Beruflich Dokumente
Kultur Dokumente
Development and Validation of an Object Oriented CFD Solver for Heat Transfer and Combustion Modeling in Turbomachinery Applications
Candidato:
Relatore:
Correlatore:
Coordinatore:
Hrvoje Jasak
Abstract
A new object oriented code named OpenFOAM, released by OpenCFD Ltd and aimed at solving continuum mechanics problems, has been provided with suitable tools for predicting ows involving heat transfer and combustion in conditions typical of turbomachinery applications. OpenFOAM supports operatorbased implicit and explicit second and fourth-order Finite Volume (FV) discretization in three dimensional space and on curved surfaces. Expansions have been added in this directions: Development of an All-Mach pressure based solver for co-located arrangement, development of RANS eddy viscosity isotropic and anisotropic models with particular interest of wall treatment, development of combustion models for premixed, partially premixed and diusion ames and development of radiative heat transfer models. Tests for validation here presented concern geometries for impingement as well as lm and internal duct cooling turbine blade. While for combustion and radiation typical testcases for the validation was simulated. All the simulations results were compared against in-house and from literature experimental data.
ii
iii
iv
Heisenberg reportedly said on his deathbed that he would ask God two questions: why relativity and why turbulence? "I really think he may have an answer to the rst question."
vi
Acknowledgements
The author would like to express his gratitude to Prof. Bruno Facchini for agreeing to coordinate and to supervise this thesis and to Prof. Hrvoje Jasak for agreeing to act as a co-supervisor. I also would like to thank Stefano Traverso for his encouragement. As well I would like to thank all the guys in our group for their numerous discussions over the course of the code development. An heartfelt thankyou to all those people that trust my capabilities, to my family for everything they have done during these long years of studying and to my half sweetheart. Financial support for this study was provided by ANSALDO ENERGIA and is greatfully acknowledge.
vii
viii
Contents
Nomenclature 1 Basic CFD
1.1 1.2 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Physical principles and Mathematical model . . . . . . . . . . . . 1.2.1 Continuity equation . . . . . . . . . . . . . . . . . . . . . 1.2.2 Momentum equation . . . . . . . . . . . . . . . . . . . . . 1.2.3 Energy equation . . . . . . . . . . . . . . . . . . . . . . . 1.2.4 A model for the uid . . . . . . . . . . . . . . . . . . . . . Discretization approaches . . . . . . . . . . . . . . . . . . . . . . 1.3.1 Finite Volume Method . . . . . . . . . . . . . . . . . . . . 1.3.1.1 Approximation of Volume Integrals: Source Terms 1.3.1.2 Approximation of Surface Integrals . . . . . . . 1.3.1.3 Diusive ux . . . . . . . . . . . . . . . . . . . . 1.3.1.4 Convective ux . . . . . . . . . . . . . . . . . . . Finite Approximation: Interpolation and Dierentiation Schemes 1.4.1 Upwind Interpolation UDS . . . . . . . . . . . . . . . . . 1.4.2 Linear Interpolation CDS . . . . . . . . . . . . . . . . . . 1.4.3 Self Filtered Centered Interpolation SFCD . . . . . . . . . 1.4.4 Algebraic Equation . . . . . . . . . . . . . . . . . . . . . . Grid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Solution method . . . . . . . . . . . . . . . . . . . . . . . . . . .
xxiii 1
2 4 4 5 7 8 9 11 11 12 13 13 13 14 14 15 15 15 16
1.3
1.4
1.5 1.6
2 OpenFOAM
2.1 2.2 2.3 2.4 Introduction to C++ . . . . . . . OpenFOAM software . . . . . . . Matrix structure in OpenFOAM Working with main classes in OF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
21
22 23 25 26
29
30 30 31 32 36
ix
Co-located grid: Rhie - Chow interpolation Compressibility eects . . . . . . . . . . . . Critical aspects . . . . . . . . . . . . . . . . Solver validation . . . . . . . . . . . . . . . 3.9.1 GAMM tests . . . . . . . . . . . . . 3.9.2 Goldman test . . . . . . . . . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
37 39 40 42 42 43
4 Turbulence Modelling
4.1 4.2 4.3 4.4 The physics of turbulence . . . . . . . . . . . . . . . . . Models for Turbulence . . . . . . . . . . . . . . . . . . . RANS approach . . . . . . . . . . . . . . . . . . . . . . Standard k turbulence model . . . . . . . . . . . . . 4.4.1 Equation for turbulent kinetic energy . . . . . . 4.4.2 Equation for turbulent kinetic energy dissipation 4.4.3 Drawbacks of standard two equation model . . . 4.4.4 Wall functions . . . . . . . . . . . . . . . . . . . 4.4.5 Stability . . . . . . . . . . . . . . . . . . . . . . . Failures in predictions for heat transfer . . . . . . . . . . Low Reynolds k . . . . . . . . . . . . . . . . . . . . 4.6.1 Abe Kondoh Nagano . . . . . . . . . . . . . . . . 4.6.2 Chien . . . . . . . . . . . . . . . . . . . . . . . . 4.6.3 Chen Lien Leschziner . . . . . . . . . . . . . . . 4.6.4 Hwang Lin . . . . . . . . . . . . . . . . . . . . . 4.6.5 Lam Bremhorst . . . . . . . . . . . . . . . . . . . 4.6.6 Lien Leschziner . . . . . . . . . . . . . . . . . . . 4.6.7 Realizability . . . . . . . . . . . . . . . . . . . . Two Layer . . . . . . . . . . . . . . . . . . . . . . . . . . Anisotropic Turbulence Model . . . . . . . . . . . . . . . 4.8.1 Model Formulation . . . . . . . . . . . . . . . . . The k class . . . . . . . . . . . . . . . . . . . . . . . 4.9.1 Original k . . . . . . . . . . . . . . . . . . . 4.9.2 Baseline model . . . . . . . . . . . . . . . . . . . 4.9.3 Shear Stress Transport model . . . . . . . . . . . 4.9.4 Automatic Near Wall Treatment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
45
47 48 49 53 53 54 55 55 57 57 58 59 60 61 62 63 63 64 65 66 67 69 70 71 72 73
4.5 4.6
75
77 77 78 78 79 79 80 81 83
5.4 5.5
5.6
5.7
5.5.2 Laminar diusion ames modelling . . . 5.5.3 Mixture fraction . . . . . . . . . . . . . 5.5.4 Diusive amelets . . . . . . . . . . . . 5.5.5 Turbulent combustion diagram . . . . . 5.5.6 Turbulent diusion ames modelling . . 5.5.7 Presumed pdf amelets models . . . . . Premixed Flames . . . . . . . . . . . . . . . . . 5.6.1 Combustion regimes and types of ame 5.6.2 Turbulence and ame scales . . . . . . . 5.6.3 Spectral Diagram . . . . . . . . . . . . . 5.6.4 Premixed turbulent combustion models 5.6.5 Bray-Moss-Libby model . . . . . . . . . 5.6.6 Level-Set Model: G-Equation . . . . . . Partially Premixed Flames . . . . . . . . . . . . 5.7.1 Partially Premixed Flames Modelling .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
105
6.3
6.4
6.5
6.6
Radiation physic . . . . . . . . . . . . . . . . . . . . . . . . . . . 107 Denitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107 6.2.1 Radiative Intensity . . . . . . . . . . . . . . . . . . . . . . 107 6.2.2 Irradiance . . . . . . . . . . . . . . . . . . . . . . . . . . . 108 6.2.3 Radiant energy ux vector . . . . . . . . . . . . . . . . . . 108 6.2.4 Black Body Emissive Power . . . . . . . . . . . . . . . . . 108 Radiative Properties of Surfaces . . . . . . . . . . . . . . . . . . . 109 6.3.1 Emittance . . . . . . . . . . . . . . . . . . . . . . . . . . . 109 6.3.2 Absorbance . . . . . . . . . . . . . . . . . . . . . . . . . . 110 6.3.3 Reectance . . . . . . . . . . . . . . . . . . . . . . . . . . 110 Radiative Transfer Equation . . . . . . . . . . . . . . . . . . . . . 110 6.4.1 Interaction with Participating Media . . . . . . . . . . . . 110 6.4.1.1 Absorption . . . . . . . . . . . . . . . . . . . . . 111 6.4.1.2 Emission . . . . . . . . . . . . . . . . . . . . . . 111 6.4.1.3 Scattering . . . . . . . . . . . . . . . . . . . . . . 111 6.4.2 Derivation of RTE . . . . . . . . . . . . . . . . . . . . . . 112 6.4.3 Boundary conditions for RTE . . . . . . . . . . . . . . . . 112 Solution Methods For RTE . . . . . . . . . . . . . . . . . . . . . 113 6.5.1 Optical Thickness . . . . . . . . . . . . . . . . . . . . . . 113 6.5.2 Spectral Nature Modelling . . . . . . . . . . . . . . . . . . 113 6.5.2.1 Modelling of Radiative Properties Gaseous Combustion Product . . . . . . . . . . . . . . . . . . 113 6.5.2.2 Radiative properties of soot . . . . . . . . . . . . 115 6.5.3 Directional Nature Modelling . . . . . . . . . . . . . . . . 116 6.5.4 Spherical Armonics Method . . . . . . . . . . . . . . . . . 117 6.5.5 Finite Volume Method . . . . . . . . . . . . . . . . . . . . 119 Radiative Heat Transfer Eect On Overall System Energy Balance121
xi
123
Generalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124 Flat Plate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125 Impingement Cooling . . . . . . . . . . . . . . . . . . . . . . . . . 127 7.3.1 7.3.2 7.3.3 ERCOFTAC C25 Axial-symmetric Impingement . . . . . 127 Single hole case . . . . . . . . . . . . . . . . . . . . . . . . 131 Multirow Impingement Cooling . . . . . . . . . . . . . . . 134 Sinha Test . . . . . . . . . . . . . . . . . . . . . . . . . . . 138 Multirow Eusion Cooling . . . . . . . . . . . . . . . . . . 141 Fan-Shaped Cooling Holes . . . . . . . . . . . . . . . . . . 143 7.4.3.1 7.4.3.2 SAA Geometry . . . . . . . . . . . . . . . . . . . 145 PAA Geometry . . . . . . . . . . . . . . . . . . . 150
7.4
7.5
Blade internal cooling - Trailing Edge . . . . . . . . . . . . . . . 153 7.5.1 Trailing Edge Cooling System: AITEB2 . . . . . . . . . . 153 7.5.1.1 7.5.1.2 Geometry G1.0 . . . . . . . . . . . . . . . . . . . 158 Geometry G1.2A and geometry comparison . . . 159
161
Results: Diusion Flames . . . . . . . . . . . . . . . . . . . . . . 162 Numerical Modeling . . . . . . . . . . . . . . . . . . . . . 162 8.1.1.1 8.1.1.2 8.1.2 Libraries generation . . . . . . . . . . . . . . . . 162 Coupling with CFD code . . . . . . . . . . . . . 163
Model Validation: Blu Body SANDIA Burner . . . . . . 164 Numerical Modeling . . . . . . . . . . . . . . . . . . . . . 176 Model Validation: Lean Premixed Test Rig . . . . . . . . 177 . . . . . . . . . . . . 180 Numerical Modeling . . . . . . . . . . . . . . . . . . . . . 180 Model Validation: GE10 Burner . . . . . . . . . . . . . . 181
8.2
8.3
189
Results: P1 Model . . . . . . . . . . . . . . . . . . . . . . . . . . 190 RTE Numerical Modeling . . . . . . . . . . . . . . . . . . 190 Model Validation: Concentrical Cylinders Testcase . . . . 191 . . . . . . . . . . . . . . . . . . . . . . . 194 RTE Numerical Modeling . . . . . . . . . . . . . . . . . . 194 False Scattering . . . . . . . . . . . . . . . . . . . . . . . . 196 Model Validation: RADIARE . . . . . . . . . . . . . . . . 197 9.2.3.1 9.2.3.2 9.2.4 Case 1 . . . . . . . . . . . . . . . . . . . . . . . . 197 Case 2 . . . . . . . . . . . . . . . . . . . . . . . . 199 . . . . . 202
xii
205
10.1 Numerical Modeling . . . . . . . . . . . . . . . . . . . . . . . . . 206 10.1.1 Participating Media Absorption Coecient Model: Gaseous Media . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206 10.1.2 Soot absorption coecient . . . . . . . . . . . . . . . . . . 206 10.1.3 Soot modelling . . . . . . . . . . . . . . . . . . . . . . . . 206 10.1.4 Coupling with CFD code . . . . . . . . . . . . . . . . . . 207 10.2 Model Validation: Craneld Jet-k Burner . . . . . . . . . . . . . 208
Conclusions
217
xiii
xiv
List of Figures
1.1 1.2 1.3 1.4 1.5 1.6 2.1 2.2 3.1 3.2 3.3 3.4 4.1 4.2 Fluid element for conservation laws. . . . . . . . . . . . . . . . . 5 Stress component in the x direction. . . . . . . . . . . . . . . . . 6 Heat ux vector. . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 Typical bi-dimensional quadrilateral control volume. . . . . . . . 11 Typical structured C grid example. . . . . . . . . . . . . . . . . . 16 Typical hybrid tetrahedral hexaedral unstructured mesh example. 17 Polyhedral mesh example. . . . . . . . . . . . . . . . . . . . . . . UML diagram for GeometricField class. . . . . . . . . . . . . . . Bidimensional control volume. . . . . . . Prole Mach number in upper and lower Density gradient contour plots. . . . . . Stator Blade analysis. . . . . . . . . . . . . . wall. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 26 33 42 43 44
4.3
Da Vinci sketch of turbulent ow. . . . . . . . . . . . . . . . . . . 47 LES predictions of shear stress (top row), rate of strain (middle row), and eddy viscosity (bottom row) for the xy component (left column), yz component (middle column) and xz component(right column). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67 Comparison of analytical correlation and DNS data for the anisotropy factor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69 Premixed ame propagation speed . . . . . . . . . . . . . . . . . Examples of strained ames: a) steady strained one-dimensional diusion ame and b) counterow premixed ames . . . . . . . . Diusion ame structure . . . . . . . . . . . . . . . . . . . . . . . Example of ame structure in Z-space, methane-air equilibrium ame . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Regimes in nonpremixed turbulent combustion . . . . . . . . . . Shapes of the -function for dierent value of Z and . . . . . . Schematic representation of a premixed ame . . . . . . . . . . . Schematic representation of both amelet regimes. If Ka < 1 amelets are corrugated (a) while for 1 < Ka < 100 amelets are thickened (b) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Peters' combustion regimes diagram . . . . . . . . . . . . . . . . 80 81 82 86 87 91 93
5.9
95 96
xv
5.10 Schematic representation of a lifted jet diusion ame . . . . . . 103 6.1 6.2 6.3 6.4 7.1 7.2 7.3 7.4 7.5 7.6 7.7 7.8 7.9 7.10 7.11 7.12 7.13 7.14 7.15 7.16 7.17 7.18 7.19 7.20 7.21 7.22 7.23 7.24 7.25 7.26 7.27 7.28 Symbols for the denition of radiative intensity . . . . Scattering due to a small particles . . . . . . . . . . . Solution methods for the dierential radiative equation Coordinate system for FVM angular discretization . . . . . . . . . . . . . . . . . . . . . . . . . . 107 111 116 120
Turbulence quantities proles. . . . . . . . . . . . . . . . . . . . . 126 Stanton number prole. . . . . . . . . . . . . . . . . . . . . . . . 127 ERCOFTAC - Schematic view of experimental set up. . . . . . . 128 ERCOFTAC - Geometry detail of the grid around the stagnation point. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128 ERCOFTAC - Nusselt number distribution along radius. . . . . . 129 Turbulent kinetic energy distribution on symmetry plane [m2 /s2 ]. 130 Impingement single hole - Grid. . . . . . . . . . . . . . . . . . . . 131 Impingement single hole - Heat transfer coecient on impinged wall along symmetry line. . . . . . . . . . . . . . . . . . . . . . . 132 Impingement single hole - Heat transfer coecient distribution on impinged wall [W/(m2 K)]. . . . . . . . . . . . . . . . . . . . . 133 Impingement single hole - Velocity magnitude on cutting hole symmetry plane [m/s]. . . . . . . . . . . . . . . . . . . . . . . . . 134 Impingement single hole - Turbulent kinetic energy on cutting hole symmetry plane [m2 /s2 ]. . . . . . . . . . . . . . . . . . . . . 135 Impingement ve holes - Grid. . . . . . . . . . . . . . . . . . . . 135 Impingement ve holes - Heat transfer coecient along center lines.136 Impingement ve holes - Velocity magnitude on symmetry plane [m/s]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136 Impingement ve holes - Temperature distribution on symmetry plane [K]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137 Impingement ve holes - Heat transfer coecient distribution on impinged wall [W/(m2 K)]. . . . . . . . . . . . . . . . . . . . . . . 137 Sinha - Calculation domain. . . . . . . . . . . . . . . . . . . . . . 138 Sinha - Near wall mesh details. . . . . . . . . . . . . . . . . . . . 139 Spanwise distribution of eectiveness. . . . . . . . . . . . . . . . 140 Laterally averaged and local center line eectiveness. . . . . . . . 140 Eectiveness distribution over the wall. . . . . . . . . . . . . . . 141 Mesh details . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142 Spanwise averaged adiabatic eectiveness. . . . . . . . . . . . . . 143 Comparison between laterally averaged and local center line lm cooling eectiveness. . . . . . . . . . . . . . . . . . . . . . . . . . 143 SAA Computational grids: coarser 7.25(a) and ner 7.25(b) . . 144 PAA Computational domain . . . . . . . . . . . . . . . . . . . . 145 Dierences between numerical and experimental domain . . . . . 145 Local lateral for three location downstream for SAA smooth conguration BR=0.73 . . . . . . . . . . . . . . . . . . . . . . . . 146
xvi
7.29 Local lateral for three location downstream for SAA smooth conguration BR=1.31 . . . . . . . . . . . . . . . . . . . . . . . . 7.30 Local lateral for three location downstream for SAA sharp conguration BR=0.74 . . . . . . . . . . . . . . . . . . . . . . . . . . 7.31 Local lateral for three location downstream for SAA sharp conguration BR=1.29 . . . . . . . . . . . . . . . . . . . . . . . . . . 7.32 Numerical adiabatic eectiveness map for SAA smooth conguration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.33 Experimental adiabatic eectiveness map for SAA smooth conguration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.34 Numerical adiabatic eectiveness map for SAA sharp conguration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.35 Experimental adiabatic eectiveness map for SAA sharp conguration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.36 Spanwise averaged eectiveness < > for SAA geometry for smooth (7.36(a), 7.36(b)) and sharp (7.36(c), 7.36(d)) congurations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.37 Local lateral for three location downstream for PAA smooth conguration BR=0.75 . . . . . . . . . . . . . . . . . . . . . . . . 7.38 Local lateral for three location downstream for PAA smooth conguration BR=1.36 . . . . . . . . . . . . . . . . . . . . . . . . 7.39 Numerical adiabatic eectiveness map for PAA smooth conguration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.40 Experimental adiabatic eectiveness map for PAA smooth conguration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.41 Spanwise averaged eectiveness < > for PAA geometry and smooth conguration . . . . . . . . . . . . . . . . . . . . . . . . 7.42 Typical compound cooling system of a blade. . . . . . . . . . . . 7.43 Computationl domain and boundaries . . . . . . . . . . . . . . . 7.44 Computational grids: pedestal and rib details for unstructured and structured meshes . . . . . . . . . . . . . . . . . . . . . . . . 7.45 Geometry G1.1A: heat transfer coecient map . . . . . . . . . . 7.46 Geometry G1.1A: heat transfer coecient distribution . . . . . . 7.47 Geometry G1.0: heat transfer coecient map . . . . . . . . . . . 7.48 Geometry G1.0: heat transfer coecient distribution . . . . . . . 7.49 Geometry G1.0: contour of x vorticity component on six sections normal to x axis . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.50 Geometry G1.2a: heat transfer coecient map . . . . . . . . . . 7.51 Geometry comparison: heat transfer coecient distribution . . . 7.52 Geometry comparison: velocity vector plot . . . . . . . . . . . . . 7.53 Geometry comparison: static pressure distribution . . . . . . . . 8.1 8.2
149 150 150 151 151 151 153 154 155 156 156 157 158 159 159 159 160 160
Diusion ame model owchart . . . . . . . . . . . . . . . . . . . 164 BBody burner: geometry and ame images . . . . . . . . . . . . 165
xvii
8.3 8.4 8.5 8.6 8.7 8.8 8.9 8.10 8.11 8.12 8.13 8.14 8.15 8.16 8.17 8.18 8.19 8.20 8.21 8.22 8.23 8.24 8.25 8.26 8.27 8.28 8.29 8.30 8.31 8.32 8.33 8.34 8.35 8.36
BBody burner: schematic representation of the ame zones with some sampling sections . . . . . . . . . . . . . . . . . . . . . . . . 165 BBody mesh . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166 BBody burner: mixture fraction (left) and temperature contours (right) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167 BBody burner: OH and CO mass fraction contours . . . . . . . 168 BBody burner: details of velocity eld contour in the recirculation zone with streamlines . . . . . . . . . . . . . . . . . . . . . . 169 BBody burner: mixture fraction variance contour near blu body. 169 BBody burner: mixture fraction average proles. . . . . . . . . . 170 BBody burner: mixture fraction variance proles. . . . . . . . . . 171 BBody burner: temperature proles. . . . . . . . . . . . . . . . . 172 BBody burner: species proles at Y=13 mm. . . . . . . . . . . . 173 BBody burner: species proles at Y=30 mm. . . . . . . . . . . . 173 BBody burner: species proles at Y=45 mm. . . . . . . . . . . . 174 BBody burner: species proles at Y=65 mm. . . . . . . . . . . . 174 BBody burner: species proles at Y=90 mm. . . . . . . . . . . . 175 BBody burner: species proles at Y=120 mm. . . . . . . . . . . . 175 Localization of mean ame front by linear interpolation. . . . . . 176 Vanderbilt combustor geometry and boundary conditions. . . . . 178 Contour plots of temperature and progress variable source for tfc model (left) and of temperature and iso-g for level-set model (right).178 Averaged temperature proles at three dierent locations x/d = 0.1, x/d = 0.8 and x/d = 1.0. . . . . . . . . . . . . . . . . . . . . 179 CO Mole fraction. . . . . . . . . . . . . . . . . . . . . . . . . . . 180 Flowchart of BML for partial premixed ame. . . . . . . . . . . . 181 GE 10: schematic representation . . . . . . . . . . . . . . . . . . 182 GE 10: schematic representation of the burner, property of GE-NP182 GE 10: used mesh, property of GE-NP . . . . . . . . . . . . . . . 184 GE 10: schematic representation of geometrical model and ame 184 GE 10: plot of mixture fraction near combustor inlet . . . . . . . 185 GE 10: velocity contours, left ST AR CDT M right F OAM . . 186 GE 10: vector plot of velocity eld near combustor cap by (Andreini, 2004) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186 GE 10: vector plot of velocity eld near combustor cap by the implement model . . . . . . . . . . . . . . . . . . . . . . . . . . . 186 GE 10: progress variable contours from right to left: KPP closure, TFC closure and ST AR CDT M KPP closure . . . . . . . . . . 187 GE 10: temperature contours from right to left: KPP closure, TFC closure and ST AR CDT M KPP closure . . . . . . . . . . 187 GE 10: ame lift, ST AR CDT M . . . . . . . . . . . . . . . . . 187 GE 10: ame lift, FOAM-KPP . . . . . . . . . . . . . . . . . . . 187 GE 10: ame lift, FOAM-TFC . . . . . . . . . . . . . . . . . . . 187
xviii
8.37 GE 10: closure, 8.38 GE 10: closure, 9.1 9.2 9.3 9.4 9.5 9.6 9.7 9.8 9.9 9.10 9.11 9.12 9.13 9.14
progress variable source term from right to left: TFC closure and ST AR CDT M KPP closure . progress variable source term from right to left: TFC closure and ST AR CDT M KPP closure .
