993.  M. Darbandi and G. E. Schneider. Application of an allspeed ﬂow algorithm to heat transfer problems. Numerical heat transfer, 35:695–715, 1999.  C. M. Rhie. Pressure based navier-stokes solver using the multigrid method. AIAA Journal, 27(8):1017–1018, 1989.  W. Shyy and M. E. Braaten. Applications of a generalized pressure correction algorith for ﬂows in complicated geometries. Advances and Applications in Computational Fluid Dynamics - ASME Winter Annual Meeting, pages 109–119, 1988.  W. Shyy and M. E. Braaten. Adaptive grid computation for inviscid compressible ﬂows using a pressure correction method. AIAA, ASME, SIAM, and APS, National Fluid Dynamics Congress, pages 112–120, 1988.  K. C. Karki and S. V. Patankar. Pressure based calculation procedure for viscous ﬂows at all speeds in arbitrary conﬁgurations. AIAA Journal, 27(9).  J. Rincon and R. Elder. A high resolution pressure based method for compressible ﬂow. Journal of Computation and Physics, 3:217–231, 1997.  F. S. Lien and M. A. Leschziner. A pressure-velocity solution strategy for compressible ﬂow and its application to shock/boundary-layer interaction using second-moment turbulence closure. Journal of ﬂuids engineering, 115:717– 725, 1993.  E. S. Politis and K. C. Giannakoglou. A pressure-based algorithm for high speed turbomachinery ﬂows. International journal for numerical methods in ﬂuids, 25:63–80, 1996.  J. J. McGuirk and G. Page. Shock capturing using a pressure-correction method. AIAA Journal, 28(10):1751– 1757, 1990. 15
 F. Moukalled and M. Darwish. Tvd schemes for unstructured grids. International Journal of heat and mass transfer, 46:599–611, 2003.  F. Moukalled and M. Darwish. Normalized variable and space formulation methodology for high-resolution schemes. Numerical heat transfer. Part B, fundamentals, 26:79–96, 1994.  F. Moukalled and M. Darwish. A uniﬁed formulation of the segregated class of algorithms for ﬂuid ﬂow at all speeds. Numerical heat transfer. Part B, fundamentals, 37:103– 139, 2000.  F. Moukalled and M. Darwish. A robust pressure based algorithm for multiphase ﬂow. International journal for numerical methods in ﬂuids, 41:1221–1251, 2003.  I. Senocak and W. Shyy. A pressure based method fot turbulent cavitating ﬂow computations. Journal of Computation and Physics, 176:363–383, 2001.  L. Davidson. An introduction to turbulence models. Technical Report Publication 97/2, Chalmers University of Technology, 2003.  J. Bredberg. On two equation eddy-viscosity models. Technical Report Internal report 01/8, Chalmers University of Technology, 2001.  V. C. Patel, W. Rodi, and G. Sheuerer. Turbulence models for near wall and low reynolds number ﬂows: a review. AIAA Journal, 26:1308–1319, 1993.  F. S. Lien. Computational modeling of 3-D ﬂow in complex ducts and passages. PhD thesis, University of Manchester, Institute of science and technology, 1992.  K. Abe, T. Kondoh, and Y. Nagano. A new turbulence models for predicting ﬂuid ﬂow and heat transfer in separating and reattaching ﬂows-i. ﬂow ﬁeld calculations. International Journal of Heat Mass Transfer, 37:139–151, 1994.  D. A. Yoder and N. J. Georgiadis. Implementation and validation of the chien k − ε turbulence model in the wind navier-stokes code. Technical Report 209080, NASA, Glenn Research Center, Cleveland Ohio, 1999.  F. S. Lien, W. L. Chen, and M. A. Leshziner. Lowreynolds-number eddy-viscosity modelling based on nonlinear stress-strain/vorticity relations. Engineering turbulence modelling and experiments, 3:91–100, 1996.  C. B. Hwang and C. A. Lin. Improved low-reynoldsnumber k − ε model based on direct numerical simulation data. AIAA Journal, 36(1):38–43, 1998.  C. K. G. Lam and K. Bremhorst. A modiﬁed form of the k −ε model for predicting wall turbulence. Journal of ﬂuids engineering, 103:456–460, 1981.  P. A. Durbin. On the k − ε stagnation point anomaly. International journal of heat and ﬂuid ﬂow, 17(1):89–90, 1996.  G. Medic and P. A. Durbin. Towards improved prediction of heat transfer on turbine blades. Journal of turbomachinery, 124:187–192, 2002. Copyright c 2007 - by ASME
 D. Lakehal, G. S. Theodoris, and W. Rodi. Threedimensional ﬂow and heat tranfer calculations of ﬁlm cooling at the leading edge of a symmetrical turbine blade model. International journal of heat and ﬂuid ﬂow, 22: 113–122, 2001.  F. R. Menter. Two equation eddy viscosity turbulence model for engineering applications. AIAA Journal, 32: 1598–1604, 1994.  F. R. Menter. Zonal two equation k − ω turbulence models for aerodynamic ﬂows. AIAA Paper, (93-2906), 1993.  A. Andreini, C. Carcasci, S. Gori, and M. Surace. Film cooling system numerical design: adiabatic and conjugate analysis. ASME paper, (HT2005-72042), 2005.  K.C. Karki. A calculation procedure for viscous ﬂows at all speeds in complex geometries. PhD thesis, University of Minnesota, 1986.  L. G. Goldman, , and K. L. McLallin. Cold-air annular cascade investigation of aerodynamic performance of core-engine-cooled turbine vanes: I solid vane performance and facility description. Technical report, NASA Technical Memorandum, 1977.  K. Wieghardt and W. Tillman. On the turbulent friction layer for rising pressure. Technical Report TM-1314, NACA, 1951.  D. E. Coles and E. A. Hirst. Computation of turbulent boundary layers. In AFOSRIFP Stanford Conference, volume 2. Stanford University, 1969.  B. Facchini and M. Surace. Impingement cooling for modern combustors: experimental analysis of heat transfer and effectiveness. Experiments in Fluids, (40):601–611, 2006.  K. M. B. Gustafsson. Experimental studies of effusion cooling. PhD thesis, Chalmers University of technology, Department of Thermo and Fluid Dynamics, 2001.  A. Azzi and D. Lakehal. Perspectives in modeling ﬁlm cooling of turbine blades by transcending conventional two-equation turbulence models. Journal of turbomachinery, (124):472–484, 2002.  A. Azzi and B. A. Jubran. Numerical modeling of ﬁlm cooling from short length steam-wise injection holes. Heat and mass transfer, (39):345–353, 2003.  D. Lakehal. Near-wall modeling of turbulent convective heat transport in ﬁlm cooling of turbine blades with the aid of direct numerical simulation data. Journal of turbomachinery, (124):485–498, 2002.  T. Simon. Film-cooling lateral diffusion and hole entry effects. NASA Lewis, (1996 Coolant ﬂow management workshop), 1996.
Copyright c 2007 - by ASME
This action might not be possible to undo. Are you sure you want to continue?