Flowchart of P1 solving algorithm . . . . . . . . . . . . . . . . . 191 Flowchart of FMV-RTE solution . . . . . . . . . . . . . . . . . . 195 Bidimensional control volume for RTE . . . . . . . . . . . . . . . 196 RADIARE case 1: well-stirred combustion chamber . . . . . . . 198 RADIARE perfectly stirred combustor wall radiative heat ux contourt. Emissivity=0.5 . . . . . . . . . . . . . . . . . . . . . . 198 RADIARE perfectly stirred combustor wall radiative heat ux vs z coordinate. Emissivity=0.5 . . . . . . . . . . . . . . . . . . . 199 RADIARE perfectly stirred combustor wall radiative heat ux vs z coordinate. Emissivity=0.8 . . . . . . . . . . . . . . . . . . . 199 RADIARE case 2: cylindrical combustion chamber. . . . . . . . 200 RADIARE case 2 temperature distribution for short-ame (upper) and long-ame (lower) . . . . . . . . . . . . . . . . . . . . . 200 RADIARE case 2 quarter mesh . . . . . . . . . . . . . . . . . . . 201 RADIARE case 2 wall irradiance by short-ame . . . . . . . . . 202 RADIARE case 2 wall irradiance by long-ame . . . . . . . . . . 203 Spherical domain unstructured mesh . . . . . . . . . . . . . . . . 203 Spherical domain %-error on enthalpy sink vs absorption coecient204
10.1 Diusion ame combined with soot and radiative heat exchange modelling owchart . . . . . . . . . . . . . . . . . . . . . . . . . . 208 10.2 Jet-k: schematically representation . . . . . . . . . . . . . . . . . 209 10.3 Jet-K burner : axial plots . . . . . . . . . . . . . . . . . . . . . . 211 10.4 Jet-K burner : radial plots at 100 mm from burner . . . . . . . . 212 10.5 Jet-K burner : radial plots at 300 mm from burner . . . . . . . . 213 10.6 Jet-K burner : absorption coecient plots . . . . . . . . . . . . . 214 10.7 Jet-K burner : temperature contours left P1, right FVM . . . . . 215 10.8 Jet-K burner : enthalpy radiative sink contours left P1, right FVM 215 10.9 Jet-K. From left to right contours of: mixture fraction Favre's average, hydroxide mass fraction, soot source term [s1 ] and soot mass fraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
xix
xx
List of Tables
4.1 4.2 Standard coecients values for standard k . . . . . . . . . . . Standard coecients values for Abe et al. (1994) Low Reynolds k . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3 Standard coecients values for Chien Low Reynolds k . . . . 4.4 Standard coecients values for Chen et al. Low Reynolds k . 4.5 Standard coecients values for Hwang and Lin Low Reynolds k . 4.6 Standard values for Lam and Bremhorst Low Reynolds k coecients. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.7 Standard coecients values for Lien and Leschziner Low Reynolds k . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.8 Standard coecients values for Two Layer k . . . . . . . . . . 4.9 Standard values for k coecients. . . . . . . . . . . . . . . . 4.10 Standard values for k BSL coecients. . . . . . . . . . . . . 4.11 Standard values for k SST coecients. . . . . . . . . . . . . 7.1 7.2 7.3 7.4 7.5 7.6 7.7 7.8 7.9 7.10 7.11 7.12 7.13 7.14 8.1 8.2 Acronyms for the various turbulence models. . . . . . . . . . . Typical values for the underrelaxation factors used. . . . . . . . Flat plate - Computational boundary conditions. . . . . . . . . ERCOFTAC - Flow conditions. . . . . . . . . . . . . . . . . . . Impingement single hole - Flow conditions. . . . . . . . . . . . Impingement single hole - Computational boundary conditions. Sinha - Flow conditions. . . . . . . . . . . . . . . . . . . . . . . Sinha - Computational boundary conditions. . . . . . . . . . . Eusion - Flow conditions. . . . . . . . . . . . . . . . . . . . . . Eusion - Computational boundary conditions. . . . . . . . . . Blowing Ratio and Reynolds number for hot tests . . . . . . . . Geometries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Boundary conditions . . . . . . . . . . . . . . . . . . . . . . . . Grid characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54 60 61 62 63 63 64 66 70 72 73 124 124 125 129 132 132 139 140 142 142 144 154 155 155
BBody INLETS boundary conditions . . . . . . . . . . . . . . . 166 GE10: inlet air condition, TCD and PCD stand for Temperature and Pressure at Compressor Discharge . . . . . . . . . . . . . . . 183 P1 cylinders: temperatures of inner T1 and outer T2 cylinders. . 192
9.1
xxi
9.2 9.3
P1 cylinders: %Error on nondimensional heat ux P si versus optical thickness . . . . . . . . . . . . . . . . . . . . . . . . . . . 193 Constants value for RADIARE case 2 temperature distribution . 200
10.1 Jet-K: Sandia absorption coecient constants . . . . . . . . . . . 206 10.2 Jet-K: boundary conditions. . . . . . . . . . . . . . . . . . . . . . 209
xxii
Nomenclature
Symbols
A a b C Cp C1 , C2 C1 , C2 c c d E f f1 , f2 f F G h0 H h h HT C i, j, k i J j k+ k k l l lt
generic coecient matrix, generic diagonal coecient in algebraic equation generic o diagonal coecient in algebraic equation mass imbalance coecient for eddy viscosity constant pressure specic heat coecients in specic turbulent dissipation equation coecients in turbulent dissipation equation speed of light in vacuum speed of sound specic energy damping function for eddy viscosity damping functions for equation external force per unit volume external force, surface mass ux explicit part of turbulent kinetic energy production specic total enthalpy remainder specic static enthalpy specic thermochemical static enthalpy heat transfer coecient generic counters specic internal energy generic ux vector generic face ux k non-dimensional turbulent kinetic energy = u2 absorption coecient, turbulent kinetic energy thermal conductivity turbulent length scale for eddy viscosity turbulent length scale for dissipation turbulent length scale
[ ],[A] [A] [kg/s] [] [J/(kgK)] [] [] [m/s] [m/s] [m/A] [J/kg] [] [] [N/m3 ] [N ], [kg/s] [kg/(m3 s)] [J/kg] [] [J/kg] [J/kg] [W/(m2 K)] [] [J/kg] [/(m2 s)] [/(m2 s)] [] [m1 ], [m2 /s2 ] [W/(m K)], [m2 /s2 ] [m] [m] [m]
x a
xxiii
mass normal versor number of cell turbulent kinetic energy production term pitch pressure turbulent Prandtl number Prandtl number rate of heat source term source term per unit volume heat ux vector gas constant perfect gases universal constant 2 Reynolds turbulent number = k
[kg] [] [] [kg/(m3 s)] [m] [P a] [] [] [W ] [/s] [/(m3 s)] [W/m2 ] [J/(kg K)] [J/(kg K)] [] [] [] [] [s1 ] [W/m3 ], [m2 ] [] [] [s] [K] [s] [%] [P a] [m/s] [m/s] [m/s] [m/s] [m/s] [m/s] [W ] [m] [m] [] [m] [] []
ky Reynolds y number = Reynolds' number light beam versor rate of strain tensor work done by body forces per unit volume, surface Schmidt's number non dimensional temperature turbulent time scale temperature time 2 3k turbulent intensity = u U U Reynolds stress tensor velocity vector pseudo velocity friction velocity equivalent friction velocity generic velocity component velocity components in x,y,z direction rate of work generic cartesian coordinate cartesian coordinates non-dimensional wall distance yu wall distance mixture fraction mixture fraction variance
Greek symbols
[m2 /s], [ ]
xxiv
ij x, y +
absorbance constant for turbulent kinetic energy production term k class v specic heat ratio, blending factor, w diusivity coecient Kronecker delta dimensions of bidimensional control volume turbulent kinetic energy dissipation non-dimensional turbulent kinetic energy dissipation = u4 emittance adiabatic eectiveness averaged adiabatic eectiveness Von Karman's constant = 0.41 distance weighting factor, bulk viscosity viscosity eddy viscosity kinematic viscosity density coecients for eddy viscosity in k , and equation shear stress generic scalar, viscous dissipation 1 RT - compressibility factor net ux through surface specic turbulent kinetic energy dissipation volume solid angle
[] [] [ ], [ ], [ ] [1/sm] [] [m] [m2 /s3 ] [] [] [] [] [] [ ], [kg /(m s)] [kg /(m s)] [kg /(m s)] [m2 /s] [kg/m3 ] [] [P a] [], [W/m3 ] [kg/J] [/s] [s1 ] [m3 ] [strad]
<> t k , ,
Subscripts
b b c E, W, N, S e, w, n, s EE g m n nb P R ref
body blackbody, or burnt state (BML model) coolant east, west, north, south cell neighbor east, west, north, south face right (east) neighbor of east neighbor hot gas wavelenght [nm] counter for inner iteration normal direction neighbor cell of reference node referred to radiation reference
xxv
s T u w
Superscripts
n
outer iteration counter correction, Reynolds uctuation Favre uctuation guessed Reynolds averaged Favre averaged
xxvi
Introduction
One of the most demanding task in gas turbine design is the proper evaluation of all phenomena involved with heat transfer. Of particular interest are the compound cooling systems for the hot components of the engine. Precise investigations for the zones in which heat transfer to solid walls is higher is fundamental as long as actual design criteria force cooling system to more and more limiting conditions. Cooling system is extremely important in order to raise the gases temperature at turbine inlet that results in an improvement of the machine performances and consequently in fuel economy. Thinking about blade and combustion chamber cooling, the need for improvements in overall performances tend to increase temperature in turbine, at the same time the amount of available cooling air is wanted to be as low as possible to keep work losses at the minimum. For combustors the situation is not so dierent with the new lean premixed ames design criteria reducing the quantity of air for liner cooling. The combustor is the device that gives the power to the machine and by doing this it burns fuel and produces pollutants, it is then clear that a proper combustor design is fundamental in order to meet either performances or pollution reduction goals. So the combustor design has two main tasks, both aecting each other: the reaction zone design and the combustor cooling system design. Accurate heat transfer measurements are however very expensive to support due to the complexity of geometries, the high costs of measuring apparatus and the long set up necessary to collect reliable data. This is why CFD analysis is becoming more and more popular in each phase of the design process. Nevertheless, CFD simulations for evaluation of thermal loads and eectiveness of the cooling devices in gas turbine engines are one of the most complex to face. Complexity arises especially because of the need for advanced physical modelling and exibility in geometrical mesh handling. Typical example for the necessity of particular models is the treatment of turbulence both in combustion and heat transfer analysis. Complexity grows because the good functioning of these models is quite case dependent. Viceversa, other kinds of cooling devices are very demanding in terms of mesh handling: internal ducts are usually improved with non aerodynamic turbulator such as ribs or pins, where a very ne grid is required to resolve the complex structure of the local ow. The capability to handle hybrid unstructured meshes and the availability of
xxvii
a quite large set of turbulence models become a must for a CFD code aimed at solving heat transfer related problems. Unfortunately these requirements result to be quite strict and not many softwares worldwide can satisfy them. That is why, in the last decade, both industrial and academic researchers have been widely using only few well known commercial solvers. Commercial solvers aspire to resolve many kinds of ow and so they are often provided with a big amount of dierent packages for various models. This generality often results in poor performances in terms of calculation times due to a waste of system resources for a large part of packages that are not used in standard simulations. Apart from this, the main drawback, according to expert users, stands in how commercial CFD codes behave like a black box solution maker. Especially in case of disagreements with experiments, it is not of secondary importance to perfectly understand under which hypothesis numerical predictions could be generated. Information on the process used to obtain convergence can be found only on often poor of details user manuals: source codes are obviously inaccessible. Advanced users in heat transfer and combustion modelling applications sometimes need the use of ad hoc models or modications suitable for specic cases. User subroutine features provided by commercial packages become quickly inadequate as the complexity of modications 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. This work is a project aimed to build a software package containing many of the CFD tools needed for turbomachines cooling and combustion systems design. The objective of this thesis 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) is an object-oriented numerical simulation toolkit written in C++ language. Besides its advanced basic native CFD features its essential characteristic, in opposition to commercial solvers, 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 reduction in debugging time. This thesis describes the attempt to build a CFD package suitable for typical steady state heat transfer analysis and to add the combustion and radiative heat exchange modelling. As will be described later, to reach such goal it was necessary to introduce specic features in the package in order to overcome the limitations of built-in approaches: rst of all a compressible steady state solver capable of handling transonic ows, then a set of turbulence two-equations closures with particular reference to a detailed near wall treatment. Concerning
xxviii
combustion modelling the software is now capable of simulating diusion and partially premixed ames aected by soot formation and radiative exchange. As conrmation of the work done a set of validation applications were performed. These capabilities have been tested by proper testcases and we will focus our attention on the validation of the code with some complex congurations typical of heat transfer and combustion problems.
Thesis Outline
Chapter 1 is based on a review of classical arguments concerned with Computational Fluid Dynamics. All the dierent parts numerical solution methods are composed by, are briey presented with no aim of completeness but as hint of the approach used in this thesis. Chapter 2 presents main features of the base code OpenFOAM with an overview on the C++ philosophy of programming, a description of how the linear systems are built and an example of the structure of one of the most common classes. Moreover an example on how to write a postprocessing utility is given. Chapter 3 describes the solver used. First general themes regarding solution algorithms such as pressure velocity coupling and segregation are introduced. Thus the ancestor of the solver used is described and nally expansion to colocated approach and compressibility eects are presented. Chapter 4 introduces the problem of modeling turbulence as rst thing. Eddy viscosity RANS models are introduced rst deriving standard k model, then presenting a number of dierent models for Low Reynolds grids. Finally the specic OpenFOAM class for turbulence models is described and commented. Chapter 5 describes the RANS equations set for a combustion system together with perfect gases equations of state and chemical kinetics. Chapter 5.5 presents the diusion ame modeling starting from laminar ames models that are further used in turbulent combustion modelling after a brief description of combustion regimes. It also deals with the soot formation which is peculiar of diusion ames. Chapter 5.6 describes those models suitable to simulate premixed ames modelling and provides details on combustion regimes. Chapter 5.7 shows how the merge of diusion and premixed ames models is performed in order to analyse partially premixed combustion systems. Chapter 6 briey presents the radiative heat exchange phenomenon that leads to the denition of the radiative transfer equation. Then are reported the solution methods for this equation that represent the radiative transfer models. Chapter 7 reports the results for the simulations done to validate the models proposed. After an initial test over a at plate, typical geometries for turbomachinery cooling devices are presented. Chapter 8 reports the implementation procedure of the diusion, premixed and partially premixed ames models appliyed to reference test case and indus-
xxix
trial combustor . Chapter 9 shows how the Finite Volume Method and P1 models for radiative transfer has been implemented. The Finite Volume is validated by three testcases, the rst two by RADIARE program, the last one, representing a spherical blackwalled domain, is intended to test the capability to work with unstructured meshes as well as to analyse the behaviour of the model in troublesome conditions. While the P1 is validated by comparing obtained results to an analytical solution by Modest (2003). Chapter 10 describes how the radiative transfer modelling has been coupled to the diusion ames one. Moreover has been implemented a soot formation model, being this specie heavily produced in jet engines combustors and strongly aecting the radiative transfer. The model has been validated by the Craneld University Jet-K testcase representing a diusion Jet-A ame. Chapter 10.2 gives the conclusions of this research work.
xxx
Chapter 1
Basic CFD
This chapter introduces and discusses the basilar aspects of Computational Fluid Dynamics starting from a general overview, with the aim of organizing such branch into the more complex environment of uid ow predictions, and going through the structural parts in which numerical methods can be subdivided. In particular all basic aspects of numerical prediction techniques are touched, starting from the governing equations of the mathematical model, giving a brief introduction to discretization methods and nite approximation and mentioning the methods for solving the resulting system. In the following particular interest will be placed on the Finite Volume Method.
2 4
4 5 7 8
1.3
Discretization approaches . . . . . . . . . . . . . .
9
11 11 12 13 13
1.4
Finite Volume Method . . . . . . . . . . . . . . . . . 1.3.1.1 Approximation of Volume Integrals: Source Terms . . . . . . . . . . . . . . . . . . . . . 1.3.1.2 Approximation of Surface Integrals . . . . 1.3.1.3 Diusive ux . . . . . . . . . . . . . . . . . 1.3.1.4 Convective ux . . . . . . . . . . . . . . . .
13
. . . . 14 14 15 15
1.5 1.6
. . . .
15 16
Chapter 1.
Basic CFD
1.1 Overview
Prediction of heat transfer and uid ow processes can be obtained by two main methods: experimental investigation and theoretical calculation. Even if reliable information about physical processes could only be obtained by actual measurement, experimental investigations involving full scale equipment often result to be too expensive. The alternative is so to use scaled models and conditions, at the end extrapolate the results to full scale. This scaling, however, is not completely free from errors: general rules for correct scaling are often unavailable, phenomena may not be scalable (i.e. combustion, turbulence, etc ...), measuring instruments errors may weight more. Such problems can be avoided with the help of a mathematical model able to suitably represent the physical process. Theoretical predictions consist in working out the consequences of such model. For uid dynamics problems, the mathematical model basically consists in a set of partial dierential equation. If classical mathematics techniques were to be used for solving such equations, there would be little hope of predicting many cases of practical interest with closed form solution. Hopefully, with the development of numerical methods and the availability of large digital computers, closure can be found almost for any practical problem. Furthermore, computer analysis oers several advantages with respect to experimental investigation: low cost, speed, complete and detailed information, capability in simulating both realistic and ideal conditions. At the same time, numerical calculation are not free from disadvantages: it can happen that a suitable mathematical model for describing the physical conditions cannot be found, it is possible for prediction with a very limited objective not to be cheaper than experiments, problems involving complex geometries, strong non linearity may be harder, longer and again more expensive to solve. It appears clearly now how experiments and computations must coexist and interact to have wide and reliable predictions (Patankar, 1980). Let's now analyze in detail what a numerical solution method is composed by. As above written, the starting point of a numerical method is the Mathematical Model, i.e. the set of partial dierential equations and boundary conditions. Fluid dynamic science teaches that exact conservation laws describe the behavior of all ows: no matter the type of ow, it will respect the general governing equations. General purpose methods however are often impractical, if not impossible, to solve so it is more convenient to include simplications in the mathematical model and develop a solution method designed for that particular set of equations. Then a suitable Discretization Method, approximating the set of dierential equations by a system of algebraic equations for the variables at a number of discrete points in space and time, is necessary. The most important discretization methods are: Finite Dierence Method (FDM), Finite Volume Method (FVM) and Finite Element Method (FEM). The discrete locations where the variables want to be calculated, are dened by the Numerical Grid. The numerical grid is a discrete representation of the ow domain (both
Basic CFD
Chapter 1.
in space and time) through the use of a nite number of subdomains such as elements, control volumes etc... Then a Finite Approximation technique has to be selected taking in consideration the choice for the discretization method and the numerical grid. This choice inuence a lot the accuracy of the solution as well as the development, coding, debugging and the speed of the solution method. More accurate approximations involve, in fact, more nodes and give usually a fuller coecient matrix. A compromise between accuracy and eciency is always necessary. Once this large system of non-linear algebraic equations has been built by discretization techniques, it must be solved using a Solution Method. Such methods use successive linearization of the equations and the resulting linear systems are almost always solved by iterative techniques. Usually there are two levels of iterations: inner iterations, within which the linear equation are solved, and the outer iterations, that deal with the non linearity and coupling of the equations. As last point, it is important to determine suitable Convergence Criteria. It is fundamental to well set stopping conditions for both the inner and the outer cycles in order to obtain accurate solution in an ecient way. Once dened, numerical methods must be checked to posses certain properties in order to establish whether a method is appropriate or not. The most important properties are:
Consistency: discretization should become exact as the grid spacing tends to zero. In other words truncation error, i.e. the dierence between exact and discretized equation, must go to zero as t 0 and x 0. Stability: errors appearing in the course of numerical solution process do not magnify. For iterative methods, stable methods are the ones that do not diverge. Convergence: the solution of the discretized equation tends towards the exact solution as the grid spacing tends to zero. It is a very dicult property to demonstrate, it is usually accepted to test grid-independence for a solution. Conservation: solution must respect conservation of physical quantities both on local and global scale. It is a very important property because limits solution error. Even if on ne grids non-conservative schemes can also lead to correct solutions, conservative ones are usually preferred. Boundedness: solution should lie within proper bounds. Boundedness is dicult to guarantee and often unbounded schemes have stability and convergence problems too. Realizability: guarantee a model to give physically realistic solutions for the phenomena it is representing. Usually in connection with phenomena too complex to be directly simulated.
Chapter 1.
Basic CFD
Accuracy: is the property of well approximating the exact solution, in other words limiting modeling, discretization and iteration errors.
D = + (U Dt t
),
(1.1)
meaning that the rate of increase of a scalar of uid particle, on the left hand side of Eq. 1.1, is equal to the rate of increase of uid element plus the net rate of ow of the scalar out of uid element, on the right hand side of the same equation (Anderson, jr., 1995). Secondly an appropriate model for the uid is proposed: simplications for the cases of interest such as ideal gas and Newtonian uid are introduced. Third passage in manipulating the set of governing equations, i.e. steadystate simplication and turbulence modeling, is explicitly done in chapter 4 but has been widely used in all the cases of need in the previous chapters too.
Dm =0. Dt
(1.2)
With reference to Fig. 1.1 the rate of increase of mass in the uid element is:
( x y z) = x y z , t t
(1.3)
Basic CFD
Chapter 1.
thus the net rate of ow of mass into uid element is:
(1.4)
Eq. 1.3 and Eq. 1.4 together give the well known dierential continuity equation in conservation form:
+ t
(U ) = 0 .
(1.5)
DU = Fs + Fb . Dt
(1.6)
It is easier to separate external forces into surface forces Fs and body forces Fb . The rst ones being dened upon control volume boundaries, the second ones on the volume itself. It is common to separate surface forces into pressure, the hydrostatic part of the stress tensor, and viscous stress, the deviatoric part. To derive x-momentum equation it is better start with a balance of surface
Chapter 1.
Basic CFD
p 1 x x 2 p + p+ x p
xx 1 x y z x 2 1 xx 1 x + xx + x y z 2 x 2 p xx = + x x xx
x y z ,
(1.7)
yx
yx 1 yx 1 yx y x z + yx + y x z = x y z, (1.8) y 2 y 2 y
and on (T,B):
zx
zx 1 zx 1 zx z x y + zx + z x y = x y z. (1.9) z 2 z 2 z
Putting Eq. 1.7, Eq. 1.8 and Eq. 1.9 together we obtain the x-component of the momentum equation:
(u) + t
uU =
(p + xx ) yx zx + + + fbx , x y z
(1.10)
(v) + t
and z-direction:
v U =
xy (p + yy ) zy + + + fby , x y z
(1.11)
(w) + t
wU =
xz yz (p + zz ) + + + fbz . x y z
(1.12)
Basic CFD
Chapter 1.
DE =Q+W . Dt
(1.13)
Let's determine the rate of heat addition and the work done on the particle in terms of more suitable physical quantities such as temperature gradients, shear and normal stresses.
Referring to Fig. 1.3, the net rate of heat transfer due to heat ow in the x-direction is:
qx +
qx 1 qx 1 x qx x x 2 x 2
y z =
qx x y z , x
(1.14)
and including contributions for the y and z directions, the rate of heat transfer added to the particle per unit volume is:
qx qy qz = x y z
q .
(1.15)
q = k T,
giving nally:
(1.16)
Q=
(k T ) x y z .
(1.17)
Chapter 1.
Basic CFD
The net rate of work done on the particle by surface forces acting on xdirection is:
(1.18)
Following the same approach for the y and z direction as in Eq. 1.18 and summing, the total rate of work done on the uid particle by surface stress is:
W =
(pU ) +
ij Ui xj
x y z .
(1.19)
Including Eq. 1.19 and Eq. 1.17 into Eq. 1.13, the nal form of the energy equation in obtained, eventually adding the rate of work done by body forces as a source term SE :
E + t
E U =
pU +
ij Ui + xj
(k T ) + SE .
(1.20)
For compressible ows it is however more correct, and surely less complicated to treat, to express energy equation in terms of another variable: total enthalpy in which pressure energy is directly included is one of the choices:
h0 = E +
p 1 2 p =i+ u + v 2 + w2 + . 2
(1.21)
Substituting Eq. 1.21 into Eq. 1.20, one may obtain energy equation in terms of total enthalpy:
h0 + t
h0 U =
ij Ui xj
p ij Ui + + t xj
(k T ) + Sh ,
(1.22)
Basic CFD
Chapter 1.
First hypothesis is the assumption of the uid to be a perfect gas, meaning that it follows the well-known equation of state:
p = RT.
(1.23)
Second assumption concerns the modeling of the viscous stress. Fluids of interest are considered Newtonian, meaning that viscous stresses, in analogy with hookean elasticity, are linearly proportional to the rates of deformation. A rst coecient named dynamic viscosity, relates stress to linear deformation, a second coecient called second viscosity or bulk viscosity takes into account volumetric deformation eect:
ij =
Ui Uj + xj xi
+ ij
U .
(1.24)
It may noticed that, following such denition, mechanical and thermodynamic pressure may not coincide. In fact following this denition ij is not implicitly traceless. Since little is known about it is common to avoid this inconsistency simply assuming that = 2 . It is also true that the term U is usually 3 very small and some authors proposed to directly neglect the eects due to volumetric deformation (White, 1991). Third comes the modeling of thermophysical properties. Even if the validation runs have been performed with constant thermophysical properties, uid properties are treated, not to loose generality for the proposed model, as function of both temperature and pressure. All these assumptions and simplications lead to the nal set of governing equations:
= 0, = Ui Uj 2 Uj + ij xj xj xi 3 xj p T = + k + + Sh , t xj xj = RT , +
(1.25) (1.26)
+ Fbi ,
(1.27) (1.28)
= ij
1 Ui = xj 2
Uj Ui + xi xj
2 3
(1.29)
Chapter 1.
Basic CFD
approach is to convert a partial dierential equation, valid on the entire domain, into a set of discrete algebraic equation, one for every node considered. The value at the node is, of course, put in relation with neighbor nodes, the simultaneous satisfaction of all the equation in the set then give the numerical solution. The most popular discretization techniques are presented and discussed:
Finite Dierence Method (FDM) approximates conservation equations in dierential form substituting partial derivatives via truncated Taylor series expansions or polynomial tting. Even if, in principle, it can be applied to all kind of grids, actual applications are limited to structured grids where grid lines are used as local coordinate lines. In such a way in fact, it is easy to obtain higher-order schemes. The biggest drawback of FDM is that it does not enforce conservation, consequently it is very hard to get reliable simulations of complex geometries and use is restricted to the simple ones. Finite Volume Method (FVM) works with the integral form of the conservation equations. The domain is divided into nite volumes whose centroid represent the calculation node. Grid just dene boundaries in between dierent volumes and need not to be related to any metrics. Interpolation is used to express variable values at the surface in terms of nodal values. This method is conservative by construction as long as surface integrals for volumes sharing the same face are equal for both of them. The disadvantage of FVM in comparison with FDM is that building higher than second order schemes for 3D simulation is much more dicult due to the three levels of approximation introduced: interpolation, dierentiation and integration. Because of its very physical approach, ease to be understood and implemented, FVM is the most widely used approach. Finite Element Method (FEM) may appear similar to FVM, the distinguish feature is the weight function: conservation equations are multiplied by a weight function before being integrated. Solution is supposed to adhere within each volume to a shape function constructed from values at the corners of the elements. Such an hypothesis is substituted into the conservation equations whose derivative with respect to nodal value is put to zero, selecting so the residual minimizing allowed function. The main advantage in using FEM is the ability in dealing with arbitrary geometries, while the main drawback, common to all integral methods, is that the resulting matrix may result not well structured meaning that ecient solving method are dicult to implement.
Discussing which of the previously described methods should be used, exact satisfaction of scalar conservation over control volumes has been focused to be the most important properties to search for. It is fundamental in fact that all
10
Basic CFD
Chapter 1.
the simplications and discretizations to be introduced in the governing equations, always respect exactly basilar physical principles they are built on. The physical laws to be obeyed are all conservation laws so conservation should be perfectly observed. Exact conservation can be imposed externally or intrinsically respected by the discretization method. The second choice seemed the most logic and that is why implicitly conservative methods were selected. Finite Volume Method has so been chosen, as a consequence from now on it will be the method of reference and will be discussed more in detail, see Sec. 1.3.1.
J ndS =
S S
(U n)dS
S Convection
( n)dS =
Dif f usion
q d .
Source
(1.30)
Such an equation applies over each control volume and the entire domain as well, underlying once again the main feature of FVM: global conservation. To obtain an algebraic equation the three integral must be approximated by quadrature formulae. In the following discussion about approximation techniques the bi-dimensional quadrilateral CV in Fig. 1.4 has been considered.
11
Chapter 1.
Basic CFD
QP =
qd = q .
(1.31)
The rst and easiest approximation just replaces the mean value with the value at the CV center node
Q P = qP .
(1.32)
Such approximation is obviously exact if q is constant or varies linearly, in any other case is second order accurate. Higher order of accuracy schemes can be obtained via shape function or interpolation.
J ndS =
S S
Jn dS =
k Sk
jdS =
k
k .
(1.33)
To exactly calculate surface integrals, the value of j is needed everywhere on the face Sk . This is of course impossible as long as only nodal values of are computed, it is so necessary to introduce some approximations. Usually two level of approximation are introduced:
the integral is approximated in terms of the variable values at one or more locations on the cell face (face approximation ) the cell-face values are approximated in terms of nodal values (nodal approximation ).
For each type several approximation modes have been proposed. For the face approximation:
midpoint rule - the value over the face is equal to the value at cell-face center e = je Se . This approximation is of second-order accuracy. trapezoid rule - the value on the face is the mean between face extremes 1 (cell vertex) shared with 'neighbor' cells e = Se 2 (jne + jse ). This approximation is second order too. Simpson's rule - the value on the face is a combination of the value at the 1 center of the face and the values at the vertices e = Se 6 (jne + 4je + jse ). This approximation is forth order.
Of course to maintain such levels of accuracy, nodal approximation of, at least, the same order must be chosen. Extension in three dimensions is quite direct for both the midpoint and trapezoid rule, but gets more complicated for
12
Basic CFD
Chapter 1.
higher-order approximation. Nodal approximation is discussed later Sec. 1.4 being a task requiring deeper treatment. It has already been cleared that J is composed by two dierent contributions: convection and diusion. These two terms indeed behave in a quite dierent manner as will be shown in the following.
( n)dS =
S k
Sk ( )k =
k
Sk ( )k ( )k .
(1.34)
Let's see how it is not that dicult to compute gradients on the boundary face e: E P . (1.35) ( )e = xE xP It is however possible to compute the gradient in other way that is:
( )P =
Sk k ,
k
(1.36)
and nd the corresponding value on the face e interpolating ( )P and ( )E with techniques to be proposed later in Sec. 1.4.
(U n)dS =
S k
Sk (Un )k =
k
Sk (Un )k k =
k
Fk k . (1.37)
Transport of across the boundary faces has been computed in quite a direct way, the problem has now shifted in nding an expression for k involving only nodal values. This pretty much is the most crucial passage of the entire discretization process and so the following section is entirely dedicated to interpolation and dierentiation schemes.
13
Chapter 1.
Basic CFD
e = P + (xe xp )
+
P
(xe xp )2 2
2 x2
+H .
P
(1.39)
UDS neglect all terms but the rst so it is a rst order scheme. The rst neglected term behave like a diusive ux:
d je = e
,
e
(1.40)
meaning that scalars are diused normally and parallel to the ow. The coecient of numerical diusion is proportional to grid dimension and to mass ux, moreover is magnied for multidimensional oblique to the grid ows. To avoid such inaccuracy, fundamental in case of shocks or rapid changes, mesh must be rene a lot because the scheme is only rst order accurate.
e = E e + P (1 e ) ,
(1.41)
with e = (xe xP )/(xE xP ). Taylor series expansion for E at point xP shows that Eq. 1.41 is second order accurate vanishing all rst order terms in the truncation error. In fact:
e = E e + P (1 e )
(xe xp )(xE xp ) 2
2 x2
+H .
P
(1.42)
Having eliminated the numerical diusion, the leading term of the truncation error is quadratic and may produce oscillatory solutions. In spite of such oscillatory behavior CDS, being the simplest second order, is one the most widely used dierencing scheme.
14
Basic CFD
Chapter 1.
SF CD = e CDS + (1 e )U DS . e e e
(1.43)
The blending factor will be as close as possible to unity in case of steep gradients in order to get sharp discontinuities, while in other more uniform zones stability is reached with close to zero. The main drawback of this scheme is the introduction of additional non linearity due to the dependency of on values, resulting for some steady ow in oscillatory solutions (Sta, 2004).
AP P +
nb
Anb nb = QP .
(1.44)
If both the AP and the Anb are known, this is the case if the ow eld is given, the equation is linear and putting together the equation for each node it is possible to built up a matrix representing the dierential equation over the entire domain:
A=Q,
(1.45)
in which matrix A is in general quite sparse, because inuence for P is usually limited to a small number of other points, actually for FVM is strictly limited to the number of neighbor nodes. is the solution vector and Q is the source term both of dimension equal to the number of cells.
1.5 Grid
As already mentioned, the solution domain should also be discretized into a nite number of control volumes. The discrete representation of the spatial domain is called grid or mesh. Meshes can rstly be subdivided into:
Structured mesh: uid domain of interest is covered with three families of lines (, , ) never intersecting other lines of the same family and normally intersecting all lines of the other families in sequence. Nodal point Pi,j,k is dened by the intersection of the three lines (i , j , k ). In this grid arrangement it is always possible to identify neighbor nodes moving on the three lines intersecting at the nodal point.
15
Chapter 1.
Basic CFD
Unstructured mesh: uid domain of interest is covered with a number of nite volumes Pi of arbitrary shape sharing the boundaries. Contrarily to the structured grid where exibility in gridding complex geometries is limited by the sti organization of mesh elements, unstructured mesh oers the maximum freedom in dening cells both in terms of shape and location.
A comparison between a typical example of a structured and an unstructured mesh can be done looking at Fig. 1.5 and Fig. 1.6. On the contrary of FDM, FVM can also deal with unstructured grid. This is fundamental for heat transfer involved simulations where typically non-aerodynamic geometries are encountered. The main dierence between FDM and FVM however is that in FVM grids CV are dened via cell boundaries more than cell centers, even if this could be done. Once the boundary surfaces are dened the coordinate of cell center are calculated assuring a higher degree of accuracy in substituting the mean with the center value. Viceversa if centers are dened and boundary calculated, Central Dierence Scheme for derivatives are more accurate at CV faces because the face is midway between the two nodes.
A new class of methods, the so called meshless methods, has been proposed recently to avoid handling meshes that sometimes can result in being the most expensive step in the entire solution process. The idea is not to create a mesh but just to store data for the cell centers, actually it is a no-sense to talk about cells when there are no more, so they will be referred to as nodes. An area of inuence is hence dened for each node generating a local mesh. Local control volumes can overlap without loosing conservative properties: conservation is respected no matter the shape, dimension and orientation of the control volumes.
16
Basic CFD
Chapter 1.
however of secondary importance to underline that the ow eld cannot be known a priori and scalar quantities for which conservation equations were derived (, Ui , h0 , p) are responsible for the development of the ow eld itself. So the equations are not independent one from the other but on the contrary dierent scalar conservation laws result to be tightly coupled. In other words the Navier-Stokes Equations cannot be solved separately but must be treated like a real system meaning that in the conservation equation for scalar 1 , all other s cannot be considered known but should be treated as variables too. This results in a strong non-linearity of the system, think about convective or viscous dissipation terms for example. The mechanism used to eliminate such non-linearity is a guess and correct approach leading to an iterative cycle at the end of which the solution will satisfy all the equations at the same time. Starting from an initial guess of all s, the system is solved taking the values for i as constant in the equation for j . This solution is used to update coecients of the matrix representing the system and the cycle repeats until solution is reached. As the reader may easily believe it is not so straight forward that such a method really leads to convergence. Many dierent algorithms can be used to improve convergence properties for this cycle usually called outer cycle, most of which separate the system in equations to be solved sequentially and not contemporary. These are called segregated solvers, in contrast with simultaneous solvers, and will be analyzed in detail later in Sec.3.1. Once the Navier-Stokes Equations are linearized, again they appear in the nal form:
A=Q
(1.46)
where A is, this time, the sparse coecient matrix for all the ve conservation laws and the equation of state, is the solution vector for all variables and Q is
17
Chapter 1.
Basic CFD
the source vector. Linear systems are not dicult to solve, think about Gauss elimination techniques or Cramer rule, but, since typically the number of cells for a CFD simulation varies from some thousands to hundreads of millions and CPU time is a very requested and still limited resource, solving in an ecient and fast way it is a must. Methods for solving linear systems are commonly classied into two main groups: direct methods and iterative methods. Direct methods handle the matrix with linear algebra rules to reduce it in a form from which extraction of nal solution can be done quite easily. Every well posed linear system can be solved directly but the sparse matrix A (every node depends on the value of not so many other nodes) gives decomposition matrices that are not sparse increasing the computational cost considerably. Iterative methods instead starts from a guessed solution and use the system to successively rene it. If each iteration is cheap and the number of iteration is small enough, this is usually the case in CFD, iterative methods result in being much faster than direct ones. This iterative cycle is called inner cycle. The algorithms used in this thesis are the Incomplete Cholesky preconditioned Conjugate Gradient (ICCG), Incomplete Cholesky preconditioned Biconjugate Gradient (BICCG) and the Algebraic Multi Grid (AMG). For details on such methods see Ferziger and Peric (2002); Lacor (2006); Murthy and Mathur (2002). To avoid instability due to the non-linearity of the system as a consequence of the coupling between the equations, underelaxation techniques are widely used (Patankar, 1980). The concept standing behind underrelaxation is to relax the changes in variable values multiplying the increments from the previous value by an underrelaxation factor (more commonly referred to as relaxation factor) with values in the interval [0, 1]:
n = n1 + (n n1 ) ,
(1.47)
where n on the right hand side in the exact result of the linear system at outer iteration n while n on the left hand will the best available value for at iteration n. Eq. 1.47 is sometimes referred to as explicit relaxation, meaning that matrix A is not modied. It is however possible to express the same relation in an implicit way involving a modication of the matrix itself. Considering the linearized equation for P : P AP n +
nb
Anb n = QP , nb
(1.48)
and using the Eq. 1.47 it is possible to write a new expression for P at iteration n:
n = n1 + P P
QP AP n + P
nb
Anb n nb
AP
nb
n1 P
1 AP n1 . (1.49) P
Anb n = QP + nb
18
Basic CFD
Chapter 1.
This technique is known as implicit relaxation. Stability is increased because P the diagonal dominance of the matrix A is higher: A n > AP n . UnfortuP P nately optimum underrelaxation factors are problem dependant, moving towards lower values to increase stability and to higher values to speed up convergence. A general strategy could be to start with low relaxation factors for early iteration and move to higher values reaching convergence. It is now time to clarify what really convergence is and how it can be established. Using iterative methods, the most important parameter to check how exactly a prediction satisfy the equation is the residual dened as: Res = Q A . Res is a vector, with cell number size, and it is not a very convenient factor to establish whether or not an equation is converged. The global parameter usually used is: GRes =
n
Res =
n
(Q A ) ,
(1.50)
where n correspond to the number of cells in uid domain. Equations are checked separately so 5 dierent residuals plus the ones for turbulence will be followed. Usually, and this is the case, residuals are normalized with respect to some mean value, trying to account relative errors more than absolute. In this work the normalization factor is dened as follows:
ref S Sref
Norm
As already hinted, two nested iterative cycles are present: the inner one is directly related to the solution of linear system at iteration n, the outer one regards the algorithms of segregated solvers going from the initial guess at step 0 to actual iteration at time n. Convergence criteria must be set up and controlled for both inner and outer iteration cycles. Outer iterations control parameter is Initial Residuals, while inner iterations are managed by Final Residual: InitialResidual FinalResidual
= =
n (Qi0
19
Chapter 2
OpenFOAM
The aim of this chapter is to introduce the reader into the main features of OpenFOAM as support code for CFD simulations. In particular focus will be posed more on the way it is born and developed than on its practical using. OpenFOAM in fact is not thought to be a ready-to-use software, even if it can be used as a standard simulation package, but to oer a backing to CFD programmers in building their own codes. Being at the core just a C++ library, rst a brief introduction to such programming language is necessary. Then OpenFOAM itself is described in detail giving both a philosophical introduction and practical examples on how to work with its classes and its applications.
Introduction to C++ . . . . . . . OpenFOAM software . . . . . . . Matrix structure in OpenFOAM Working with main classes in OF
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
22 23 25 26
21
Chapter 2.
OpenFOAM
Procedural Programming: the focus is on the processing, i.e. the algorithm needed to perform the desired computation. Languages provides facilities for passing and returning the most suitable arguments to the most suitable functions in the most suitable way. Discussions regard the concept of most suitable. Modular Programming: also known as the data-hiding principle, is concerned with subdiving programs into smaller modules data are hidden within. This need reects a constant increase in program size. Data Abstraction: this paradigm is dealing with user-dened data types. Data types suitable for the problem are newly dened providing a full set of operations for each type. Connected and fundamental for doing it, are the concepts of operator overloading and polymorphism, also called late binding. Object-Oriented Programming: the concept introduced with such a paradigm is the concept of inheritance in connection with class hierarchy. Base classes are dened as the common structure and then specialized into derived classes. This can actually be done even with previous paradigm but now the possibility of sharing (inherit) members from base to derived class is introduced. Generic Programming: the aim is to parametrize algorithm in such a way they can work for a variety of suitable types and data structures. Generic Programming introduces the concept of containers: classes that can hold a collection of elements of dierent types.
C++ was designed to support data abstraction, object-oriented and generic programming in addition to traditional C programming techniques such as procedural and modular programming. It was not meant to force one particular programming style upon all users (Stourstroup, 1997). It would be better to clarify some of the above introduced concepts, in particular: polymorphism, inheritance and containers. Connected with data abstraction paradigm is the concept of operator overloading meaning that the same function can actually have dierent meaning for
22
OpenFOAM
Chapter 2.
dierent classes. Polymorphism is the capability of deciding at run-time which of the several overloaded member functions to call. Inheritance is the ability of one class to inherit capabilities or properties from another class. To x ideas an example is given. Consider the class of the Human-beings, it has some specic properties (two arms walk, intelligence, etc) plus some others (giving birth to live young, warm blood, etc) derived from the class Mammal which is belonging to. Equally Mammals have inherited some of their properties to other base classes such as Vertebrate or Animals. In such a way a sort of chain is built to correctly link a property to the right class. Thinking about containers the best example is a list. Lists are basically a collection of elements stored in a row. It is possible to dene a list of integer as well as a list of people, meaning that the class denition is not bound by the type of objects it is listing. It is common to refer to containers as templates. Templates are a compile-time mechanism so that their use result in no run-time overhead compared to hand written code (Davis, 2005).
fvm::ddt =
23
Chapter 2.
OpenFOAM
fvm::d2dt2 = fvm::div =
2 t2
, ,
2 i x2 i
i xi
fvm::laplacian =
In addition implicit treatment of source terms is done by: fvm:Sp and fvm:SuSp. Explicit equivalent for the previous operators are dened and furthermore other common operators such as curl, gradient, etc, are implemented. Building dierent types of partial dierential equations is now only a matter of combining in a dierent way the same set of basic dierential operators. To give an example of the capability of such a top-level code, let's consider a standard equation like momentum conservation:
U + t
U U
U = p.
(2.1)
It can be implemented in an astonishingly almost natural language which is ready to compile source C++ code:
24
OpenFOAM
Chapter 2.
(a)
(b)
In addition, even if never directly tested by the author, automatic mesh motion and runtime topological changes such as attach/detach boundaries, cell layer addition/removal are allowed. OpenFOAM was thought not really to be only a ready to use code but to be as much exible as possible in dening new models and solvers in the simplest way. Its strength in fact 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. Moreover, the possibility of using top-level libraries containing a set of models for the same purpose which refer to the same interface, guarantees programmers for smooth and ecient integration with the built-in functionality. New models can be added to the appropriate model table at link-time and become available in the same manner as the supplied models. Most of the selections necessary to set up calculations are done at runtime, meaning that options can change while the code is running. For further information about how to use and how to program OpenFOAM see OpenCFD (2006a,b).
25
Chapter 2.
OpenFOAM
clarify consider line i of array A, it corresponds to equation for cell of index i. The number of non-zero coecients on this line equals the number of faces plus the diagonal coecient that multiply the solution at cell i itself. On the left side of diagonal coecient (rows j from 1 to i-1), coecients for faces between cell i and lower index are stored. l array is actually storing coecients for owner cells of faces bounding cell i not owned by cell i itself. At the contrary, coecients for cells who share boundaries owned by cell i, once again the cells with index higher than i, are stored in the u array. OpenFOAM programmers usually works with FVM, so a special class of matrix fvMatrix in which the inuence is only limited to adjacent cells is dened and normally used.
To interpret these graphs one should read it in this way: a box represents a class, a solid arrow stands for public inheritance and a dashed arrow indicates usage, with the edge of the arrow labeled with the variable responsible for the relationship. Meaning that GeometricField has a boundaryField and is a Field<Type>. As hinted above at, GeometricField could represent any tensor eld; it consists of an internalField and a boundaryField. The rst one, belonging to Field<Type> class, stores a list of values of appropriate tensor rank for each computational point, the second, namely a GeometricBoundaryField, stores boundary values subdiving it into one patchField for each boundary surface. Reference to dimensions is stored in the dimensionSet that basically is the vector of the exponents for the fundamental units. It could be of 5 [kg, m, s, K, mol] or 7 [kg, m, s, K, mol, A, cd] elements.
26
OpenFOAM
Chapter 2.
GeometricField is a template and gives a good example of generic programming: this virtual class in fact contains Fields of scalar, vector and tensor type dened on surfaces and volumes. One can in fact dene objects like a surfaceScalarField as well as a volTensorField. This completes introduction to one of the most common classes in OpenFOAM has been done.
27
Chapter 3
Contents 3.1 3.2 3.3 3.4 3.5 3.6 3.7 3.8 3.9
Segregated solvers . . . . . . . . . . . . . . . . . . Pressure velocity coupling . . . . . . . . . . . . . Problems associated with density based solvers SIMPLE Algorithm . . . . . . . . . . . . . . . . . Pressure checkerboarding . . . . . . . . . . . . . Co-located grid: Rhie - Chow interpolation . . . Compressibility eects . . . . . . . . . . . . . . . Critical aspects . . . . . . . . . . . . . . . . . . . . Solver validation . . . . . . . . . . . . . . . . . . .
3.9.1 3.9.2
. . . . . . . . .
30 30 31 32 36 37 39 40 42
42 43
29
Chapter 3.
(3.1)
where [mn] is a vector of dimensions equal to the number of cells n times the number of unknowns m. For computational costs reasons, this system is usually not solved entirely but sequentially, in other words the system matrix is decomposed into one smaller matrix for each equation and such matrices are solved in sequence for the scalar of reference:
A[n,n] [n] Au[n,n] u[n] Av[n,n] v[n] Aw[n,n] w[n] Ap[n,n] p[n] Ah[n,n] h[n]
= = = = = =
In this way it is clear that the coupling between the dierent equations must be treated with another strategy. An algorithm to iteratively link such discretized equations is needed in order to reach convergence. It is easy to guess that a number of dierent algorithm has been proposed to organize the steps in which the process should be parted.
30
Chapter 3.
regimes. In particular, it usually happens that dierent Mach conditions simultaneously arise in the same domain. Such situations makes the accurate solution of viscous ows governing equations a complex task since this kind of ows does not fall completely into any of the categories cited before. In order to solve such ows specic solvers should be created trying to enlarge range of applicability to the maximum, at the best to All-Mach ows. The most famous method proposed to change behavior of the incompressible Navier-Stokes Equations and t them over the compressible shape is the socalled articial compressibility technique: a fake time derivative is added to the incompressible conservation equations rendering them hyperbolic and solvable with standard time-marching techniques developed for the compressible form of the equations. Such methods are applied to steady state solution and require a dierent set of variables in the compressible and incompressible ow regime. That is why a dierent path to solution has been searched for.
+( t
J) = Q ,
(3.8)
J=
U
convective
dif f usive
(3.9)
In addition an equation of state to describe the relationship between the various thermodynamic variables is needed. This set of non-linear coupled equation should be solved for the unknowns , U , p, h. It is so customary to associate each of the equations to one basic variable: density for continuity, velocity components for momentum, enthalpy for energy. In addition the equation of state can be considered as the equation for pressure. At very low Mach number however, the pressure and the density become very weakly related and in the idealized limit of incompressible ows, the density is completely decoupled from the pressure. For such a reason neither pressure nor density can be directly associated with the incompressible continuity equation and it assumes the role of a compatibility condition on the velocity eld. For a sequential solution of the equations a mechanism to couple continuity and momentum should be searched, otherwise it is also possible to use the fully compressible equations even in incompressible ows. Indeed all uids are compressible and the pressure is always related to the density even though the relationship may become very weak. However, because equations are solved numerically, constraint are imposed by numerical considerations: truncation and round-o errors.
31
Chapter 3.
The perfect gas law for a nearly incompressible ow in which the speed of sound can be considered constant, states that:
p = RT =
c2 c2 p = .
(3.10)
p 1 = . 0 u2 M a2 0 0 0
(3.11)
This expression clearly points out, that even though changes in density may become small or innitesimal, as long as Ma number tends to zero too their ratio is nite and changes in pressure may remain large. If sensibility in solving both terms is established to be:
p 0 u2 0 0
(3.12) (3.13)
then it is imposed the Mach number to be grater than 10 2 . In addition for M a = 0 there is a singularity point for the Navier-Stokes Equations, rendering them improperly scaled in the limit of incompressibility. The above discussion highlights the problems associated with the use of density as a primary variable for computing low Mach ows or mixed compressible and incompressible ows (Karki, 1986b). Development of a computational scheme valid for the whole range of Mach number, needs a switch to pressure as a primary independent variable in preference to density.
32
Chapter 3.
ae ue a n vn
= =
(3.14) (3.15)
To initialize the SIMPLE calculation a pressure eld p is guessed, hence Eq. 3.14 and Eq. 3.15 are solved with such pressure distribution to obtain a rst guess value for velocity u and v :
ae u e
a n vn
= =
anb u + (p p )y + Qe , nb P E
anb vnb + (p p )x + Qn . S N
(3.16) (3.17)
A pressure corrector is then dened as the dierence between the correct and the guessed pressure, and similarly is done for velocity correctors too:
p u v
= = =
p p , uu , vv .
Subtraction of Eq. 3.16 from Eq. 3.14 and of Eq. 3.17 from Eq. 3.15 gives two equations for pressure and velocity correctors:
ae ue a n vn
= =
(3.21) (3.22)
Given that correctors tend towards zero as long as solution is converging, dropping terms such as anb unb and anb vnb is not giving any error when convergence is reached. This is however the biggest approximation of the SIMPLE algorithm and its consequences will be discussed later in this section. Rewriting Eq. 3.21 and Eq. 3.22, one obtains:
ue vn
= =
(3.23) (3.24)
33
Chapter 3.
where de = y and dn = x . Thus far only momentum equations has been ae an considered, continuity should be used to impose mass conservation:
(3.26)
Eq. 3.23 and Eq. 3.24 can be inserted in Eq. 3.26 to give:
Fe + e de y (pP pE ) Fw + w dw y (pP pW ) + Fn + n dn x (pP pN ) Fs + s ds x (pP pS ) = 0 , (3.27)
AP pP =
nb
anb pnb + b ,
(3.28)
where:
aE aW aN aS AP
= = = = =
e de y , w dw y , n dn x , s ds x , anb ,
nb Fw Fe + Fs Fn .
b =
At this point all the necessary equations have been deduced: they should be included into an organic iterative process to obtain a solution. The overall algorithm foresees the following steps: 1. Guess a pressure eld p . 2. Discretize and solve the momentum equations using the guessed value p for the pressure terms. Eq. 3.16 and Eq. 3.17 yield the u and v . 3. Find the pressure correction source term b calculating mass imbalance with Eq. 3.34. 4. Discretize and solve the pressure correction equation: Eq. 3.28. 5. Calculate velocity corrections using Eq. 3.23 and Eq. 3.24. 6. Correct pressure and velocity elds: Eq. 3.18, Eq. 3.19 and Eq. 3.20.
34
Chapter 3.
7. Solve discrete equations for each scalar using the continuity satisfying velocity eld for the convection term. 8. Check convergence: IF satised stop, ELSE use p as a new guess p and start again from 2. It is now time to return to Eq. 3.21 to clarify the consequences of dropping the nb anb . It has already been underlined how at convergence nothing is changing, so the impact is limited on the rate of convergence. Such dropping in fact places the entire burden of correcting velocities upon the pressure corrector. This will lead to a velocity eld still satisfying continuity but also to an overpredicted pressure eld. The pressure correction equation is indeed susceptible to divergence unless some under-relaxation is used during the iterative process. Implicit underrelaxation is usually preferred because it also increase the diagonal dominance of the matrix, but this cannot be case. Being a corrector the history of convergence of p is without any particular meaning, in other words values of p at previous iteration do not resolve the same, or almost the same, equation, as a consequence p n1 can be very dierent from p n and implicit relaxation will violate continuity. What is done to avoid this obstacle is to relax the new values for pressure and not the corrector:
pnew = p + p p .
(3.35)
To avoid overpredictions of pressure, velocities are also relaxed. A correct choice of under-relaxation factors is essential for cost-eective simulations. Too large a value may lead to oscillatory or even divergent solution while small values comport extremely slow convergence. Unfortunately, the optimum values of under-relaxation factors are ow dependent and must be sought on a caseby-case basis (Malalasekera and Versteeg, 1995; Murthy and Mathur, 2002). Other variants of the same method have been proposed in the course of years:
SIMPLER (SIMPLE-Revised): use a discretized equation for pressure to obtain p from an initial guess of velocity v and pressure p elds (Patankar, 1980). SIMPLEC (SIMPLE-Consistent): use the same algorithm of SIMPLE but with a better approximation for Eq. 3.21 and Eq. 3.22 (Doormal and Raithby, 1984). PISO (Pressure Implicit with Splitting Operators): add a second corrector step to the standard SIMPLE algorithm meaning that p = p + p is only a starting guess for the second corrector step (Issa, 1986).
At this point the method is completely described, it is not however secondary to better precise how to treat diculties arising with the use of such a solver. In particular, more attention must be paid on the problem of checkerboarding solutions and extension to fully compressible ows.
35
Chapter 3.
Fe =
FP + FE . 2
(3.36)
where the interpolated ux Fe is referred to the interpolated velocity Unfortunately this approach leads to inaccurate results, in fact it allows for non physical checkerboarding solutions. Checkerboarding proles are due to the fact that linear interpolation brings continuity, i.e. Eq. 3.25, to be independent from the value of velocity on the cell of reference, in other words conservation of mass at cell P depends only on neighboring values, as a consequence odd and even nodal values result to be decoupled:
Fi = Fe Fw + Fn Fs =
1 (FE FW + FN FS ) . 2
(3.37)
If such a pattern is satisfying momentum equation too, the checkerboarding would persist in the nal solution; unfortunately this is the case. To demonstrate it, take the x-direction momentum equation Eq. 3.14 and make a balance over cell P:
uP
= dP (pw pe ) + ... pW + pP pE + pP = dP 2 2
(3.38)
+ ... = dP
pW pE 2
+ ... . (3.39)
The dependency of uP on pP , using linear interpolation, is canceled and velocity is only feeling neighboring pressure values. This means that velocity and pressure at the same location do not inuence each other, as a consequence checkerboarding patterns of velocity and pressure will satisfy momentum equation.
36
Chapter 3.
Dramatically both momentum and mass conservation equation are solved by the checkerboarding pattern and so once it is formed it will never smooth. Even if dicultly perfect checkerboarding proles for velocity or pressure arise, due to irregularities in the mesh, boundary conditions or physical properties, such tendency often shows itself in unphysical wiggly elds. To avoid this unacceptable behavior two methods are proposed in literature:
use a staggered grid: uid domain discretization on which velocities and pressures are computed at dierent locations, use a co-located grid but a dierent interpolation method for face mass uxes.
First approach takes advantage of storing values for velocity and pressure precisely at the points required for the implementation and no interpolation is needed. Primary drawback for the staggered grid arrangement is the increment of metric complexity due to the use of dierent grid for dierent variables. This method results as a consequence to be very expensive computationally speaking for curvilinear coordinates and inapplicable to unstructured mesh, so in practice it has been abandoned in the course of years. Second approach prevents checkerboarding by expressing face velocities in terms of adjacent cell pressure rather than alternate values. Even if more complex to understand, this approach is less dependent on geometries and gridding strategies and moreover it is the only chance to continue using a co-located code. Advanced interpolation techniques are so computed in order to avoid such checkerboarding on co-located grids.
u0 = uP + P
dl p A
= uP +
P
(pE pW ) , 2AP
(3.40)
u0 = e
u0 + u0 P E , 2
(3.41)
ue = u0 e
dl p A
= u0 e
e
(pE pP ) . 2Ae
(3.42)
37
Chapter 3.
Substituting Eq. 3.40 into Eq. 3.41 and into Eq. 3.42, a nal expression for face velocity is obtained:
ue =
(uP + uE ) 1 + 2 2
(pE pP ) . 2Ae
(3.43)
Similar expressions may be derived for the other faces. Putting together such expressions and summing mass uxes Fi , it is clear that explicit dependence on cell values do not cancel out as with linear interpolation. Continuity and momentum equation now do feel pressure at point P meaning that wavy gradients are smoothed and so checkerboarding is avoided. It is interesting to see how the behavior of continuity equation changes with the introduction of this interpolation also known as added dissipation scheme. Rewrite Eq. 3.43 considering Ae = AE = AP , this happens in case of uniform grid and uniform ow:
ue =
Writing down the Taylor series expansion for pressure and reorganizing:
2p x2
so
=
P
(3.45)
ue
= =
2p x2
x2 =
E
x3 .
(3.46)
ue uw
= =
3p x3 4p x4
x3 +
e 4
1 4AP
3p x3
x3 =
w
x = 0 ,
P
(3.47)
it comes out that a fourth order derivative is added to continuity. Derivative terms of even order are known in literature as dissipation terms, as a consequence a more diusive behavior is expected from this equation (Anderson, jr., 1995). More diusion means more inuence of the center value on its direct neighbors sweeping away all unphysical wiggles. With the implementation of such interpolation scheme the problem of checkerboarding patterns is solved and the same overall algorithm can be applied also in this case of co-located variables Davidson (2005); Murthy and Mathur (2002); Mangani (2006). Before going through the derivation of a dierent pressure correction equation valid for compressible uid ows, it is better to specify how to use values of pressure and velocity known at step 6 of the SIMPLE algorithm, for solving conservation equations for the other quantities. Solution of passive scalars are
38
Chapter 3.
better computed via the use of face ux, on which continuity is respected, more than cell velocity, on which only momentum is exactly satised. If this advice is followed in fact, conservation equations are perfectly conservative.
= + .
Using a compressibility factor dened as:
(3.48)
1 = , p RT
(3.49)
it is possible to write:
= p .
(3.50)
39
Chapter 3.
As a consequence:
Fe = (u)e Ae = ( + )e (u + u )e Ae = ( u )e Ae + ue Ae + e u Ae + e ue Ae , (3.51) e e
that, dropping last term because innitesimal of second order and substituting with Eq. 3.50 (upwinding is used in the following), gives:
Fe = Fe + ue Ae + pP u Ae . e e
(3.52)
Repeating the same reasoning for all faces a dierent pressure correction equation is obtained:
Fe + de y (pP pE ) + pP u y e e Fw + dw y (pP pW ) pW u y w w + Fn + dn x (pP pN ) + pP u x n n Fs + ds x (pP pS ) pS u x = 0 . (3.53) s s
AP pP =
nb
anb pnb + b ,
(3.54)
where:
aE aW aN aS AP
= = = = =
de y, e dw y w + u y, w dn x, n ds x + s anb
nb Fw
u x, s uw y
u x s
u y e
u x, n
(3.59) (3.60)
b =
Fe + Fs Fn .
Eq. 3.54 include two dierent types of terms for the pressure correction pP : rst type involving velocity, a convective term pP u y , and second involving e pressure dierence, a diusion like term de y (pP pE ). The relative impore tance of the two terms depends on the Mach number of the ow. At low Mach numbers, the diusional part dominates and the equation exhibits an elliptic behavior. In the supersonic regime, the convective term is much larger than the diusional term and the mass ux is governed solely by the upstream pressure. This reects the correct hyperbolic behavior of pressure for supersonic ows. The nature of the equation is such that the transition from the subsonic to supersonic regime is gentle and the transonic ow calculations are smoothly treated.
40
Chapter 3.
for the pressure corrector and the underrelaxation of the pressure corrector equation. The starting problem was to understand which was the more suitable boundary condition for the pressure corrector. Being a ctitious variable at convergence equally zero, every boundary condition satisfying this constraint could be considered physical for the problem. However it is true that on boundaries on which the pressure is specied there is no need for a correction (p = p ) and so p is xed to zero. At the same time boundaries on which the mass ow is specied need no mass correction that means no dierence of p across that boundary: zero Neumann condition. Information about boundary conditions is propagated into the domain along characteristic lines. Characteristics for compressible uid ow are the Mach waves. Theory of characteristics reveals how, at the inlet boundary, one of the conditions should be extrapolated from interior values. Extrapolated boundaries are described mathematics with Neumann type boundary conditions. The null gradient constraint is standardly applied to pressure, a condition that respects the physics, therefore a mathematical enquiry should establish whether or not such condition on pressure correction equation results in a well posed problem. In case it results to be untrue and other types of boundary conditions are needed to respect well posedness, a solution to maintain physical constraint is needed. In the weakly compressible form derived in Sec. 1.2.4, pressure corrector equation is a Poisson type equation that is an elliptic problem. A rst derivative constraint, Neumann boundary condition, is suitable for this kind of problem: mathematically the problem results to be well-posed. In the fully compressible version however a convective term is added and the behavior of the equation turns hyperbolic. For hyperbolic equations, unless one uses the unsteady form, resulting in an initial boundary value problem, a Dirichlet boundary condition is needed at the inlet too, to satisfy well-posedness constraint. Physically, it has already been underlined, specifying pressure corrector means to impose a pressure. To avoid inlet pressure not to change, it is updated at every iteration no more via the pressure corrector but via extrapolation from the interior using both value and gradient on rst cell. This passage is somehow the bottleneck of the entire process. To improve the speed of convergence for inlet pressure better would be to increase the diusive behavior of the equation near the inlet. This can be done using, instead of the zero xed value boundary condition, a mixed (Robin type) boundary condition:
cw + (1 c)
= cref + (1 c)
w
.
ref
(3.61)
This boundary condition imposes a value, depending on previously dened refer ence quantities ref and n , both zero in the case, to a linear combination of wall value and normal gradient weighted on scalar c. Again it is better underline how the combination of this boundary condition on the pressure corrector and the extrapolation of inlet values for pressure from the internal domain, is mathematically and physically valid for the entire range of Mach number.
ref
41
Chapter 3.
It has already been shown (see Eq. 3.35) how underrelaxation for the pressure corrector is necessary to obtain convergence. Furthermore an additional weighting factor mix , here improperly referred to as a relaxation factor, has been included in the pressure correction equation, scaling convective terms, to balance the relative inuence of the diusion term over convective term:
AP =
nb
anb + mix (u y u x + u y + u x) . w s e n
(3.62)
Just to show how the nature of the two relaxation factors is completely dierent, typical values for them two are indicated: p = 0.01 and mix = 0.9.
Mach Number
Mach Number
0.75
0.75
0.5
0.5
1.5
2.5
3.5
4.5
(a) Mach prole in transonic (b) Mach prole in supersonic (c) Mach prole in supersonic ow. ow. ow with two-bump geometry.
gives the Mach number distributions along the walls and density gradient magnitude contour plot shown in Fig. 3.2(a) and Fig. 3.3(a). In the supersonic case,
42
Chapter 3.
M ain = 1.65, the ow results supersonic all along the bump: Mach number distributions and density gradient magnitude contour plots are shown in Fig. 3.2(b) and Fig. 3.3(b). These results correspond to reference solutions from literature Peric et al. (1993); Karki (1986a). Fig. 3.2(c) and Fig. 3.3(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. 3.2(b), Fig. 3.2(c) and Fig. 3.3 the second bump does not inuence the ow upstream indicating that the solution algorithm correctly reproduces the hyperbolic behavior of the ow.
43
Chapter 3.
44
Chapter 4
Turbulence Modeling
This chapter is focused on the modelling of turbulence with the aim of nding closure to the set of conservation equations proposed in chapter 1, namely from Eq. 1.25 to Eq. 1.28. After a brief introduction to turbulent phenomena, classical approaches to turbulence modelling are presented with special interest in the RANS approach. Main part concerns the presentation of the many turbulence models capable of partially avoiding such defects in the prediction of turbulent eects, implemented in OpenFOAM for this thesis.
The physics of turbulence . . . . Models for Turbulence . . . . . . RANS approach . . . . . . . . . . Standard k turbulence model
4.4.1 4.4.2 4.4.3 4.4.4 4.4.5
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . . . . . . . .
47 48 49 53
53 54 55 55 57
4.5 4.6
Equation for turbulent kinetic energy . . . . . . Equation for turbulent kinetic energy dissipation Drawbacks of standard two equation model . . . Wall functions . . . . . . . . . . . . . . . . . . . Stability . . . . . . . . . . . . . . . . . . . . . . .
57 58
59 60 61 62 63 63 64
. . . . . . .
. . . . . . .
65 66 69
67
70 71
45
Chapter 4.
Turbulence Modelling
4.9.3 4.9.4
72 73
46
Turbulence Modelling
Chapter 4.
phenomena may be described with the following characteristics (White, 1991; Malalasekera and Versteeg, 1995):
Fluctuations in time superimposed upon a mean value for each quantities: (t) = + (t). Connected with this is the extension to three dimensions also for bidimensional ows: velocity uctuates always in 3D. Eddies or uid packets intermingle and ll the shear layer. Eddies with a wide range of length scales, coexist at the same time and at the same place. Random variations in uid properties. Such variations are not whitenoise shaped, each quantity has a specic continuous energy spectrum. Self-sustaining motion. New eddies replace those lost by viscous dissipation. Kinetic energy is handed down from bigger to smaller eddies in what is termed the energy cascade. Large scales (energy-containing subrange) contain most of the energy while smallest scales (dissipation subrange) dissipate such energy into heat. Scales laying in between belong to the so called inertial subrange: a transport region for the cascade process. Mixing is much stronger than in laminar ows. Turbulent eddies actively increase diusion by moving in the three dimensions. Heat transfer and friction are strongly enhanced.
Transition from laminar and turbulent regime always occurs when a certain, called critical, Reynolds number is reached. Starting from such a description, one cannot hope turbulence to be isotropic: only the smallest eddies in fact are non-directional, with bigger ones strongly dependent on mean ow direction.
47
Chapter 4.
Turbulence Modelling
Direct Numerical Simulation (DNS): it is not a model, Navier-Stokes Equations are computed in their most general form meaning that the complete spectrum of involved frequencies and length scales are solved. This approach is only feasible at current time for Low Reynolds number ows, due to limitation on computer resources, and used mainly as validation test for other approaches or as an help in understanding turbulence physics. Large Eddy Simulation (LES): remembering that energy is contained mainly in the largest scales, this model reduces the range of interest only to the biggest vortices. It employs, in fact, a time dependent three dimensional computation of the large-eddy structure and a model for the small scales. It basically consists in ltering in space the Navier-Stokes Equations with a high-pass lter, resolving for scales that actually are the energy-containing scales and modelling dissipation subrange behavior. The cut-o scale usually lies into the inertial subrange (Davidson, 2006). LES is becoming more and more popular in the CFD community, particular complexities in treating boundary conditions and the need for wide computer resources still limit the use to simple geometries and specic areas of interest in which turbulence modelling is more than fundamental (turbulent combustion, wake eects, chemical reactions). Reynolds Averaged Navier Stokes (RANS): Navier-Stokes Equations are averaged in time on a period big enough to contain also lowest frequency oscillation. The unsteady behavior of the turbulent ows is completely neglected, turbulence become a steady phenomenon simply considering the eects of uctuations onto the mean ow. At the state of art RANS simulations are the standard for ows involving heat transfer of industrial interest. In fact unsteady phenomena result in being determinant for such simulations. The ease of implementation and the speed in solving, supported by a good accuracy in modelling mean ows, are reasonably believed to make this approach still practised in middle-term. Since this thesis is treating turbulence with a RANS approach, a deeper
48
Turbulence Modelling
Chapter 4.
introduction to such an approach is needed. This will be the subject of next section.
Hybrid: this class is aiming in taking advantage of low computational cost of the RANS approach and the good condence of LES simulation. The idea is to use the two approaches in dierent areas where ow conditions are diverse. In particular LES is used only in the zone were RANS simulations are likely to fail, namely separating zone.
= 0, = Ui Uj 2 Uj + ij xj xj xi 3 xj p T = + k + + Sh , t xj xj = RT . +
(4.1) (4.2)
+ Fbi ,
(4.3) (4.4)
f (x) = f f = lim
1 t t
f (x, )d .
0
(4.5)
Before applying this operator to the complete set of equations, it would be better see what happen if it is applied to instantaneous uid properties:
=+ =+ .
(4.6)
The average of the mean value is equal to itself, the average of the uctuating part is zero:
(4.7) (4.8)
= 0,
all this comes directly out from the denition of the period t to be big enough to include all eects of unsteady turbulence. If products appears inside the averaging operator however, uctuation do not completely cancel out:
= ( + )( + ) = + + + = + . (4.9)
It must be underlined that also Favre (density-weighted) average:
1 f (x) = f f = lim t t
f (x, ) d =
0
f (x) ,
(4.10)
49
Chapter 4.
Turbulence Modelling
is very important for compressible ows because it exactly respects the following simplication (Knight, 1997):
Ui Uj = Ui Uj + Ui Uj .
(4.11)
Remembering such a clarication, in the following derivation Reynolds and Favre averaging will be distinguished. After this introduction to averaging, it is better start using this operator on the Navier-Stokes Equations. First start with continuity:
(4.12)
Ui (Ui Uj ) p + = + t xj xi xj
Ui Uj + xj xi Ui Uj + xj xi
2 Uj ij +Fbi 3 xj 2 Uj ij +Fbi 3 xj
p + xi xj
Ui Uj + xj xi
2 Uj ij 3 xj
+ Fbi , (4.13)
Uj Ui Ui Uj = + + + 2 xj xj xi xi
(4.15)
The RANS equations look pretty much like the Navier-Stokes Equations, Eq. 4.1, except for an extra term in momentum and energy equation, depending on uctuating quantities. These terms appear inside the divergence operator and they are often moved to the right hand side as companion terms of laminar viscous stress and conduction terms. That is why they are in literature known as Reynolds stresses and turbulent heat uxes. RANS approach will now used from now on in this thesis and so to clarify notation all the bars, apart from where specically needed, are dropped with the implicit notation that a = a, a and a = a , a . After the Reynolds Averaging, temporal dependency is of no interest anymore, so simplication of steady-state can nally be introduced, just drop the
50
Turbulence Modelling
Chapter 4.
= 0, = Ui Uj 2 Uj + ij Ui Uj xj xj xi 3 xj T = (k h0 Uj ) + + Sh , xj xj = RT . +
(4.16) (4.17)
+ Fbi ,
(4.18) (4.19)
It comes out that modelling turbulence is another way of saying for nding a way to treat the two terms: Ui Uj and h0 Uj . These terms introduce 9 more unknown into the Navier-Stokes Equations system, that before was determined but now is undetermined with 9 degrees of freedom. Constraints should be introduced if the system want to be solved: introducing these constraints is modelling turbulence. Modelling these terms is also known as the closure problem, explicitly pointing out that the number of unknowns [15: density, velocity (3), pressure, Reynolds stresses (6), enthalpy, turbulent heat uxes (3)] is larger than the number of equations (6). Dierent levels of approximation as been proposed to model in a more complex or simpler way those terms, depending on the needed accuracy and the available computer resources. Main class of models are quickly introduced again in order of decreasing computational complexity. To better understand energy equation has been neglected (drop o 1 equation and 4 unknowns) and turbulent heat ux is implicitly treated like Reynolds stress.
Dierential Stress Models, more commonly called Reynolds Stress Transport Model (RSTM) or Second Moment Closure models, solve a separate scalar transport equation for each stress component and closure is reached. The biggest drawback of this category of models can be guessed easily: it is computationally very expensive compared with the following methods. Moreover, stability is far from being easy to obtain and, even if they include more turbulence physics than eddy viscosity models with advection and production terms (energy-in terms) derived exactly, they still require modelling for important terms quote: redistribution and dissipation terms (Apsley, 2006). Eddy Viscosity Models: the RSTMs are the only models not taking advantage of Boussinnesq assumption, other models in fact, to prevent solving additional dierential equations, relate Reynolds stress tensor to strain rate or, in other words, to velocity gradients: Ui Uj = T = t Uj 2 Uk Ui + ij xj xi 3 xk 2 kij 3
(4.20)
51
Chapter 4.
Turbulence Modelling
where last two terms are added to be consistent with continuity and turbulent kinetic energy denition, that is to have the proper trace for turbulent stress tensor (Davidson, 2003; Knight, 1997). The problem now shifts to a way of guessing the eddy viscosity t , that is why the following classes of turbulent models are called Eddy Viscosity Models. The main advantage of such methods is the facility of implementation in viscous solvers: just replace molecular viscosity with an eective viscosity ef f = + t and the same for thermal diusivity with an eective thermal diusivity ef f = + t . This thesis is proceeding further with eddy viscosity models so the last and really implemented form of the Navier-Stokes Equationsis presented. Moreover a change in variable for the energy equation has been done solving for static enthalpy instead of total:
= = + = =
0,
(4.21) (4.22)
Uj 2 Uj Ui ef f + ij xj xj xi 3 xj h p (ef f ) + Ui + + Sh , xj xj xi RT .
2 kij + Fbi , 3
(4.23) (4.24)
Even if theoretical foundation is limited to simple shear ows, eddy viscosity models are the most commonly used because they are very cheap in CPU time, well known in their pros and cons and global predictions usually do not fail so much. At the contrary, the main drawback is the little turbulence physics they are based on: anisotropy and history eects are completely neglected. Due to the mathematical structure of the model in fact, in the best case one can hope to accurately represent only one Reynolds Stress. Applicability of such models should be restricted in cases where stress are almost mono directional. This is the case almost never but often it is possible to individuate a major stress, the idea is to well predict the main one and neglect if the others are not that well modelled. This category is subdivided again into:
Algebraic Models also called zero equation models because do not solve any extra partial dierential equation and t is obtained from mean ow properties and a prescribed, geometry dependent characteristic length. One Equation Models, in which a length scale is also prescribed but an extra partial dierential equation is solved for k. Two Equations Models where transport equations are specied for both k and its rate of dissipation (absolute or specic ), and t is then computed locally from the value of these scalars.
52
Turbulence Modelling
Chapter 4.
The fundamental limitation of zero equation models is the assumption that Reynolds Stress can be modelled on mean ow: turbulence does not in fact respond instantly to changes in mean ow but rather adjusts over a time scale typical of the turbulent structure. One equation models, even if feels the inuence of uctuating properties via turbulent kinetic energy equation, do still depend on geometry via the denition of a characteristic length. First, in order of complexity, class of model to be independent on the denition of an appropriate length scale is the two equations class. This property is fundamental in predicting many type of ows, for example recirculating ows: mixing length in fact cannot be used for ows in which diusion and convection are not negligible. The following will be consistent with the choice of two equations models as a good compromise between area of applicability and computational feasibility.
lt = Cmu
k2 , k2 .
(4.25)
t = C k 2 lt = C
(4.26)
(k) (kUj ) + = t xj xj xj
k xj Ui Uj U Ui Ui i . (4.27) xj xk xk
U U U + Uj p 2 j i i
Terms on the rst line need no further modelling so eorts should be directed in modelling the last three blocks. First of these three comes the turbulent diusion of kinetic energy: transport of velocity uctuations by the uctuations themselves, almost always modelled by use of gradient-diusion assumption:
U U U + Uj p 2 j i i
t k k xj
(4.28)
53
Chapter 4.
Turbulence Modelling
where k is the so called Turbulent Prandtl number whose value is usually close to unity. Second term represents the rate of production (positive source term) due to the mean ow and is commonly approximated, with the help of Boussinesq approximation, by:
Pk = Ui Uj
Ui = t xj
Ui Uj 2 Uk + ij xj xi 3 xk
Ui 2 Uj k . (4.29) xj 3 xj
In order to make equations the most implicit as possible, in computational uid dynamic it is common to split the implicit and the explicit part of the production term:
Pk =
G
explicit
2 Uj . k 3 xj
implicit
(4.30)
Last term is the sink term or dissipation for which a second equation is needed:
Ui Ui = . xk xk
(4.31)
(k) (kUj ) t k + ( + ) t xj xj k xj
= Pk .
(4.32)
() (Uj ) t + ( + ) t xj xj xj
= C1 Pk
2 C2 . k k
(4.33)
In Tab. 4.1 standard values for the introduced constants are given.
C 0.09
C1 1.44
C2 1.92
k 1.0
1.3
Such equations should be solved together with the Navier-Stokes Equations and, as one could guess and see, coupling is pretty strong. But as long as Eq. 4.32 and Eq. 4.33 are much stier than the others, they are usually solved segregated in order to increase convergence properties. In addition it is necessary to use under-relaxation techniques in iterative methods.
54
Turbulence Modelling
Chapter 4.
w = e,w
UP , yP
(4.34)
55
Chapter 4.
Turbulence Modelling
where the subscript P stands for rst node from the wall, e,w is the eective kinematic viscosity at the wall + + 1 for yP < y + (4.35) e,w = yP + + + 1 for yP > y y + ln(1+(y + y + ))
P
+ and y is a non-dimensional matching length whose typical value is 11.6 for smooth walls. In the equation for k , Pk and should be replaced at the rst node with an averaged value: av Pk
= =
(w /) u0 yP
2 + + ln[1 + (yP y )]
+ + (yP y ) + + 1 + (yP y )
(4.36) (4.37)
av
u3 0 ln yP
yP y
+1 ,
(4.38)
and
y =
2kP . 0.5 C u0
(4.39)
As a consequence dissipation equation is not solved in the rst layer of cells being the near wall node value superimposed (Apsley, 2006). Furthermore near wall temperature proles should also be corrected when High Reynolds turbulence models are used. The log-law can be adapted to provide an equation that relates the value of temperature at the near wall node TP , to that at the wall Tw and to the value of the wall heat ux rate qw .
+ 0.25 0.5 TP = Cp C kP
Tw TP , qw
(4.40)
where
+ TP = t
1 ln
+ 0.25 + yP exp(C y ) + y
+ 0.25 + y C
Pr 1 t
(4.41)
This equation can be interpreted in two dierent ways depending on physical boundary condition for temperature. If xed wall heat ux simulations need to be performed, the numerical condition for T will be xed value given by:
+ Tw = TP + TP
qw . 0.25 0.5 C p C k P
(4.42)
If instead temperature is imposed on the boundary, the temperature gradient must be held xed numerically:
qw = k
T n
0.25 0.5 = Cp C kP w
Tw TP . + TP
(4.43)
56
Turbulence Modelling
Chapter 4.
Here notation can be confusing so it would be better underline that k on the left hand side is the thermal conductivity while kP on the right hand side is the turbulent kinetic energy at rst node (Iacovides, 2004; Apsley, 2006). The values of Tw and qw depend on TP so they change during the iteration process. When convergence is reached the value of TP , Tw , qw will however exactly satisfy the wall function.
4.4.5 Stability
After this quite long digression, some words more should be spent on the problems connected with the standard k model. If k is applied on regions of low turbulence, numerical instability may arises. The problem is connected with the modelling of the destruction term in the equation: 2 /k . Even if both quantities goes to zero, the rate of decrease is not always correct (Davidson, 2003). Even if they are known to fail in well predicting adverse pressure gradient ows, turbulence model of the k class are still the most used two equations models world wide (Wilcox, 1998).
57
Chapter 4.
Turbulence Modelling
of the damping functions and some extra sources in dissipation equation. It is known from literature that in high strain rate regions eddy viscosity models overpredicted turbulent kinetic energy: this problem is sometimes referred to as stagnation point anomaly (Durbin, 1996). The cause of such an overprediction stands in the modeling of the production term Pk supposed to be proportional to the rate of strain Sij . When Sij is too large, and that happens not only near stagnation points but also in swirling ows and even in turbine passages, such proportionality fails to be physical or better the errors with the simplications become unacceptable. To avoid such overprediction of turbulent kinetic energy, turbulent time scale, normally computed as the ratio between k and , is bounded. Such a bound was derived from a sort of realizability constraint for Reynolds stress tensor (Medic and Durbin, 2002). Having modeled near wall behavior, now the focus shifts towards well matching good near wall predictions with suitable modeling of ow structures far from the wall. Being the standard k model still one of the best in predicting High Reynolds zones, Two Layer k models have been implemented. Such methods, in fact, consist in patching together a one equation model in the near wall layer and a two equations High Reynolds model, namely standard k model or the renormalized group variant, in the outer layer (Rodi, 1991). It has already been mentioned how k models are likely to fail in predicting adverse pressure gradient ows. Another class of model, writing the second equation in terms of specic dissipation , is instead better performing for separating ows. Several models have been proposed all sharing the same defect in depending from freestream values. Interesting is the k in the SST (Shear Stress Transport) variant: it includes the modication of the standard k to avoid sensitivity to (quite arbitrary) freestream values of (Menter, 1994, 1993). The basic idea is similar to Two Layer models: two dierent approaches are merged together to model the two dierent ow regions. The sublayer and logarithmic model is the standard k , 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 , written in terms of , has been preferred due to its good compromise in predicting dierent kind of ows. Summarizing it is kind of easy to obtain turbulence models that well perform in wall bounded k or in frees shear ows k but since most complex ows include both types of regions, the task is to well match both behaviors.
58
Turbulence Modelling
Chapter 4.
t = C f
k2 .
(4.44)
Little is known about such function but to maintain agreement with standard k in the High Reynolds zone it must tend to unity in the fully turbulent logarithmic layer. In addition an experimental curve can be obtained calculating an expression for f for a bidimensional boundary layer and substituting measured data for k + , + , uv + :
f =
uv + + . C k +2 (dU + /dy + )
(4.45)
It exhibits an almost constant value for y + < 15, a linear increase up to y + = 60 and a gradual approach toward unity. Equation for turbulent kinetic energy dissipation is also damped in its source terms and an extra term is added:
t () (Uj ) + ( + ) t xj xj xj
= 2 C2 f2 + E , (4.46) k k
C1 f1 Pk
The function f2 is introduced primarily to incorporate Low Reynolds number eects in the sink term. The eects of such a function is generally limited to the viscous sublayer and nuances in the shapes do not exert large inuence on the overall results. E and f1 usually inuence magnitudes at peak or at the wall. Equation for k remains substantially invariant with only the addition of an extra dissipation terms for some models. Damping functions are in general expressed in terms of two turbulent non dimensional numbers: Rey = ky/ , (4.47)
Ret
k 2 / ,
(4.48)
where here, and from now on, y is wall distance if not specied. After this description of the commonalities between such models, each model is analyzed in its own specic formulation.
t () (Uj ) + ( + ) t xj xj xj
= C1 Pk 2 C2 f2 , (4.49) k k
59
Chapter 4.
Turbulence Modelling
t k (k) (kUj ) + ( + ) t xj xj k xj
= Pk .
(4.50)
This model uses another non-dimensional number to express the damping functions, it is again a sort of Reynolds number calculated via turbulence dissipation:
y0.25 . (4.51) 0.75 The production term in turbulence dissipation equation is not damped so only two damping functions are present in this model: y = f f2 = = 1 exp 1 exp y 14.0 y 3.1
2 2
1+
Ret 200
2
, .
(4.52) (4.53)
1 + 0.3 exp
In Tab. 4.2 standard values for the introduced constant are given. Wall bound-
C 0.09
C1 1.5
C2 1.9
k 1.4
1.4
Table 4.2: Standard coecients values for Abe et al. (1994) Low Reynolds k .
w = 2
k , y2
(4.54)
where this condition is intended imposed on rst near wall node. For k , it is better remind the no-slip condition on solid surfaces.
4.6.2 Chien
The following is the model proposed by Chien (1982), known to the author from (Yoder and Georgiadis, 1999) and referred to as CH later on.
() (Uj ) t + ( + ) t xj xj xj C1 Pk
=
(4.55) (4.56)
2 C2 f2 2 2 exp 0.5y + , k k y = Pk 2 k . y2
t k (k) (kUj ) + ( + ) t xj xj k xj
The viscosity damping function:
f = 1 exp 0.0015y +
is expressed this time in terms of:
(4.57)
y+ =
yu .
(4.58)
60
Turbulence Modelling
Chapter 4.
Looking carefully at such an equation, it appears how dicult can be to implement y + for cells far from the wall. It is not that easy in fact to dene what really u is for an internal cell. This computation was done with reference to the friction velocity value of the closest wall cell. The other damping function is instead modeled with the standard turbulent Reynolds number:
f2 =
1 0.22 exp
Ret 6
(4.59)
In Tab. 4.3 standard values for the introduced constant are given.
C 0.09
C1 1.35
C2 1.8
k 1.0
1.3
Wall boundary value for will be held xed to 0, as one may notice in fact an extra term has been introduced into turbulence kinetic energy equation taking into account the correct near wall behavior of the sink term in that equation. It is unnecessary to repeat that wall boundary condition for k is a null Dirichlet condition.
f f2
5.29 Rey
(4.62) (4.63)
but source term in equation does not remain unchanged. An extra production term is doping the near wall behavior:
Pk = 1.33f2 Pk + 2
k exp 0.00375Re2 y y2
(4.64)
61
Chapter 4.
Turbulence Modelling
C 0.09
C1 1.44
C2 1.92
C3 -0.33
k 1.0
1.219
Table 4.4: Standard coecients values for Chen et al. Low Reynolds k .
In Tab. 4.4 standard values for the introduced constant are given. Wall boundary condition for are specied in such a way:
w =
Pk k +2 2 . y
(4.65)
= =
, 2| k
0.5
(4.66)
| .
(4.67)
An extra term depending on will also be added in the kinetic energy equation:
() (Uj ) t + ( + ) t xj xj xj C1 G
(k) (kUj ) t k k + ( + ) = Pk 0.5 . (4.69) t xj xj k xj xj xj Apart from the denition of this new parameter y : , k used in the denition of the viscosity damping function y = y
3 f = 1 exp 0.01y 0.008y
(4.70)
(4.71)
the main dierence stands in the tuning of constants with the change in y : y , (4.72) k = 1.44 1.1 exp 10 y = 1.3 1.0 exp . (4.73) 10 The eect of this tuning is to damp the diusivity factor, without constraining the production terms. In Tab. 4.5 standard values for the introduced constant are given. Wall boundary value for and k as well are xed to 0.
62
Turbulence Modelling
Chapter 4.
C 0.09
C1 1.44
C2 1.92
Table 4.5: Standard coecients values for Hwang and Lin Low Reynolds k .
() (Uj ) t + ( + ) = t xj xj xj 2 (Uj ) 2 , (4.74) C1 f1 G C1 f1 C2 f2 k 3 xj k t k (k) (kUj ) + ( + ) t xj xj k xj The three damping functions are: f f1 f2 = = = = Pk .
(4.75)
1+
20.5 Ret
, .
1 exp Re2 t
In Tab. 4.6 standard values for the introduced constant are given. Wall
C 0.09
C1 1.44
C2 1.92
k 1.0
1.3
Table 4.6: Standard values for Lam and Bremhorst Low Reynolds k coecients.
boundary condition for is of type zero gradient normal to walls. Obviously k is respecting the no slip condition.
() (Uj ) t + ( + ) t xj xj xj
= 2 C2 f2 , (4.79) k k
C1 (Pk + Pk )
63
Chapter 4.
Turbulence Modelling
t k (k) (kUj ) ( + ) + t xj xj k xj
= Pk .
(4.80)
Like in CLL model, two damping functions and a modied production term lead all the modication to the original equations:
f f2 Pk
Re2 t
0.00222Re2 y
In Tab. 4.7 standard values for the introduced constant are given. Wall
C 0.09
C1 1.44
C2 1.92
k 1.0
1.3
Table 4.7: Standard coecients values for Lien and Leschziner Low Reynolds k .
w = C1
Pk C2 f2
(4.84)
For the companion model, called LNR (Lien, 1992), everything but the following remain invariant:
f D Pk
= = =
4.6.7 Realizability
Excessive levels of turbulence kinetic energy are predicted by standard twoequation models in region of large rate of strain. This was originally recognized in stagnation point ows, but it will be seen here that it is a more widespread anomaly. The problem is connected with a big overproduction of k, in case a moderate level of k is subjected to a large of rate of strain. Such over production can be due to, an underestimation of the sink term or/and an overestimation of turbulent viscosity. These ideas could be merged together into a bound for the local turbulent time scale Ts (Medic and Durbin, 2002). In detail, rst step is to reformulate the expression for t :
t = C f u2 Ts ,
(4.88)
64
Turbulence Modelling
Chapter 4.
where u is the velocity scale and Ts is the local turbulence time scale. For the previously proposed models u = k and Ts = k/ and so equation is rewritten in terms of Ts :
t () (Uj ) ( + ) + t xj xj xj
= C1
Pk C2 . Ts Ts
(4.89)
Ts = min
k , 6C S
(4.90)
in which S = 1 G and = 0.6 are usually used. As a consequence at large rate 2 of strain Pk is growing like |S| rather than |S|2 . The option of selecting the realizability constraint or not has been implemented in all previous models with a switch to be controlled by the user. This correction will be referred to in future chapters, when brevity is a must with the postscript Real.
f = 1 exp
Rey A
(4.91)
In order to avoid that non-turbulent zones far from the wall were calculated with the one equation model, a user controlled parameter denes the wall layer thickness meaning that above such value f was not checked any more and the standard k directly applied. The k l models are based on the following equations, expressing both dissipation and eddy viscosity via a turbulent length scale:
k2 , l = C kl . =
(4.92) (4.93)
65
Chapter 4.
Turbulence Modelling
l l
= Cl yf , Rey = Cl y 1 exp A .
(4.94) (4.95)
In Tab. 4.8 standard values for the introduced constants are given.
C 0.09
Cl 2.5
A 5
A 62.5
66
Turbulence Modelling
Chapter 4.
isotropic even for relatively simple ows such as a three-dimensional boundary layer ((Davenport and Simpson, 1982)). For the particular case of lm cooling ows, eddy viscosity anisotropy was examined by Acharya et al. ((S. Acharya and M. Tyagi and A. Hoda and F.Muldoon, 1999)). In this work the dierent eddy viscosity components arising in a lm cooling ow were assessed by taking the ratio of the computed shear stress and its respective strain rate component as computed in a LES . The vertical (xy ) eddy viscosity can be regarded as the turbulence viscosity used by conventional EV M s as all of them are tuned on ows, such as two dimensional boundary layer, backward facing step and so on, where the mean ow is aected mainly by this component which is therefore the dominating one. Looking at gure 4.2, where the results obtained by Acharya et al. are reported, it can be seen that the xz eddy viscosity is considerably greater than the xy component, that is the wall-normal component. This suggests that common EV M s are forced to underestimate the lateral turbulent diusion. Even if Fig. 4.2 shows a very complex situation which can hardly be aorded by simple anisotropy corrections anyway it clearly indicate that directional modications of eddy viscosity are likely to improve the predictive capabilities of standard EV M s.
Figure 4.2: LES predictions of shear stress (top row), rate of strain (middle row), and eddy
viscosity (bottom row) for the xy component (left column), yz component (middle column) and xz component(right column).
67
Chapter 4.
Turbulence Modelling
EVMs. An ad hoc correction is achieved by multiplying the eddy viscosity associated to the cross-Reynolds stress by an anisotropy factor to be computed algebraically. In the approach of Bergeles et al. the correction was applied just to the momentum equations leaving the remaining of the equations unaltered. A wall function approach was used and was assumed to be constant in the boundary layer assuming a value of 4.5. Outside the boundary layer a value of 1 was assumed. The value of the anisotropy factor in the boundary layer was inferred by experimental data collected by Quarmby and Quirk (A. Quarmby and R. Quirk, 1974). More recently, a more rened approach was used by Lakehal et al. in a series of papers on this topic ((A. Azzi and D. Lakehal, 2001), (A. Azzi and B.A. Jubran, 2001), (A.Azzi and D.Lakehal, 2002), (D.Lakehal and C.S. Theodoridis and W.Rodi, 2001), (D.Lakehal, 2002)). First of all they used a zonal low Reynolds treatment for wall boundaries. Then, an algebraic expression based on DNS data was used to compute the anisotropy factor as a function of the non-dimensional wall distance y + and also, the energy as well as the turbulence kinetic energy equations, in addition to momentum equation, were corrected to take in account for the anisotropy of the turbulence eld. The model derivation is based on physical reasoning. Following the conceptual scheme of Azzi et al., the transport equations for the Reynolds stress tensor may be written as follows, if homogeneous near-equilibrium turbulence, unaected by external forces is assumed:
ui uj =
(4.96)
Pij = ui uk
(4.97)
Under the addition hypothesis that the ow is homogeneous in the stream-wise (x) direction and that v w is negligible with respect to u v and u w the only non zero shear stresses are
u w = 2
u u ; u v = 3 z y
(4.98)
In the previous expressions we assumed a cartesian reference coordinate system with the x axis parallel to the stream-wise direction, the y axis in the span-wise direction and the z axis normal to the plate in a jet cross ow. From Eqs. 4.96, 4.97 and 4.98 the anisotropic eddy viscosities 2 and 3 may be recast as follows:
2 = Cs
kw 2
; 3 = Cs
kv
(4.99)
2 = Cs
w 2 k2 w 2 k2 ; 3 = Cs k k
(4.100)
68
Turbulence Modelling
Chapter 4.
Where = v 2 /w 2 is the ratio of the span-wise and wall-normal uctuations. Eq. 4.100 shows that eddy viscosity for the cross stresses in the stream-wise (2 ) and spanwise (3 ) directions have to be dierent according to an anisotropy factor which is found to depend on the ratio of the normal stresses v 2 and w 2 . Actually, approaching the wall, the normal stress w 2 decays at a much faster rate with respect to the v 2 because of the damping eect of wall itself. This eect is completely overridden by conventional EVMs which use a scalar eddy viscosity. In order to close the anisotropic eddy viscosity model dened by equations 4.100 the function has to be dened. As said above, Bergeles assumed a peace-wise constant prole for posing = 4.5 within the boundary layer and = 1 outside. Lakehal et al. (D.Lakehal and C.S. Theodoridis and W.Rodi (2001); A.Azzi and D.Lakehal (2002)) used the channel ow DNS data of Kim et al. (J.Kim and P.Moin and R.Moser (1987); W.Rodi and N.N.Mansour and V.Michelassi (1993)) to dene a proper prole for as a function of the nondimensional wall distance y + . The actual prole is shown in gure 4.3. This function is strictly valid only for channel ow without separation. In order to obtain a more general functional relation the dependence of on Y + is relaxed and nally the relation is given the following expression:
Figure 4.3: Comparison of analytical correlation and DNS data for the anisotropy factor .
2 = M ax[
(4.101)
69
Chapter 4.
Turbulence Modelling
model able to predict eects of adverse pressure gradients and to integrate through the viscous sub-layer (Wilcox, 1998). After that, many variants has been proposed making k models the second most used class of two-equation turbulence models. As hinted in Sec. 4.5, both k and k class of models have been shown their own strengths and weaknesses in the course of the years. k in particular is recognized as the best choice for free shear ows (mixing layer, jets and wakes) and reached good accuracy also in wall bounded ows. k certainly performs better in wall bounded ows, especially the ones with adverse pressure gradients, although its freestream values dependency is notorious, in addition it is credited with a higher numerical stability (Bredberg, 2001).
4.9.1 Original k
The model introduced now is proposed in (Wilcox, 1998) as the evolution of the well-known k model presented by Wilcox in 1988. New dissipation coecients, maintaining the high precision for boundary layers and removing overprediction of free shear layer spreading rates, have been introduced. Thus such model is applicable to both wall-bounded and free shear ows and the minimum requirement is satised. Denition of eddy viscosity directly follows from the k class one but due to main authors dierent notation sometimes there is ambiguity regarding the old standard k C constant. Therefore it is not superuous to write:
t =
k .
(4.102)
Then the equation for turbulence frequency is presented together with standard k equation:
() (Uj ) t + ( + ) t xj xj xj
= C1 Pk C2 2 , k
(4.103)
= Pk k .
(4.104)
k 2
w = 10.0
6.0 . C2 y 2
(4.105)
70
Turbulence Modelling
Chapter 4.
Again this expression is intended as an imposition on the rst near wall node, noticing that the value for is going towards innity as distance from the wall is reducing to zero.
(4.106)
(4.107)
First the blending function F1 should be implemented, so we need some auxiliary functions such as:
CDk arg1
to nally write:
= max 2 1
(4.108)
(4.109)
4 F1 = tanh arg1 .
(4.110)
K C1 C2
= F1 + (1 F1 ) , = F1 k + (1 F1 )k , = F1 C1 + (1 F1 )C1 , = F1 C2 + (1 F1 )C2 .
Values for the introduced constants can be found in Tab. 4.10. At the rst near wall node is specied to be:
w = 10.0
6.0 . C2 y 2
(4.115)
71
Chapter 4.
Turbulence Modelling
0.09
C1 0.4404
C2 0.0828
C1 0.5976
C2 0.075
k 1
k 0.5
0.856
0.5
12 = a1 k .
(4.116)
For conventional two-equation models the principal shear stress can be computed as:
12 = t =
Productionk a1 k , Dissipationk
(4.117)
where is the vorticity and a1 is a constant. In adverse pressure gradient ows the ratio of production to dissipation could be much larger than one, meaning that Bradshaw's hypothesis is substantially violated. To satisfy Eq. 4.116 within the framework of eddy-viscosity models t should be bounded in such a way:
t =
a1 k . max[a1 , F2 ]
(4.118)
Of course original formulation of the eddy-viscosity must be maintained for free shear layers so the same blending function approach as for the baseline model is adopted. First come the equation of conservation for turbulent properties:
(4.119)
= Pk k .
(4.120)
arg2 F2
= max 2
500 k , 0.09y y 2
(4.121) (4.122)
2 = tanh arg2 .
72
Turbulence Modelling
Chapter 4.
CDk arg1 F1
= max 2 1
(4.123)
(4.124) (4.125)
= F1 + (1 F1 ) , K = F1 k + (1 F1 )k , C1 = F1 C1 + (1 F1 )C1 , C2 = F1 C2 + (1 F1 )C2 . 0.09 C1 0.44035 C2 0.0828 C1 0.59761 C2 0.075 k 1 k 0.5 0.856 0.5
a1 0.31
w = 10.0
6.0 . C2 y 2
(4.130)
Another variant of the k SST, later in this thesis also called simply SST, proposed to extend applicability beyond aerodynamic applications has been implemented following the instructions given in CFX guide CFX (2004). Modications regards eddy viscosity:
t =
a1 k , max[a1 , 2SF2 ]
(4.131)
CDk = max 2 1
k , 1010 xj xj
(4.132)
2 Uj Pk = min G k , 10 , 3 xj
and the value of C1 = 0.5532.
(4.133)
73
Chapter 4.
Turbulence Modelling
of the most common issues in the preparation of accurate calculation meshes for heat transfer analysis of complex geometries is the assurance of a proper y + distribution, particularly when the ow eld is distinguished by relevant variation of ow Mach number. Therefore, in order to increase grid independence a mixed approach between wall-function and Low Reynolds was added to the k SST model. The idea, see Menter (1993), is to blend the two approaches via a blending function calculated algebraically from the non dimensional wall distance. Both turbulent production and turbulent specic dissipation are imposed on the rst node mixing the eects of the Low Reynoldsand the High Reynolds contributions:
(4.134) (4.135)
where is calculated with an algebraic expression for y + dened with u = max(u,LR , u,HR ) . The same blending is applied to thermal quantities following Kader universal law B. (1982).
74
Chapter 5
Reynolds Averaged Navier-Stokes equations of mass fraction . . . . . . . . . . . . . . . . . . . . . . . . . 77 Perfect gases law and thermodynamical properties 77 Chemical Kinetics . . . . . . . . . . . . . . . . . . . 78
5.3.1 5.3.2 5.3.3 Arrhenius model . . . . . . . . . . . . . . . . . . . . Reaction mechanisms . . . . . . . . . . . . . . . . . Chemical kinetics in turbulent ow . . . . . . . . . . 78 79 79
5.4 5.5
80 81
5.6
5.5.1 5.5.2 5.5.3 5.5.4 5.5.5 5.5.6 5.5.7 5.6.1 5.6.2 5.6.3 5.6.4 5.6.5
Premixed Flames . . . . . . . . . . . . . . . . . . .
Combustion regimes and types of ame Turbulence and ame scales . . . . . . . Spectral Diagram . . . . . . . . . . . . . Premixed turbulent combustion models Bray-Moss-Libby model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
93
83 84 84 86 87 89 90 94 94 96 97 98
75
Chapter 5.
5.7
5.6.6 5.7.1
76
Chapter 5.
ui Yk Yk + t xi
xi
T + Sck ScT
Yk + k xi
(5.1)
where Sck and ScT are respectively the laminar and turbulent Schimdt numbers dened as: T Sck = ScT = (5.2) Dk DT where Dk and DT are respectively the laminar and turbulent diusions of k th specie that in this work, following a common procedure, are assumed to be equal to corresponding Prandtl's numbers dened in Eq. 5.4. Finally the last term of RHS k is the average reaction rate for specie k that requires a supplementary model.
(5.3)
with viscous heating T V neglectable if M a2 1.0 and xR representing i the heat sink due to radiative exchange. P r and P rT are laminar and turbulent Prandt's numbers dened as:
Pr =
Cp
P rT =
T Cp
(5.4)
where Cp (T ) = k Yk Cpk (T ) is the mixture mean specic heat at constant pressure. The thermochemical enthalpy is written as:
T
h(T ) =
Tref
Cp (T )dT +
Yk h0 (Tref ) k
(5.5)
having dened h0 (Tref ) as the formation enthalpy of k th specie. k The above reported RANS equations set, coupled with a turbulence model represents a powerful instrument to model reacting turbulent ow.
77
Chapter 5.
heat at constant pressure as the weighted average of species specic heats, without dropping their dependence on temperature. Indeed while considering uxes with high temperature gradients, such as those involving combustion, it is necessary to model each k th species specic heat at constant pressure as function of temperature, while the eect of pressure is neglected by using the perfect gases model.
hk =
R Wk
and
Cpk =
hk T
(5.6)
p
Specic heats can be evaluated starting from molecular properties of each specie, unfortunately this approach involves highly complicated functions, then the dependence on temperature is modelled with polynomial expressions. In present work, for each k th specie, has been used the expression reported in Eq. 5.6 where R = 8.31445 [kJ/(kg K)] is the universal perfect gases constant and Wk is the molar mass of specie k . Again from Eq. 5.6 we can notice that thermochemical enthalpy is formed by two part: the rst one, dependant on temperature, is the sensible enthalpy while the other, temperature independent, is the formation enthalpy referred to standard conditions (usually T = 298.15 K and p = 101325 P a) The model of perfect gases gives also a relation between temperature, pressure and density, the perfect gases equation of state:
pk =
k RT Wk
(5.7)
where Wk is the molar weight of k th species, and pk is its partial pressure. Using the Dalton's law for ideal gases mixtures (p = k pk ) and dening the mixture averaged molar mass as in Eq. 5.8, we can dene the equation of state for the mixture as reported in Eq. 5.9.
W =
1
Yk k Wk
(5.8)
p=
RT W
(5.9)
1 1 + 2 2 + 3 3 + . . . + k k 1 1 + 2 2 + 3 3 + . . . + k k
78
Chapter 5.
The speed of reaction is dened as the variation, in the unit of time, of each specie concentration weighted to its own stoichiometrical coecient:
q=
where ck indicates the molar concentration of a generic specie. The net reaction speed is dened as:
q = qf qr
where the subscript f and r stand respectively for forward and rearward reactions. In equilibrium condition the net speed is zero so qf = qr . Rearward and forward reactions speeds are usually modelled as a function of species concentrations:
K K
qf = kf
k=1
ck k
qr = kr
k=1
ck k
Where k is are the reaction order with respect to various components and their sum gives the global reaction order. kf and kr are the kinetics constants of the reaction and are dependant on temperature. These constants are evaluated by the Arrhenius' equation:
k = AT exp
Ea RT
(5.10)
where A is the frequency factor while Ea is the activation energy. A, Ea and are characteristics of each reaction.
79
Chapter 5.
into their averaged and uctuating components. This is a direct approach for k closure called moment closure that is very rarely used. Representation by Taylor's series introduces a new series of correlations among uctuating terms (such as Yk T ) that cannot be neglected a priori, hence a specical algebraical like modelling procedure is required, that is very hardly generalizable. Then it is usually preferred to close the expression of k by models that keep in account, in a more or less empirical way, the problem physics: these are the so-called turbulent combustion models that are described in following sections.
Considering Fig. 5.1, we can dene n as the normal to ame front, u the
reactants stream speed, and SL as the laminar ame front speed (zero for diusion ames, subsection 5.5), relative to reactants stream so, w, the ame front absolute speed, can be calculated as:
w = u + SL n
This denition allows to rewritten the ame stretch as:
K=
where
t
u + SL
(5.12)
The rst term of Eq. 5.12 represents the distortion eect of ame surface due to reactants speed gradients. It is a term owed to the lack of ow
80
Chapter 5.
uniformity and it is also present in steady ames: it is commonly called strain. Most common examples of strain aected laminar ames are the diusion steady strained ame (Fig 5.2(a)) and the premixed symmetrical counter ow ames (Fig. 5.2(b)). The rst one having a strain rate equal to a = ux , is steady due to strain itself. The second one having a strain x u rate equal to a = yy is a twin ame stabilized by two opposite reactant jets, where the ame fronts runs parallel to the stagnation plane.
The second term of Eq. 5.12 is due to curvature eect of ame surface. And it is equal to zero in diusion ames, because of the absence of a ame speed.
(a)
(b)
Figure 5.2: Examples of strained ames: a) steady strained one-dimensional diusion ame
and b) counterow premixed ames
Stretching induces modication of chemical processes then it cause modications of laminar premixed ames speed. If reactants ow is turbulent, the local ame stretch dened as in Eq. 5.12 represents the instantaneous stretching (for premixed ame and, by dropping the term dependant on ame speed, it can be representative of diusion ame too). Theoretically it should be possible to dene a Favre averaged form of stretching (K ), but practically the averaging procedure would produce a large number of Reynolds stresses terms. Hence it is, usually, preferred to model the ame stretch with the Favre averaged strain rate, expressed as function of turbulence eld characteristic variables ((Andreini, 2004)).
81
Chapter 5.
These ame are called non-premixed ames or diusion ames since diusion of reacting species is the rate-controlling processes ((Peters, 2000)). Diusion combustion has its main applications in Diesel engines, furnaces and airplanes engines. In many applications non-premixed ames not only involves separated uxes but fuel and oxidizer may be solids or liquid, then chemical-physical changes, such as evaporation, are needed. In Fig. 5.3 is shown a prototype of the diusion ame congurations, where from the right side comes the oxidizer (which may be diluted in other gases) while form the left side comes the fuel (diluted or not). Fuel and oxidizer diuse towards the reaction zone where they burn and generate heat. Temperature reaches its maximum in this zone and diuses away from the ame front towards the fuel and oxidizer streams. Again keeping the attention on Fig. 5.3 we can make some important considerations ((Poinsot and Veynante, 2001, chap. 3)):
Far away on each ame side, the gas is either too rich or too lean to burn. Chemical reactions take place only in those limited regions, where fuel and oxidizer are mixed adequately. The most favorable conditions occur where fuel and oxidizer are in stoichiometric proportions: a diusion ame usually lies along the points where mixing produces a stoichiometric mixture. Flames as the one reported in Fig. 5.3 are steady only when strain is applied to the ame, i.e. when the fuel and oxidizer streams are pushed against each other at given speeds. Diusion ames do not exhibit a reference speed, contrarily to premixed ones (chapt. 5.6): ames are unable to propagate towards fuel because of the lack of oxidizer and the lack of fuel in oxidizer stream makes impossible the propagation even in this side. The lack of propagation implies that the reaction zone does not move signicantly relatively to the ow eld.
82
Chapter 5.
A diusion ame does not have a reference thickness: an unstretched ame grows up indenitely with time while the thickness of a stretched ame depends essentially on stretch and can assume a very wide range of value.
This chapter rstly introduces the methods suitable to treat laminar diusion ames, then an extension of these model is performed to treat turbulent combustion.
5.5.1 Soot
A peculiar product of diusion combustion is soot. In ideal conditions hydrocarbons combustion processes lead to the formation of carbon dioxide (CO2 ) and water vapor (H2 O). Instead, while occurring real conditions, in fuel reach zones, combustion and pyrolysis of hydrocarbons produce intermediated species and radicals that may lead to soot formation. Soot formation analysis should be consider since ((Da Soghe, 2006)):
soot is a product of incomplete combustion processes so the more it is produced the worse is the combustion process productivity; soot is one the most dangerous pollutant for human health; soot is an highly radiatively active specie (cfr. subsubsection 6.5.2.2) that explain why it is considered in present work, and moreover this feature is exploited in steam generators and furnaces; soot is used in the production of inks and pneumatics.
Soot is mainly made of carbon and, in smaller proportions, of hydrogen, oxygen and nitrogen. Soot particles density is smaller that carbon ones, it may vary between 1700 1900 kg/m3 and depends on particles porosity. Usual dimensions of soot lay interval between 5 80 nm. A detailed description of soot formation processes and relative models can be found in (Da Soghe, 2006), (Pasquini, 2006) and (Lautenberger, 2002) and the processes involving soot can be summarized as:
Surface growth due to the merger of soot molecules to already formed particles
Particles coagulation and fragmentation provided by collisions between particles that may merge or fragment
Particles oxidation is the only process that may reduce soot mass and is
performed on particles surface involving reactions with oxygen and OH that produce CO and CO2 .
83
Chapter 5.
These processes are usually occurring at the same time and are strongly dependant on combustion system features. Usually, in common combusted, the produced soot (rising over 99 %) is almost completely oxidized before exiting the combustion system.
A mixing problem that provides the mixture fraction Z , introduced in subsection 5.5.3, as a function of space and time coordinates A ame structure problem where the links between ame variables and the mixture fraction are used to construct all other ame variables, i.e. the ame structure in Z -space
mj =
k=1
akj Wj mk Wk
(5.13)
mj Zj = = m
n k=1
akj Wj Yk Wk
(5.14)
The laminar equation of conservation for species k , that has its RANS form in Eq. 5.1, using the Fick's law for diusion term, is
(ui Yk ) Yk + = t xi xi
Yk Sck xi
+ k
(5.15)
Adding species conservation equations Yk in the same way as in Eq. 5.14 the j th element conservation equation is
Zj ui Zj + = t xi xi
n k=1
akj Wj Wk
Dk
Yk xi
(5.16)
84
Chapter 5.
where the chemical source terms k s do not appear since, for the mass conser vation of each element j , present in any reaction l in k species, having stoichiometric coecient kl is true that
n
akj Wk kl = 0
k=1
(5.17)
Again relying on the mass conservation of each element, if all diusivities are considered equal (Dk = D), the balance equation for the element mass fraction becomes
Zj ui Zj + = t xi xi
Zj xi
(5.18)
Considering the global reaction equation (omitting intermediates reactions) for the complete combustion of an hydrocarbon fuel Cm Hn
F Cm Hn + O2 O2 CO2 CO2 + H2 O H2 O
we can write
(5.19)
and
ZO = YO2 ,u
(5.20)
Using the original approach of Burke and Schumann of 1928, we can dene the conserved scalar can be dened as
(5.21)
this scalar becomes equal to 0 in stoichiometric condition. Normalizing beta we obtain the denition of mixture fraction:
Z=
2 1 2
(5.22)
that is equal to 0 in the oxidizer stream (denoted by the subscript 2) and equal to 1 in the fuel stream (denoted by subscript 1). After having dened the mixture fraction we can now write its transport equation remembering that we have considered all species diusivities equal to D.
Z ui Z + = t xi xi
Z xi
(5.23)
Normalization let state that every specie conservation can be achived with Eq. 5.23 with the boundary conditions: - Z=1 on fuel side - Z=0 on oxidizer side - other boundaries, such as walls and outlets, are all characterized by zero ux of each species, elements and mixture fraction. So the normal component of mixture fraction gradient is equal to zero on each of these boundaries
85
Chapter 5.
This last statements on boundary conditions together with the nature of Z equation, that is the same for all species, implies that by solving only the Z equation we have the concentration of all species Yk ,and consequently (section 5.7) the mixture mean molar weight, specic heat and formation enthalpy:
Yk = f (Z, t)
for
k = 1, N
(5.24)
mal diusivity is assumed to be equal to the species one and if boundaries are adiabatic ((Poinsot and Veynante, 2001)). Moreover the diusive ame front is identied by the isosurface:
Z (x, t) = Zst
where Zst is the stoichiometric mixture fraction value.
Yk ui Yk Z Z + Yk + + + ui t t xi Z t xi xi
continuity Zequation
Z xi
(5.25)
Z Z xi xi
2 Yk = k Z 2
86
Chapter 5.
The two terms within brackets in the LHS vanish because of continuity and mixture fraction equation (Eq. 5.23) so we obtain the amelet equation as:
Yk = D t
Z Z xi xi
2 Yk 1 2 Yk + k = + k Z 2 2 Z 2
(5.26)
Where has been introduced the scalar dissipation rate (Eq. 5.27) that has the dimension of an inverse of time (like strain) and can be interpreted as the inverse of a characteristic diusion time ((Peters, 2000)).
= 2D
Z Z xi xi
(5.27)
It measures the Z -gradient and the molecular uxes of species towards the ame. Standing to (Poinsot and Veynante, 2001) it is directly inuenced by strain, growing aside with this; moreover it gives an estimate of mixing layer thickness by D/. In the assumption of steady amelet, always true in present work, the species reactions rates can be written as:
k =
1 2
mixing
2 Yk Z 2
reaction
(5.28)
that shows how diusive ame reaction rate only depends on Z . At this point, stated that the mixture fraction appears to be the basic variable for most theories of laminar diusive ames, the modelling of this kind of ames can be decoupled into two problems.
In diusion combustion, in contrast to premixed combustion, there is no characteristic velocity scale such as the laminar burning velocity, and therefore
87
Chapter 5.
there is no ame thickness dening a characteristic length scale. There exists the strain rate a, locally imposed by the ow eld which is the inverse of a characteristic time. This can be used together with the diusion coecient Dst to dene a diusion thickness:
ld =
Dst a
1/2
There also exists the gradient of the mixture fraction eld | Z| which relates the thickness ld , dened in physical space, to a diusion thickness (Z)F in mixture fraction space:
(Z)F = | Z| ld
that, using the denition of the scalar dissipation rate (Eq. 5.27), becomes:
(Z)F =
st 2a
1/2
In addition to the (Z)F it is possible to dene a reaction zone thickness (Fig. 5.3) as the sum of the thicknesses of fuel consumption layer and oxidation layer both dened in Z -space, even if the last one is the much thicker than the rst. Figure 5.5 (log-log diagram) reports the Peters' combustion regimes diagram for diusion ames where (Z)F and (Z)R represent the just dened thicknesses of diusion zone and reaction zones. The mixture fraction uctuation Z (Zst stand for its value at stoichiometric conditions) is dened as the r.m.s. of mixture fraction variance: Z = Z 2 . With q and st are indicated scalar dissipation rates (Eq. 5.27, (Peters, 2000)) at quenching and stoichiometric conditions (Favre averaged). For large mixture fraction uctuations, where Z > (Z)F , uctuations in mixture fraction space extend to suciently lean and rich mixture such that the diusion layers surrounding the reaction zone are separated, this is the separated amelets zone. For small mixture fraction uctuations, where Z < (Z)F , which may be due to intense mixing or to partial premixing of the fuel stream, the reaction zones are connected, then we are in connected amelets zone. The line of equation Zst / (Z)R = 1 represents those uctuations that are equal to the thickness of the reaction zone. Moreover for q < st the ame is quenched. Also in gure is reported the shape of a ame that shows how local conditions, along the contour of mean stoichiometric mixture in a jet diusion ame, would t into dierent regimes. This diagram has to be used with great care ((Poinsot and Veynante, 2001)) since the local ame thickness and speed depend on the local ow conditions such as local strain rates and may be aected by unsteady eects. In a given combustor, the ame structure may strongly depend on the spatial location. Moreover some strong approximation have been introduced in its derivation.
1/2
88
Chapter 5.
The mixing problem has to provide the Favre's average of mixture fraction Z (xi , t) (Eq. 5.33) A ame structure problem where species mass fractions Yk or reaction rates k are expressed as function of mixture fraction. These functions are formulated as conditional expressions (Yk |z , k |z ), because of the inuence of multiple parameters in a turbulent ow.
To determine the average values of Yk (by which can be obtained all other variables to solve Navier-Stokes Equations, subsec. 5.2) or of k , the mean value of Z is no longer sucient: higher Z moments are need and, if possible, a probability density function (pdf ) of Z . When the pdf of Z , p(Z) is known, average species mass fraction (Yk ) or average reactions rate ( k = k ) are given by:
1
Yk =
0 1
Yk |Z p (Z ) dZ k |Z p (Z ) dZ
(5.29) (5.30)
k =
0
where Q|Z denotes the conditional average of quantity Q for a given value of the mixture fraction Z = Z , depending on Z and various quantities such as the scalar dissipation rate. p (Z ) is the Z -pdf that complies with:
1 0
p (Z ) dZ = 1
In RANS of diusion ames, the solution of species conservation can be obtained by two families of method:
89
Chapter 5.
Reaction rate approach implies the solution of species mass fraction conservations equations (Eq. 5.1) that requires the reaction rates k . These are provided by libraries built with Eq. 5.30 or by the Eddy Diusivity Concept model that assumes reaction rates to be function of species mass fraction and of turbulent mixing ((Magnussen and Mjertag, 1976)).
Yk =
0 0
(5.31)
where p (Z, K) is the pdf of mixture fraction and strain rate that are usually considered statistically independent:
p (Z) =
With:
Z 1 (1 Z) () ()
( + )
(x) =
0
et tx1 dt
= Z
= 1Z
and
Z 1Z Z
2
10
Z Z
(5.32)
90
Chapter 5.
The pdf for the stretching is usually the Dirac-delta function, assuming that the strain rate does not change along the ame front:
p (K) = K K
Once that pdf s have been dened, solutions of Eq. 5.31 can be computed and by those hypothesis presented in section 5.2, we can build amelet libraries needed for the solution of the containing the variables Q = Q Z, Z 2 , K RANS and consequently of the mixing problem. In order to dene the ame structure in space is required the evaluation of mixture fraction Favre's average, mixture fraction variance and strain rate elds. Strain rate can be calculated as function of turbulence variables, while for mixture fraction Favre's average and variance are needed proper transport equations that are solved together with RANS equations:
Z + ui Z = t xi xi
Z ui Z xi
(5.33)
where the turbulent transport term ui Z approach used for RANS (section 5.1):
ui Z =
T Z ScZ xi
(5.34)
ScZ is the Schimdt turbulent number for Z and T is the turbulent viscosity. By this assumption the closed Favre averaged mixture fraction equation form is: Z + ui Z = t xi xi D + T ScZ Z xi
(5.35)
91
Chapter 5.
Z t xi Z 2 xi
ui Z xi xi
ui Z xi
(5.36)
turbulent transport
+ 2Z
Z xi
2ui Z
Z Z Z 2D xi xi xi
dissipation
production
In this equation appear many unclosed term that can be modelled as follow ((Poinsot and Veynante, 2001)):
T ScZ
Z 2 xi
Molecular diusion terms can be usually neglected against turbulent transport. The unknown contribution in the production term ui Z is modelled using Eq. 5.34: ui Z Z T Z Z = xi ScZ xi xi
The dissipation term requires more assumptions: it can be meant to be the scalar dissipation rate of mixture fraction uctuations Z , and it can be assumed to play the same role as the dissipation of turbulent kinetic energy with the velocity eld. It can be modelled as: 2D Z Z = c Z xi xi k
2
where c is a model constant. Having modelled all these terms the mixture fraction variance equation can be rewritten as:
Z t + T ScZ
ui Z xi
=
2
(5.37)
Z 2 T Z Z + c Z xi ScZ xi xi k
92
Chapter 5.
of the reaction rate, the gas behind the ame front (Fig. 5.7) rapidly approaches the burnt state close to chemical equilibrium, while the mixture in front of the ame typically remains in the unburnt state. Therefore the combustion system on the whole contains two stable states, the unburnt and the burnt gases state. In premixed combustion both states exist in the system at the same time; they are spatially separated by the ame front where the transition from one to the other takes place. In present chapter are presented the most common turbulent premixed combustion models, with particular attention to the used Bray Moss Libby one. Anyway each model is suitable to be used in particular conditions so a combustion regimes classication is needed.
93
Chapter 5.
u 3 (r) r
being u = 2/3 k representative of eddies rotational speed. Similarly a turbulent mixing characteristic time of eddies of dimension r can be dened as:
m =
This time scale can be compared to the one characteristic of a ame (of thickness 0 0 L and with speed SL ) propagating in the considered (L = L /SL ), by the denition of the Damkhler nondimensional number:
Da(r) =
m (r) L
The correct denition of the Damkhler's number, Da (Eq. 5.38), is based r = LT , at which correspond m (LT ) = T = k/ (being k = RM S(u )).
Da =
T L
(5.38)
Da =
0 LT S L L u
(5.39)
Damkhler's number allows to determine two limit conditions. If Da >> 1 chemical reactions have a much higher speed than turbulent mixing, giving a just lightly distorted ame front but still laminar in its inner structure. This is the so called amelet regime, that allows to treat describe the combustion with single step innitely fast chemistry.
94
Chapter 5.
In the opposite case, with Da << 1 reactions are slower than turbulent mixing. In this situation the combustion system can be described with the Perfectly Stirred Reactor hypothesis ((Andreini, 2004)), that requires the Arrhenius chemical model. In order to describe the interaction between ame front and eddies having dimension comparable to Kolmogorov length scale Lk (small eddies), is introduced the Karlovitz number: 1 = L Ka = (5.40) Da(Lk ) k that can be rewritten as:
1/2 0 SL u 3/2
Ka =
that implies:
LT L
(5.41)
Ka =
L Lk
(5.42)
The Karlovitz's number is the proportional to the ratio between ame front
Figure 5.8: Schematic representation of both amelet regimes. If Ka < 1 amelets are
corrugated (a) while for 1 < Ka < 100 amelets are thickened (b)
thickness and small eddies characteristic length. The condition Ka = 1 is the Klimov-Williams criterion delimiting the amelet combustion regime. For Ka < 1 also smaller eddies are greater than ame front thickness that completely keeps its laminar structure. Finally, for Ka > 1 Lk is comparable with ame thickness and then small eddies start to interact with laminar ame inner structure. For a clearer description of this phenomena is convenient, as proposed by (Peters, 2000), to introduce a Karlovitz's number in function of the ame inner layer thickness (IN ):
KaIN =
IN Lk
(5.43)
95
Chapter 5.
KaIN
Ka 100
(5.44)
Until Ka < 100 (or KaIN < 1), the inner layer is not aected by eddies action, keeping its laminar structure. Then the introduction of the inner layer Karlovitz's number and the consequent more detailed analysis, allow to extend the amelet regimes to be within 1 < Ka < 100 where small eddies have a smaller thickness than ame diusion thickness L , having then the ability to penetrate in the preheat zone (Fig. 5.8) but having a smaller dimension than reactions zone (inner layer and oxidation layer ) that practically keeps a structure similar to a laminar ame one. The combustion regime in the range 1 < Ka < 100 is called Thin Reaction Zone and we are dealing with ThickenedWrinkled Flamelets ((Peters, 2000)). These two regimes are reported in Fig. 5.8.
96
Chapter 5.
front that would implies an interaction (between ame and turbulence). Flame front practically propagates in laminar way.
0 u /SL > 1: corrugated amelets. Aside the rising of turbulent uctuation u , turbulent motion structures are able to corrugate the ame front. Interactions between eddies and ame front are now present: can be spotted pocket of burnt and fresh gases
1 < Ka < 100: thin reaction zones (thickened wrinkled amelets) as already hinted, smaller eddies dimensions (Lk ) allow uctuations to penetrate the laminar ame preheat zone, anyway without modifying the reaction zone that keeps its well dened shape and small thickness, as well as in laminar regime. The overall eect is that the ame front is thickened and still subject to the wrinkling action of large eddies (of dimension LT )
Kolmogorov length scale is small enough that the whole ame laminar structure is subject to turbulent uctuations actions: eddies are even smaller than reaction zone and then no longer exists a laminar structure. Flame is near the Perfectly Stirred Reactor state where the mixing process can be considered innitely fast, then the process is completely controlled by chemical reactions. Practical realizations of combustion systems operating in the Distributed Reaction Zone are prevented by quenching phenomenon, distinguishing this regime. There exist many systems working near the border between Thin Reaction Zone and Distributed Reaction Zone such as gas turbine combustors, indeed getting nearer to the Distributed Reaction Zone allows to limit the reaction zone and the of the whole apparatus.
97
Chapter 5.
c=
T Tu Tb Tu
c=
(5.45)
where the subscript u indicates the unburnt gas state, and b the burnt one, clearly the progress variable c is equal to 1 in the burnt gases ow, and to 0 in the fresh gases ow. If species diusivities are assumed equal to corresponding thermal diusivities, and the specic heat (Cp ) is constant, these two denitions can be demonstrated to be analogous. In BML model is assumed an innitely fast chemistry with a single step (F + OP ) mechanism, implicitly assuming that Da 1. Innitely fast chemistry assumption implies that the amelet has got an innitesimal thickness and that its internal structure induces the instantaneous crossing from unburnt to burnt gas states. With these hypothesis the probability of having intermediate states is virtually null, so the pdf is made of two Dirac-delta ( ), one in c = 0 and the other in c = 1:
(5.46)
(x, t) + (x, t) = 1
where (x, t) and (x, t) are respectively the probability of having c = 1 and c = 0, in point x at time instant t. The BML model also assumes a constant pressure for the system (nearly true in gas turbine combustors) so the perfect gases equation of state (Eq. 5.7) can be expressed as follow:
u Tu = b Tb
Then is useful the denition of the heat release factor or expansion factor, as:
Tb u 1= 1 b Tu
(5.47)
= u + b = (1 ) u + b
Stated that
1
c = c =
0
c p(c) dc = b = b c
(5.48)
(1 c) c + u b
= b
1+ u = 1 + c 1 + c
98
Chapter 5.
1c 1 + c
(1 + ) c 1 + c
Using the found relation between Favre's and Reynolds average of c (Eq. 5.48), the Favre's average of generic variable Q is obtained by:
Q=
b u Q + Qb = (1 c) Qu + c Qb u
(5.49)
where the unburnt and burnt values are calculated by conditional averaging respect to corresponding progress variable values, being p (Q|c = c ) the pdf of Q if c = c :
+ +
Qu =
Q p (Q|c = 0) dQ
Qb =
Q p (Q|c = 1) dQ
Since scalar variables such as k th mass fraction are associated to corresponding value of laminar amelet that, for considered hypothesis only allow burnt and unburnt state, conditional averages of mass fraction is coincident with corresponding value in fresh gases and burnt gases stream so we can write:
Yk = (1 c) Yku + cYkb
(5.50)
that allows, by those hypothesis presented in section 5.2, to calculate the variables needed for the solution of the RANS. All these denition allows to state that BML model completely describes the turbulent ame by the solution of the Favre's averaged transport equation for c:
c ui (c) + (cui ) = t xi xi
+ c
(5.51)
that is performed aside RANS equations, once that models are provided for its two unclosed terms, turbulent transport c ui and source c .
Usually the turbulent transport is modelled by the common Boussinesq's hypothesis: c ui = T c ScT xi
(5.52)
This formulation can be a source of errors, indeed may occur that turbulent uctuation leads to diusion of c in opposite sense respect to the sign of progress variable gradient (Gradient Diusion, GD): this situation is called Counter Gradient Diusion (CGD) (a detailed explanation of this phenomenon can be found in (D.Veynante and L.Vervisch, 2002) and (Peters, 2000)).
99
Chapter 5.
ame surface area for each volume unit) and the ame front reaction speed L :
c = u L
(5.53)
An evaluation of L is obtained by the laminar ame speed of the 0 corresponding amelet (SL ), keeping in consideration the stretching. The simplest expression was proposed by Bray:
0 L = I0 SL
Where I0 is the laminar ame stretch factor needing an adjunct model. An expression for Flame Surface Density is the Bray one:
g c (1 c) y Ly
where g and y are model constant and Ly is the wrinkling length and is usually related to the turbulence integral scale LT . A more accurate model is the Coherent Flame Model that implies the solution of a transport equation for ((Andreini, 2004) and (Poinsot and Veynante, 2001))
c = u ST Fk | c|
(5.54)
where FK is a coecient that takes on account laminar ame stretching (0 < Fk < 1). The proposed ST expression is:
0 ST = Au 3/4 SL
1/2
DT
1/4
LT
1/4
(5.55)
where A is a constant (usually assumed equal to 0.5), u is the velocity 0 turbulent uctuation, SL is the unstretched laminar ame speed, DT is the turbulent diusivity and LT is the integral turbulence scale.
Kolmogorov Petroski Piskunov closure considers the progress variable source term as a function of c itself. It derivation is based on the asymptotic analysis of the progress variable equation (Eq. 5.51, for details on derivation procedure (Andreini, 2004)). The nal form is:
c = u CKP P
2 ST c (1 c) DT
(5.56)
where the turbulent ame speed is dened as function of laminar speed and Damkhler number:
0 ST = SL + u 1 + Da2 1/4
(5.57)
where u =
2/3 k .
100
Chapter 5.
G + t
0 U G = ST | G| DT | G|
(5.58)
The transport equation for passive scalar G is a conditional equation, as it is dened only on the ame front (G = G0 ): the additional condition of unicity of the gradient normally to the ame front has to be assured, so that G provides the Cartesian distance from the mean ame front at any point of the domain.
| G| = 1, G = G0
(5.59)
According to Mangani (2003), an experimentally veried Gaussian distribution has been adopted to model the Probability Density Function of G. An additional transport equation for the mean variance G 2 it's therefore required to close PDF function (Eqn. (5.60)).
G t
U G
= 2DT
DT G
2
+
2
cs G k
(5.60)
To ensure the condition of unicity for the PDF, G towards the ame front normal direction:
G=0
(5.61)
Turbulent ame speed ST is evaluated from the solution of an additional transport equation for the turbulent term T of the ame surface area ratio :
AT ST = = | G| + T = 1 + T = SL A ST = SL (1 + T )
(5.62)
101
Chapter 5.
Transport equation for T was adapted from Peters Peters (2000) and requires a re-distribution step identical to that described for G 2 .
T + (U T ) = DT T + t (u u ) : U DT ( G)2 2 c0 T + c1 T + 2 k G c2
0 SL T 2
(5.63)
(G 2 )1/2
c3
DT G
2
102
Chapter 5.
In present work has been chosen to treat the partial premixed ame by coupling the presumed shape pdf amelet model, described in subsubsection 5.5.7, with the Bray-Moss-Libby model and the G -Equation models. The followed approach is provided by (D.Veynante and L.Vervisch, 2002): the idea is that burnt (indicated by the subscript b) and unburnt (indicated by the subscript u) states variables are no longer constant but are aected by mixing process. The pdf for BML model (Eq. 5.46) has to be rewritten as, also, function of mixture fraction Z . In the hypothesis of validity of amelet approach it becomes:
(5.64)
where pu (Z, x, t), pb (Z, x, t) are respectively the unburnt and burnt pdf for diusion ames (Eq. 5.31). The Favre's average for the generic variable Q (Eq. 5.49) has to be rewritten in terms of Qu (Z, K) and Qb (Z, K), stated that now:
1 1
Qu =
Q p (Z|c = 0) dZ
0
Qb =
Q p (Z|c = 1) dZ
0
(5.65)
Q = (1 c) Qu (Z) + c Qb (Z)
(5.66)
for the G -Equation model the probability c is readily determined by integrating the normal distribution of scalar G, P (G; x), across ame front:
G0
cu =
P (G, x)dG
(5.67)
In this way c has the same meaning of a progress variable and it assumes the value 0 in burned gas and 1 in unburned gas.
103
Chapter 5.
Only these modications are carried out, and all other denition and assumption of BML G -Equation model are still valid. Finally for both models, the dependence on mixture fraction has to be considered also in source terms calculations that requires the unburnt state density and the laminar ame speed that now are both functions of the mixture fraction.
104
Chapter 6
6.3
6.4
6.5
6.6
105
Chapter 6.
106
Chapter 6.
6.2 Denitions
In this section will be dened those quantities that describe radiative heat transfer, further details can be found in (Viskanta, 2005) and (Modest, 2003).
The fundamental quantity that governs the radiation eld is the spectral intensity of radiation or spectral radiance. Considering e+ (r, t + t) e (r, t) as the amount of energy in the wave length interval and + transported across an arbitrary oriented imagi-
107
Chapter 6.
nary element of area A at a point P (of coordinates (r, Fig. 6.1) during the time interval (t + t) t and conned to an element of solid angle about the direction s. The vertex of the elementary cone is on the surface element A and the outward normal n the surface makes an angle with the direction s. The limit in Eq. 6.1 tend to a nite value, as it has been demonstrated experimentally, provided that the point P and the direction s are kept constant. The value obtained by this limit is the spectral radiative intensity. The integration of spectral radiance on the whole wave length spectrum (Eq. 6.2) leads to the denition of total radiative intensity I(r, s, t).
I (r, s, t) =
A,,,t0
lim
(6.1)
I(r, s, t) =
0
I (r, s, t) d
(6.2)
6.2.2 Irradiance
Irradiance is dened as the rate of radiant energy incident on a volume element per unit area, so it is obtained by integrating the spectral intensity on the whole solid angle as in the equation:
G (r, t) =
4
I (r, s, t) d
(6.3)
To calculate the total irradiance an integration of the irradiance on the whole wave length spectrum has to be performed as:
G(r, t) =
0
G (r, t) d =
4
I (r, s, t) d
(6.4)
where has been used the denition of total radiative intensity (Eq. 6.2).
qR, =
I (r, s, t) s d
4
(6.5)
qR =
qR, d =
I(r, s, t) s d
4
(6.6)
108
Chapter 6.
where is the Planck's constant. In his work Planck found that the spectral emissivity power for a black surface (of normal n) bounded by a transparent media with refractive index m is:
eb (T, t) =
2
Ib (T, t) s n d =
2hc2 0 m2 5 [ehc0 kT /m 1]
(6.7)
where k = 1.3807 1023 J/K is the Boltzmann's constant and c0 is light speed in vacuum. The total black body emissivity power is obtained by the integration on the whole spectrum of the spectral emissivity power as follow:
eb (T, t) =
0
5 4
eb (T, t) d = n2 T 4
(6.8)
2 where = 15h3kc2 = 5.670 108 W/m2 K 4 is the Stefan-Boltzmann constant. 0 Dened the spectral black body emissivity power, the denition of the black body spectral irradiance is easily obtained as (by the rst two terms of Eq. 6.7): Ib (T, t) = eb (T, t)/ , in which has to be pointed out that black body irradiance is isotropic so the dependence on direction is neglected. Calculation of total black body radiative intensity can also be performed with Eq. 6.2 that becomes:
Ib (T, t) =
0
Ib (T, t) d =
0
eb (T, t) n2 T 4 d =
(6.9)
6.3.1 Emittance
The rate of radiant energy emission per unit area from an isothermal body at temperature T into the solid angle about the direction s in the wavelength interval and + d is given by
(6.10)
where Ie is the spectral intensity of emitted radiation. It is possible (and usually done) to express the energy truly emitted by a surface as proportional to the maximum possible value that any surface would emit, if it were in thermodynamic equilibrium at temperature T, that is the black body emission (Eq. 6.7):
de (s, T ) =
(6.11)
(6.12)
109
Chapter 6.
is the spectral directional emittance. Similarly can be dened the spectral hemispherical emittance as (T ) = I (T ) I(T ) Ib (T ) and the total hemispherical emittance as (T ) = Ib (T ) . If the spectral emittance is the same for all wavelengths the surface is dened as a grey body and if the emittance is the same for all directions the surface is a grey diusive one that is the kind of surface used in this work.
6.3.2 Absorbance
Similar considerations to those used in previous section can be used for the denition of absorbance that represents the proportionality factor between the energy amount (coming from direction s) absorbed from a surface and the amount that would be absorbed by a black surface, that absorbs the radiation completely. Thermodynamical arguments has lead to the knowledge that absorptivity is equal, in value, to the emittance, so in most general case can be written that
(s, T ) =
(s, T )
(6.13)
If a surface is grey diusely emitting one it is also grey diusely absorbing and then:
(T ) = (T )
(6.14)
6.3.3 Reectance
The reectance of a surface is the rate of incoming radiation that leave the surface without being emitted or transmitted. By its denition it depends on two direction: the incoming radiation one and the outgoing one, and the law of reciprocity says that the directions can be exchanged without varying the value of reectance . Energy balance for an opaque (only emitting, absorbing and reecting) diusely grey absorbing-emitting surface implies that: (T ) = 1 (T ) = 1 (T ).
110
Chapter 6.
radiation crossing a unit of volume without being absorbed, emitted or reected (scattered). The phenomenon described below can be justied by the solution of the Maxwell's equations set.
6.4.1.1 Absorption
Experimental experiences evidence that the amount of radiation, expressed as radiative intensity, absorbed while crossing a participating media is proportional to radiative intensity itself and to the distance the radiation beam travels through the media. This statement allows to write that
(dI )abs = k I ds
(6.15)
where the proportionality constant k is the absorption coecient and negative sign is used because absorption induces a radiative intensity decrease.
6.4.1.2 Emission
As already reported in introduction a participating media can also emit radiation. Considerations similar to those used for surfaces, dealing with black body emission and simply emission, has lead to dene the rate of emission as proportional to the emission of a black body, at the same temperature of the media and, as for absorption, proportional to the length of beam path.
(dI )em = k Ib ds
(6.16)
where the proportionality constant is again equal to the absorption coecient due to, as for surfaces, thermodynamical guesses.
6.4.1.3 Scattering
Scattering is a phenomenon that involves direction changes (see gure 6.2) of radiation beams then it represents an attenuation (similar to absorption) for a direction and at the same time it represents an augmentation for some other directions. Since its modelling is complex and in present work can be neglected because either gases molecules or soot particles are much smaller than wavelengths of interest (Rayleigh's theory) here are not reported further detail that can be found in (Modest, 2003) and (Viskanta, 2005).
111
Chapter 6.
dI = k Ib k I (6.17) ds This equation is Lagrangian in nature since it is written as we were following a ray travelling at speed of light c, hence the length of the path can be expressed as ds = c dt, and transformed in an Eulerian coordinates the RTE is: 1 I 1 dI = +s c dt c t I = k I + k Ib
(6.18)
The unsteady term I can be usually dropped stated that the speed of t light is much greater than those typical of uid ows so that the radiative phenomenon can be considered instantaneous. Even if an explicit relationship is not reported, each quantity may be depending on location in space, time and wave length while the intensity of radiation is also function of direction. Notwithstanding its complexity this relation is built on strong simplications ((Modest, 2003)) since the media is considered:
at rest (comparing its speed the light one) non polarizing in local thermodynamical equilibrium with constant index of reection non-scattering
(rw )
I (rw , s ) n s d
ns <0
(6.19)
This kind of boundary condition is, unfortunately, dependant on domain value of radiative intensities, except for black surfaces that emit independently from direction and incoming radiation while it is dependant only on temperature.
112
Chapter 6.
=
0
k ds
where s is the beam path length, normally it is more useful to write it in term of a system reference length (L):
L
=
0
k ds
(6.20)
113
Chapter 6.
excitement of electrons. So a proper estimation of absorption coecient is hard to achieve since it will be highly uneven in the spectrum, and interaction with surrounding molecules can induce modications on energy levels, introducing another source of uncertainty. Then if a surface can be easily and correctly represented by a grey body, a gas cannot be. Even with great diculties and approximation many models have been realized for spectral treatment of gases, and can be divided in four groups:
Narrow band calculations based on the observation that absorption coecient has rapid change across the spectrum, especially when compared to those of black body emission. Then can be used a proper average value of absorption coecient on small wave length intervals where black body emission is constant. Theoretically an average that exactly represents the line-by-line absorption coecient can be realized but practically, especially when gases mixture are involved, it cannot be done. Anyway modern improvements, and simplications, of this model are obtained by noting that same values of absorption coecient are repeated in the spectrum, taking in account this repetitive behaviour has been realized the correlated kdistribution model ((Zhanga and Shib, 2005)).
Global calculations represent a wide group of method since it has had many
development being the most suitable family for analysis of combustion systems. In these systems the radiative heat ux or its divergence are the most interesting variable, and indeed global calculations are aimed to give a good prediction of them. Most important models of this category are: - Planck mean absorption (further details are reported below because this is the model used in present work) - Weighted Sum of Grey Gases ((Trivic, 2004)) that replace the absorption coecient of the non grey gas with the weighted sum of grey gases absorption coecients - Spectral line based weighted sum of grey and Absorption Distribution Function both based on previous model and on the use of spectral line databases to calculate grey gases absorption coecients.
114
Chapter 6.
- Full spectrum kdistribution based uses the narrow band k distribution on the whole spectrum by the use of fractional Planck function As already stated, in present work has been used the Planck mean absorption method that denes the absorption coecient as:
k=
k Ib d = T 4 I d b
k Ib d
(6.21)
To calculate the global absorption coecient has been used a correlation by International Workshop on Measurement and computation of Nonpremixed ames ((Da Soghe, 2006, pages 101-102)). In this correlation the absorption coecient is dependant on CO2 and H2 O partial pressures, and is dependant on the reciprocal of temperature by a polynomial series. Further details on spectral treatment of gases can be found in (Viskanta, 2005) and in (Modest, 2003).
k,soot = C0
fv
(6.22)
where fv is the soot volume fraction (Eq. 10.3), and C0 is a constant dependant on complex index of refraction. The total absorption coecient is dened as:
ksoot = 3.83fv C0
T C2
(6.23)
115
Chapter 6.
Since soot is usually present together with gaseous media, contributions to absorption coecient from both have to be considered. In present work, the mixture absorption coecient has been as the sum of gases absorption coecient (see previous section) and the soot one:
(6.24)
Optically Thin Approximation based on the observation that in the optically thin limit ( 0), the media absorbs radiant energy leaving the boundaries as well as emits itself but is not optically dense enough to receive any from other elements of volume, and the system is said to be emission dominated. Then irradiance, appearing in the heat sink due to radiative transfer (Eq. 6.46), is calculated by Eq. 6.3 considering only radiative intensity leaving boundaries.
116
Chapter 6.
Keeping in consideration the classication from Viskanta, Self-Absorbing Medium methods can be divided in four groups: 1. Directional Averaging 2. Dierential Approximation 3. Energy Balance 4. Hybrid All directional approximation methods involve some type of averaging of the radiative intensity eld with directions. The multiux methods are probably the simplest to apply and still are being used, to model radiative transfer in combustion systems, even if they suer from poor accuracy. Also DOM belongs to this group and are one the most used and accurate method used together with CFD. The rst order Dierential Approximations of the RTE yield predictions of reasonable accuracy and can be adopted for spectral and band calculations. Energy Balance Methods, such as the Zone Method, the statistical Monte Carlo Method and stochastic approaches are incompatible with CFD, due to their heaviness coming from their high accuracy. Anyway the Finite Volume Method is a good compromise between accuracy and suitability to be used with CFD, and being one of the used method, in present work, it is discussed in subsection 6.5.5. Hybrid Methods have advantages and disadvantages of the method from which they are derived, to this group belongs the P1-approximation that is the other method used in this work and further detail are in subsection 6.5.4.
k =
k kref
1 kref xref
= xref
(6.25)
where kref , xref and are respectively the reference absorption coecient, the reference leght and the reciprocal of optical thickness.
= k (I Ib )
(6.26)
1+
I = Ib
(6.27)
117
Chapter 6.
(1 a)j =
j=1
aj
(6.28)
Then the radiative intensity can be calculated by inverting Eq. 6.27 using the Newmann's series (Eq. 6.28):
I = 1 k s
1+
i
s s
Ib Ib
(6.29)
2
2 k
+ (1)
i
i k
Integrating Eq. 6.29 on the whole solid angle and stated that
s
4
d = 1 + (1)
2 i+1
(6.30)
G = 4 1 +
2
2 3k
4
4 5k
6
6 7k
+
i
(6.31)
+ 1 + (1)
2 4(i + 1)
Ib
Inverting Eq. 6.31 and using again the Newmann's series (Eq. 6.28) we obtain the approximated Simplied Spherical Armonics Equation (SPn):
4Ib =
2
2 3k
42
4 45k
446
6 945k
G + O(8 )
(6.32)
If the series on RHS of Eq. 6.32 were truncated at rst order we can write the SP1 equation in dimensional form as:
1 G = k (G 4Ib ) 3k
(6.33)
the SP1, is an approximated form of RTE with and error comparable to O(4 ), then it is easy to understand that this method is suitable for optically thick media ( 0). Also the RTE boundary condition has to be modied to be used with SP1, this was done by Marshak ((Modest, 2003, pages 475-476)) for a diusive emittingreecting surface. The boundary condition for SP1 is:
1 3k
G + G = 4Ib (Tw ).
(6.34)
As already stated, this equation is valid for diusive surfaces and it could not be in a dierent way since the SP1 neglect directional behaviour of radiative transfer.
118
Chapter 6.
In present work has been derived the SP1 form of RTE that was proposed by (Gelbard, 1961) for nuclear reactor modelling, and that is equal to spherical armonics P1 approximation and in eect it is better known with this last name. In this section has been described the SP1 asymptotic derivation, reported in (Larsen et al., 2002), due to its simple form, at least compared to the Pn derivation, involving Legendre's series ((Modest, 2003). A result of SP1 approximation, that instead represents an hypothesis for the derivation of P1 approximation, is that heat ux vector can be calculated as the gradient of irradiance (Eq. 6.35). This can be demonstrated comparing the RHS s of divergence of radiative heat ux vector equation (6.46) and of SP1 equation (6.33) it can be found out that they are equal, and it is, obviously, correct also for LHS s.
qR, =
1 3k
G .
(6.35)
s
V
I dV =
V
(sI) dV
V
s dV =
=0
I s n d
(6.36)
where the second term of RHS is dropped since the direction versor s is constant in space ((Modest, 2003)), and n represents the normal to . Then integrating RTE over the control volume and solid angle i leads to:
Is n d d =
i i V
kIb dV d
i V
kI dV d
(6.37)
In the simplest implementation of the nite volume method it is assumed, for the LHS of Eq. 6.37, that the intensity is constant across each face of the discrete control volume (leading to problems reported in chap. 9.2) as well as over the solid angle. While RHS components are considered constant in the whole control volume and equal to value assumed in the control volume centroid. By this hypothesis, equal to those of nite volume method applied to CFD, Eq. 6.37 becomes:
(6.38)
119
Chapter 6.
where, Iki is the radiative intensity through face k , of area Ak , in the i th direction, while Ibpi , Ipi and kp represent respectively the centroid black body radiative intensity, the radiative intensity in i th direction and the centroid absorption coecient. Di is a versor pointing into an average direction within solid angle element i and is dened as:
Di =
i
s d
(6.39)
Until this point the method is the same of DOM, and the discrimination between these two methods lies on the treatment of term 6.39: in DOM it is discretized with approximated quadrature (not ensuring energy conservation) while in FVM it is discretized with exact relations. In present work has been followed the method suggested by (Mishra and Roy, 2007).
(6.40)
i = i
i i i i , i + i , i + 2 2 2 2
(6.41)
After this modications, the average direction versor for nite volume method can be rewritten as:
i +
Di =
i
i 2
i + i
i 2
i 2
i 2
s sin d d
(6.42)
Once that a numerical form for RTE has been obtained it is now needed a numerical formulation for boundary conditions. For a grey diusive emittingabsorbing-reecting opaque surface this means discretizing Eq. 6.19. It can be
120
Chapter 6.
do by multiplying Eq. 6.19 by s n and integrating over all outgoing directions and considered that ns<0 n s d = we obtain:
I n s d =
ns >0
w Ibw ns >0
n s d + w
ns <0
I|n s|d
(6.43)
Then by those approximations applied to discretize the RTE Eq. 6.43 for face, q , becomes:
Iq0
i,out
(6.44)
And after some rearrangements the value of radiative intensity for boundary face q becomes:
Iq0 =
q Ibq
+ (1
q)
i,out Iqi
Di nq nq )
i,in (Di
(6.45)
qR, =
I (r, s) s d =
4
s Ib d
I d =
(6.46)
k
4
I d + k
4
By some calculation, the use of irradiance denition (Eq. 6.3) and considering that a black body emits in the same way in all direction (subsec. 6.2.4), the divergence of the radiative heat ux vector becomes:
qR, = k 4Ib
I d
4
= k (4Ib G )
(6.47)
The total form of this equation, that is the one to be used in Eq. 5.3, can be obtained by integrating over the all wave length spectrum:
qR =
qR, d = k 4Ib
Id
4
= k (4Ib G)
(6.48)
121
Chapter 7
7.4
7.5
. . . . . . 153
Trailing Edge Cooling System: AITEB2 . . . . . . . 153 7.5.1.1 Geometry G1.0 . . . . . . . . . . . . . . . . 158 7.5.1.2 Geometry G1.2A and geometry comparison 159
123
Chapter 7.
7.1 Generalities
Before starting the analysis of the obtained results, not to weigh down the treatment of the subject, some features common to all simulations will be discussed in this section. Due to the huge number of implemented turbulence models, a shortcut has been used to name most of them: the acronyms before dened and summarized in Tab. 7.1 are widely used in substitution of authors' full name.
Table 7.1: Acronyms for the various turbulence models.
k Low Reynolds by Abe et al. (1994) k Low Reynolds by Chien (1982) k Low Reynolds by Chen et al. (1996) k Low Reynolds by Hwang and Lin (1998) k Low Reynolds by Lam and Bremhorst (1981) k Low Reynolds by Lien and Leschziner (1993) k Low Reynolds by Lien (1992)
Realizability constraint postscript Two Layer k SST k SST Automatic Wall Treatment
All cases are run with the second order advection scheme SFCD described in Sec. 1.4 with a deferred approach mixing implicit and explicit derivatives:
(7.1)
to obtain blended second order accuracy at convergence, U DS,n = U DS,n1 , e e but convergence properties typical of upwind schemes. Even if case dependent, it is preferred to give a general indication for the values of suggested underrelaxation factors of the various quantities in Tab. 7.2. Convergence criteria for the inner cycle are imposed as a relative tolerance, ratio
Table 7.2: Typical values for the underrelaxation factors used.
Velocity Enthalpy Turbulent kinetic energy Turbulent dissipation Turbulent specic dissipation Turbulent viscosity Pressure corrector Density
mix 0.5 0.7 0.5 0.7 0.4 0.6 0.4 0.6 0.6 0.8 0.8 1.0 0.005 0.01 0.1 0.9 1.0
of nal residual on initial residual, of [103 105 ] for all quantities but pressure corrector that can be solved with a lower accuracy. For the outer iteration a good convergence criterion was considered initial residual for all quantities equal to 106 . However not always has been possible to obtain such levels of convergence for all quantities.
124
Chapter 7.
U k
= = =
M a RT , p , RT 3 U 2 T u2 , 2 3 k2 . lt
The plate has been considered adiabatic and the inlet temperature was Tin = 294.4 K . Details about applied boundary conditions are listed in Tab. 7.3.
Table 7.3: Flat plate - Computational boundary conditions.
Inlet temperature Inlet velocity Outlet pressure Inlet turbulent kinetic energy - k Inlet turbulent dissipation -
294.4 68.9 101400 23.6 3365 ms1 m2 s2 m2 s3
Pa
Near wall behavior of both turbulent kinetic energy and turbulent kinetic energy dissipation has been properly checked, through the boundary layer up to a very little minimum y + . Due to the relative simplicity of the case, runs could be performed on a very ne grid: rst node y + 0.1. The total number of cells for the mesh was 17641. To drop dimensions from the turbulent kinetic
125
Chapter 7.
k+ +
where:
= =
k , u2 , u4
(7.6) (7.7)
u =
(7.8)
and w is the proper wall shear stress. It is better underline the world proper to point out the dierences between Low Reynolds and wall functions treatment. Shear stress is in fact directly computed with velocity wall normal gradient:
w = w = T n [(T n) n] n ,
(7.9)
where T is the Reynolds stress tensor dened in Eq. 4.20, calculated at the wall. The abscissa the non-dimensional wall distance:
y+ =
u y .
(7.10)
Both the non dimensional k and were checked in their behavior against the
8 7 6 5 4 3 2 1 0 0 10 20 30 40 50 60 70 80 90 100
0.05
exp. AKN CH CLL HW LB LW LNR
0.30
0.25
1/(xy)
+
0.20
CH CLL HW LB
0.15
+
LW LNR
0.10
0.00 0 20 40 60 80 100
(a) k+ prole
(b) + prole
non dimensional wall distance at an axial location where the ow can be considered fully developed, Fig. 7.1. The level of agreement with the experimental is fairly good, especially for k + , with free stream values matching for almost all models. However the discrepancy between experimental and numerical simulation follows exactly the proles obtained by other researchers some of which are directly models' authors Patel et al. (1985); Chien (1982); Hwang and Lin (1998). Both for a higher guarantee of stability and for the best reliability on matching experimental results, a further test was conducted only on CLL and AKN. For the same geometry another case involving heat transfer was set up to validate mesh independence of the k SST with the automatic wall treatment.
126
Chapter 7.
0.0040
0.0035
0.0030
Crawford y
+
= 10 = 10
-6 -4 -6
0.0025
y y
= 5x10
0.0020
0.0015
0.0010
0.0005
0.0000 0 1 2 3 4 5 6
X [m]
Three dierent grids with increasing rst node distance from the wall were tested + + + (yp = 0.1; yp = 9; yp = 35). Results are reported in terms of Stanton number and tested against Incropera Incropera and DeWitt (2001), see Fig.7.2. Even if the three proles are not exactly coincident the mesh dependence is decreasing as the ow develops and moreover the agreement is greatly improved from standard wall function approach.
127
Chapter 7.
for the experimental campaign can be found in (Cooper et al., 1993) as well as in the ERCOFTAC online database. Geometry has been simplied in an
axial-symmetric jet. Correct velocity and turbulence conditions at the end of the pipe were obtained from the development of a 50D upstream extruded inlet hole, see Fig. 7.4 for details on the grid. The total number of cells for this case was 70000. The plate, heated at constant heat ux q = 200 W m2 , has been
(a) Total
Figure 7.4: ERCOFTAC - Geometry detail of the grid around the stagnation point.
computed using a xed gradient temperature wall (Neumann type boundary condition), while all other walls has been considered adiabatic, zero heat ux. Under these conditions the ow results almost incompressible. Only one test condition was reproduced for validation purposes, with ow conditions reported in Tab. 7.4. Results for this test are reported in terms of Nusselt number as a function of radial distance from the center. Nusselt number is calculated from the heat transfer coecient. Heat ux is calculated again using enthalpy wall
128
Chapter 7.
mm
Km1 ms1 m2 s2 m2 s3
Pa
gradient. Briey, equations to calculate N u from the known ow eld are:
q HT C
= =
Nu =
h , n q , T Tad HT C L , k
where HTC is the heat transfer coecient, Tad is the adiabatic wall temperature, for this low speed simulation approximately equal to inlet static temperature, L is the characteristic length: the diameter of the pipe, and k = Cp is the thermal conductivity. The turbulence models tested for this case are, after the selection over the at plate: CLL, CLL realizable, AKN, Two Layer, k SST, k SSTawt. The
450
400
Exp
350
CLL AKN
300
CLLReal SST
250
D
TL SSTawt
Nu
200
150
100
50
0 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0
r/D
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 (Durbin and M. Benhia, 1998), in the area around the stagnation point, k Low Reynolds models without the realizability constraint fail and
129
Chapter 7.
dramatically overpredict the peak in the heat transfer coecient (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 realizable CLL is well predicting such peak only shifting a bit towards higher values of the radius. Furthermore it is possible to compare the behavior of the SSTawt model with the other models. When the same Low Reynolds mesh is used its predictions are obviously collapsing on the standard SST model and agreement with the best performing models is good.
(e) ERCOFTAC - TL
Figure 7.6: Turbulent kinetic energy distribution on symmetry plane [m2 /s2 ].
For turbulent kinetic energy we can see in Figs. from 7.6(a) to 7.6(e) as turbulence models dier to each other: rst of all it may be noticed how two of the models, the realizable CLL and the k SST, have an almost laminar
130
Chapter 7.
core in the jet that is going up to the wall. So that both these two models limit production term in k equation when the rate of strain is high in this case near the stagnation point. It is shown nally how the k SST implicitly satisfy the realizability constraint proposed in Sec. 4.6.7 thanks to the limit on the eddy viscosity, see Eq. 4.118.
Main ow parameters can be seen in Tab. 7.5, these conditions have been applied at the computational domain as the boundary conditions presented in Tab. 7.6. Simulations have been validated in terms of heat transfer coecient calculated with respect to inlet static temperature circa coincident to inlet total temperature. Adiabatic simulations have been done too, in order to check whether this approximation could be done or not. Heat transfer coecient is plotted
131
Chapter 7.
kg m1 W m2
along the cutting hole symmetry line versus non-dimensional distance from the stagnation point.
700
Exp TL
600
SST CLLReal
500
HTC [Wm K ]
-1
-2
400
300
200
100
-3
-2
-1
X/D
Figure 7.8: Impingement single hole - Heat transfer coecient on impinged wall along
symmetry line.
First thing to notice from Fig. 7.8 is that, contrarily to ERCOFTAC test, realizable CLL model fails in well predicting heat transfer coecient 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. 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. From a designer point of view however, what is more important is a param-
132
Chapter 7.
eter that well represent some mean value of the heat transfer coecient. Averaging has not been done for this case so, to give an idea of the bi-dimensional distribution of HTC over the plate, map of wall heat transfer coecient are contrasted in Fig. 7.9. Scale of reference is the same used for the experimental results so it is obviously scaled on its maximum 375 and minimum value 50 [W m2 K 1 ].
(c) k- SST
[W/(m2 K)].
Figure 7.9: Impingement single hole - Heat transfer coecient distribution on impinged wall
It can be noticed how the shapes of these isolines dier: Two Layer presents almost circular contour lines while realizable CLL tends to have higher values downstream on the symmetry line at the same distance from the stagnation point. The SST model lies in between this two bounds. This tendency is however smoothed downstream. What is interesting is the bubble of higher HTC right on the hole-to-hole symmetry plane. This is due by the interaction of the jet with the wall conning the ow in addition at the jet to jet interaction. The eect of non constrained single jet in fact decreases with the distance but in this case practically the mass owed towards the right side, see Fig. 7.9, cannot be drained on that side because of the solid wall. Streamlines are thus curved by the wall and, not powerful enough to penetrate the jets, goes in between them. Crossow rounds the reference jet and increases the velocity near the symmetry plane. This results in an higher than expected HTC. In order to check which of the three distributions is more physical, probably comparisons are better done evaluating the ow eld on the symmetry plane: velocity and turbulent kinetic energy are reported in Fig. 7.10 and Fig. 7.11. Both velocity magnitude and k do not perfectly agree in terms of qualitative distributions, but if for velocity contour levels pretty much coincide, for turbulent kinetic energy this is not true, and realizable CLL presents a maximum value more than two times lower with respect to the other two models. It has already been underlined in Sec. 7.3.1, that also k SST uses a sort of realizability constraint, in this case though, the low k zone is much less developed
133
Chapter 7.
(c) k- SST
[m/s].
Figure 7.10: Impingement single hole - Velocity magnitude on cutting hole symmetry plane
and values stand in the middle between the Two Layer and the realizable CLL. It is interesting to see how the really low values of viscosity inside the potential core of the k Low Reynolds model, increase velocity and the higher heat transfer coecient is easily explained. Very important in validation process for new CFD packages is the testing against equivalent models released in commercial codes. A very wide investigation of the same geometry has been done with StarCD by Di Carmine (2004). Comparisons between the results are in perfect agreement one with the other in terms of a qualitative analysis. Quantitatively speaking the dierences between the two results are however more than acceptable with maximum errors for the heat transfer coecient, on peak values, of the order of 5%.
134
Chapter 7.
(c) k- SST
Figure 7.11: Impingement single hole - Turbulent kinetic energy on cutting hole symmetry
only the Two Layer and k - SST models have been tested against experimental results, again in terms of heat transfer coecient on the impinged wall. First the heat transfer coecient on the symmetry line is plotted as a function of the non-dimensional distance from the rst hole stagnation point. However in this case, there are two cutting holes symmetry lines so both experimental and numerical data are sampled onto the two dierent symmetry lines (path1 and path2) and then merged together in the zone where a relative minimum is located, see Fig. 7.13. 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 values are quite dierent from measured data. Higher discrepancies on the even peaks are probably due to errors in the experimental measurements (Facchini and Surace, 2006). Comparing the two models, Two
135
Chapter 7.
700
exp.
600 500
TL path TL path
1 2
1 2
K 1]
-
HTC [Wm
10
15
20
25
X / D
30
35
40
45
50
Figure 7.13: Impingement ve holes - Heat transfer coecient along center lines.
Layer predicts peak values a 10% better of the k - 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 ux. Symmetry line behavior of HTC is not a re-
(b) k- SST
Figure 7.14: Impingement ve holes - Velocity magnitude on symmetry plane [m/s].
ally good parameter to establish whether or not the cooling device is eective. Even though it quite well shows maximum and minimum values, no information on the mean value, felt by the wall, are obtained. A wall HTC map is hence presented in Fig. 7.16 to both give an idea of an averaged value and permit an evaluation of ow structures. For this last purpose is of particular help the picture of the ow eld reported in Fig. 7.14 and Fig. 7.15. It could be noticed either in the velocity or in the temperature plot, the two models are perfectly in agreement in the prediction of the structure of motion. The eects due to cross ow over the plate magnify from left to right obviously following the increase in mass ow. Furthermore two columns of holes are not completely decoupled, the eect of such coupling is seen in the smearing on the symmetry plane at the height of the corresponding jet as well as in the increase of heat transfer coecient, see Fig. 7.13, on the opposite path. As already pointed out for a single hole conguration, the decrease of HTC
136
Chapter 7.
(b) k- SST
Figure 7.15: Impingement ve holes - Temperature distribution on symmetry plane [K].
downstream the jet is faster for the Two Layer than the other model, resulting in more roundish contour plots. The bubbles of higher heat transfer coecient due to passage of more streamlines deviated by the jet results now multiplied because of the interaction of the same phenomenon on dierent adjacent holes. This can be noticed also in Fig. 7.13, looking at the local maximum in correspondence with the hole on the opposite path.
(a) Experimental
(c) k- SST
[W/(m2 K)].
Figure 7.16: Impingement ve holes - Heat transfer coecient distribution on impinged wall
137
Chapter 7.
done by Sinha et al.. It simply is a study of a single jet mixing in cross ow over a at plate. Single row conguration was mainly considered in order to have a reference geometry to compare results for dierent turbulence models. In fact, single jet simulation over a at plate are not of particular interest for turbomachinery design. Multi jet congurations are the standard and should therefore be investigated. In particular, most recent developments in drilling capabilities allow the manufacturing of wide arrays of micro-holes (diameters less than 1 mm), currently referred to as eusion cooling. Even if this technique does not produce a lm wall protection as in standard lm cooling, its most ecient cooling feature is the heat removed by the passage of coolant inside the holes (heat sink eect): the high number of holes and their high length/diameter ratio (with angles below 30 ) allows to heavily increase the overall cooling eectiveness (Gustafsson, 2001). Eusion 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 lm wall protection may not represent the main cooling eect in eusion technique, the prediction of mixing between coolant and cross ow and the corresponding assessment of adiabatic eectiveness, still represent one of the most dicult task in CFD analysis Andreini et al. (2005). Second case for the lm cooling family was so chosen to represent experimental investigations done on an eusion cooling system done at the Energy Engineering Department of the University of Florence Arcangeli et al. (2006). These experiments represent a real full coverage turbine end-wall cooling device.
fundamental geometrical dimensions such as hole diameter D = 12.7 mm and coolant inlet area Aci = 1.210 mm2 .
138
Chapter 7.
Computational domain, as usual, reproduce the minimum periodic geometry meaning that two symmetry planes are imposed, one passing through hole axis and the other half spanwise pitch. Of particular importance for this case is the computational grid in the zone around the hole, where the jet entrains in the mean ow. Magnications of the used mesh are reported in Fig. 7.18, globally the mesh is composed by 177468 cells. Experimental ow conditions are listed
in Tab. 7.7, and are traduced in computational boundary conditions listed in Tab. 7.8. Results were checked in terms of adiabatic eectiveness, a parameter
Table 7.7: Sinha - Flow conditions.
Cross ow temperature Coolant temperature Pressure Density ratio - DR Blowing-rate - M Momentum ratio - I Turbulence level - Tu Cross ow velocity
Rec 300 153 101325 2.0 0.5 0.125 0.2 20 15700 K K Pa
m s1 W m2
that describe the inuence by hot gas temperature or by the coolant on the wall temperature:
Tw Tg , Tc Tg
(7.14)
where Tw is wall local temperature, Tg stands for hot gas temperature and Tc for coolant temperature. The performances of standard turbulence models (STD model in Figs.7.19(a) and 7.19(b) were analyzed in comparison with the anisotropic model ATL in terms of laterally-averaged and local span-wise and centerline eectiveness.
139
Chapter 7.
0.7
0.7
0.6
0.6
exp.
0.5
TL ATL
0.5
exp.
0.4
0.4
TL ATL
0.3
0.3
0.2
0.2
0.1
0.1
X/D
X/D
(a) 10D
(b) 15D
0.9
0.6
0.8
exp.
0.5
0.7
exp. TL
0.4
0.6
TL ATL
ATL
0.5
< >
0.3
0.4
0.2
0.3
0.2
0.1
0.1
0.0 0 2 4 6 8 10 12 14 16 18 20 22 24
0.0 0 2 4 6 8 10 12 14 16 18 20 22 24
X/D
X/D
Local lateral eectiveness 10 and 15 diameters downstream is shown in Fig. 7.19, local centerline and laterally averaged eectiveness is also presented in Fig. 7.20. Finally a map of wall eectiveness is reported in Fig. 7.21. The improvements obtained using the anisotropic model are remarkable. The experimental lateral proles of eectiveness are perfectly reproduced both 10 and 15 diameters downstream while all the other isotropic models where failing such predictions dramatically. The centerline values also tend to be much closer to experiments
140
Chapter 7.
especially when the distance from the hole is increasing. It is clear how the eddy viscosity model predicts a too coherent jet Fig. 7.21, thus severally underestimating lateral cooling performances. However this error is smoothed in the laterally-averaged parameters and as expected the anisotropic factor just distribute in dierent way the coolant over the plate not inuencing that much the averaged eectiveness. This consideration allows the use of isotropic models when the averaged parameters are considered but as expected conrms how the local values of wall temperature cannot be considered reliable.
141
Chapter 7.
(a)
(b)
in terms of spanwise averaged adiabatic eectiveness, see Fig. 7.23. Together with experimental data, correlative approach predictions using L'Ecuyer and Soechting correlation with Sellers superposition criterion have been reported Arcangeli et al. (2006). It is in fact quite dicult to reproduce exactly such experimental proles that are better used as a reference mean value for . In agreement to what has been already noticed in the previous case, the laterally averaged eectiveness is not that much inuenced by the anisotropic factor, see
142
Chapter 7.
Fig.7.23. All models in fact qualitatively well reproduce the correlative decay of the spanwise averaged eectiveness downstream the holes. By looking at the two-dimensional eectiveness map in Fig. 7.24 however it's evident how the lateral diusion of coolant is more correct in the anisotropic case than in the other two with a more qualitative agreement with experimental data.
0.50
0.45
0.40
0.35
0.30
0.25
0.20
0.15
0.10
0.05
0.00 0 10 20 30 40 50 60
X / D
(c) SST
(d) experimental
tiveness.
Figure 7.24: Comparison between laterally averaged and local center line lm cooling eec-
143
Chapter 7.
instrumentation and data is available. Comparisons with numerical simulations have been performed on two sets of experimental results, respectively BR 0.74 and BR 1.30 1.35 for both geometries. Since gas and coolant density are very close, respectively 0.82 and 0.89 kg/m3 , BR and VR are also close. Computational results are compared with experimental data in terms of adiabatic eectiveness. The maximum absolute error in measuring the experimental adiabatic eectiveness is about 0.05 (see Facchini et al. (2007)). Both SAA shaped congurations and PAA smooth conguration with compound angle deviation were simulated (see Table 7.11), (Carmine et al., 2007). In agreement with
Table 7.11: Blowing Ratio and Reynolds number for hot tests
BR
SAA PAA
Re
10795 11129 11783 18837 19097 20800
turbulence model requirements, grid were clustered near walls so that y + < 1. This feature was obtained using at least 15-20 cells inside the boundary layer thickness. The degree of resolution of the ne mesh was obtained following the results of previous similar works (Mangani et al. (2007); Bacci and Facchini (2007); Andreini et al. (2006)). For SAA sharp and smooth congurations
(a)
(b)
Figure 7.25: SAA Computational grids: coarser 7.25(a) and ner 7.25(b)
the calculation domain for computational analysis has then been chosen based on a symmetry plane boundary condition (Acharya (2007); Bacci and Facchini (2007); Andreini et al. (2006)) and only one half of the whole test section was reproduced, resulting in a reduction of calculation time. The choice of a symmetry plane also avoided convergence problems due to numerical jet shedding and oscillating phenomena that prevent a proper convergence of the RANS analysis. In order to reproduce the 2 degrees compound angle eect of real geometry only the PAA smooth conguration was chosen. The entire shaped hole with cyclic patches was simulated instead of mid-hole with symmetry patches congura-
144
Chapter 7.
tion and a correspondent velocity component was set on inlet boundary (Figure 7.26). This strategy was chosen to have a single geometry model to perform further analysis with dierent compound angles without mesh modications. Clearly the use of the cyclic boundary conditions instead of walls inuences the ow eld near the lateral wall but reproduce correctly the relative interaction between the main ow and the jet cooling (Figure 7.27). Therefore the local interaction between the jets is conserved in the simulation.
145
Chapter 7.
1.0
1.0
Exp
Exp
1.0
X/D = 5
0.8
X/D = 15
0.8
SST TL
SST TL
0.6
X/D = 25
Exp
0.8
SST TL
0.6
TL Anis
TL Anis
0.4
0.6
TL Anis
0.4
0.4
0.2
0.2
0.2
-5
-4
-3
-2
-1
-5
-4
-3
-2
-1
-5
-4
-3
-2
-1
Y/D
Y/D
Y/D
(a)
(b)
(c)
Figure 7.28: Local lateral for three location downstream for SAA smooth conguration
BR=0.73
1.0
1.0
1.0
Exp
Exp SST
X/D = 5
0.8
Exp SST
X/D = 15
0.8
X/D = 25
0.8
SST TL
0.6
TL TL Anis
0.6
TL TL Anis
0.6
TL Anis
0.4
0.4
0.4
0.2
0.2
0.2
-5
-4
-3
-2
-1
-5
-4
-3
-2
-1
-5
-4
-3
-2
-1
Y/D
Y/D
Y/D
(a)
(b)
(c)
Figure 7.29: Local lateral for three location downstream for SAA smooth conguration
BR=1.31
1.0
1.0
Exp X/D = 5
0.8
1.0
SST TL
X/D = 15
0.8
Exp SST TL
0.6
X/D = 25
0.8
Exp SST TL
0.6
0.6
TL Anis
TL Anis
TL Anis
0.4
0.4
0.4
0.2
0.2
0.2
-5
-4
-3
-2
-1
-5
-4
-3
-2
-1
-5
-4
-3
-2
-1
Y/D
Y/D
Y/D
(a)
(b)
(c)
BR=0.74
Figure 7.30: Local lateral for three location downstream for SAA sharp conguration
with SST model. Should be to noticed that the unexpected asymmetric trend of the experimental data for y/D < 3 of the sharp conguration is a spurious eect due to an uncorrect TLC color response in the area closest to the channel wall Facchini et al. (2007). Anisotropic correction to turbulence model allows to greatly improve accuracy of numerical prediction. In particular the most interesting result is the disappearing of the double peak structure predicted by standard models. The increase in lateral diusion of jet permits a more physical redistribution of coolant between centerline and peripherals regions resulting in
146
Chapter 7.
1.0
1.0
1.0
X/D = 5
Exp
0.8
X/D = 15
0.8
X/D = 25
Exp
0.8
SST TL
SST TL TL Anis
0.6
TL Anis
0.6
0.6
0.4
0.4
0.4
0.2
0.2
0.2
-5
-4
-3
-2
-1
-5
-4
-3
-2
-1
-5
-4
-3
-2
-1
Y/D
Y/D
Y/D
(a)
(b)
(c)
Figure 7.31: Local lateral for three location downstream for SAA sharp conguration
BR=1.29
Figure 7.32: Numerical adiabatic eectiveness map for SAA smooth conguration
Figure 7.33: Experimental adiabatic eectiveness map for SAA smooth conguration
147
Chapter 7.
Figure 7.34: Numerical adiabatic eectiveness map for SAA sharp conguration
Figure 7.35: Experimental adiabatic eectiveness map for SAA sharp conguration
a more smoothed prole. Anisotropic model shows the best agreement with experimental data for X/D = 5 for lower blowing ratio and for X/D = 25 for higher blowing ratio. Figures 7.32, 7.33, 7.34 7.35 report numerical and experimental data in terms of 2D contour maps which allow a more comprehensive comparison among dierent data sets. The maps comparison shows that for numerical and experimental results there is no uid interaction between the jet cooling and the lateral walls. That validates the use of a symmetry boundary condition instead of a no-slip type boundary for the lateral domain. It can be observed how computed jet's structure is much more coherent than in experiments: experimental eectiveness iso-lines tend to smooth quickly, especially for the higher blowing ratio, showing for X/D > 15 a fairly uniform prole. It
148
Chapter 7.
also appears clearly the greater lateral spreading of anisotropic model and its related smoothing of double peak structure. It's interesting to point out a lower
0.80 0.70 0.60 0.50 0.40 0.30 0.20 0.10 0.00 -5 0 5 10 15 20 25 0.80
BR=0.73
BR=1.31
10
15
20
25
X/D
X/D
(a)
0.80 0.70 0.60 0.50 0.40 0.30 0.20 0.10 0.00 -5 0 5 10 15 20 25
0.80
(b)
BR=1.29
BR=0.74
10
15
20
25
X/D
X/D
(c)
(d)
Figure 7.36: Spanwise averaged eectiveness < > for SAA geometry for smooth (7.36(a),
7.36(b)) and sharp (7.36(c), 7.36(d)) congurations
maximum value reported by experiments in terms of eectiveness in the region just downstream the hole (0 < X/D < 5) when blowing ratio increases, while all CFD results predict a region with 0.8. The discrepancy may be due to heat conduction within plate material during experiments which is not accounted in calculations. In fact as blowing ratio increases convective heat transfer grows and tend to smooth peaks and probably to modify jet decay. In order to have a more proper comparison we should perform conjugate computations taking into account solid conduction. Finally gure 7.36 reports the comparison with experimental data in terms of averaged spanwise eectiveness. The most interesting observation is the monotonic progressive decay of predicted by all computations against a more complex distribution shown by experiments, which is almost constant over wide zones. Analyzing single results we can point out that k SST better predicts the spanwise averaged data. It probably depends on a deeper penetration of jet within freestream boundary layer. Taking into account its lack of lateral diusion it could be interesting to verify the behavior of the model with the anisotropic correction: such modication will be the subject of future investigation. Contrary to the experiments, minor dierences are observed between sharp and smooth congurations for the spanwise averaged eectiveness, since the decay is similar for both congurations.
149
Chapter 7.
X/D = 5
0.8
Exp TL
X/D = 15
0.8
Exp TL TL Anis
X/D = 25
0.8
Exp TL TL Anis
0.6
TL Anis
0.6
0.6
0.4
0.4
0.4
0.2
0.2
0.2
-5
-4
-3
-2
-1
-5
-4
-3
-2
-1
0
Y/D
-5
-4
-3
-2
-1
0
Y/D
Y/D
(a)
(b)
(c)
Figure 7.37: Local lateral for three location downstream for PAA smooth conguration
BR=0.75
1.0
1.0
1.0
X/D = 5
0.8
Exp TL TL Anis
X/D = 15
0.8
Exp TL TL Anis
X/D = 25
0.8
Exp TL TL Anis
0.6
0.6
0.6
0.4
0.4
0.4
0.2
0.2
0.2
-5
-4
-3
-2
-1
0
Y/D
-5
-4
-3
-2
-1
0
Y/D
-5
-4
-3
-2
-1
0
Y/D
(a)
(b)
(c)
Figure 7.38: Local lateral for three location downstream for PAA smooth conguration
BR=1.36
shows that for numerical and experimental results there is no uid interaction between the jet cooling and the lateral walls. That validates the use of a cyclic boundary condition in order to simulate the PAA compound angle conguration. Most of the discussions reported for SAA geometry can be repeated in this case. The comparison with experimental data shows a general overestimation of central peak eectiveness and an underestimation of lateral spreading of the jet, resulting in an underestimation of eectiveness far from centerline (Y /D > 2). The most important observation is a great underestimation of the eect of compound angle predicted by computations (see Figures 7.39, 7.40). Especially at the exit of the hole experimental maps show a strong inuence of compound angle on eectiveness values, while numerical eectiveness distribution seems to be inuenced by crossow inlet angle mainly far downstream the
150
Chapter 7.
Figure 7.39: Numerical adiabatic eectiveness map for PAA smooth conguration
Figure 7.40: Experimental adiabatic eectiveness map for PAA smooth conguration
0.80 0.70 0.60 0.50 0.40 0.30 0.20 0.10 0.00 -5 0 5 10 15 20 25 0.80
BR=0.75
Exp TL TL Anis
BR=1.36
Exp TL TL Anis
10
15
20
25
X/D
X/D
(a)
(b)
Figure 7.41: Spanwise averaged eectiveness < > for PAA geometry and smooth conguration
151
Chapter 7.
hole. The double peak structure is very reduced using the anisotropic turbulence model, but is always present unlike the SAA simulation. Various authors explain the missing of the bimodal eectiveness pattern by the inuence of the hole length on the velocity prole, involving the presence or less of recirculation bubbles downstream the diuser wall (Saimweber and Schulz (2007)). Also a non-uniform ow eld upstream the plenum of the coolant jet can lead to the suppression of the double peak in the adiabatic wall eectiveness (Gritsch et al. (2001)). Figure 7.41 shows results in terms of laterally averaged eectiveness. A quite good agreement is observed at least for lower blowing ratio. With high BR conductive eects already noticed for the SAA geometry become relevant.
152
Chapter 7.
and interaction with external ow risky in term of aerodynamic performances, other techniques results very complicated to be inserted. That is why attention was posed on the axial cooling duct at the trailing edge of a turbine blade. The channel is ribbed and interaction with pedestals is also simulated. Ribs are usually inserted to increase the overall eciency and eectiveness of the cooling device, to reduce the air mass ow and more uniformly distribute the heat transfer coecient on the wall. At the same time some drawbacks, such as higher manufacturing costs and higher pressure loss, should be considered.
153
Chapter 7.
respect to the ow direction, inside the channel dened by a couple of pedestals. Two congurations with the same pedestal arrangement but with dierent rib shapes were considered. Square and circular ribs were compared with a smooth not ribbed surface Carmine et al. (2008). Numerical data were compared with in-house collected experimental data and geometries G1.0, G1.1A and G1.2A were studied (see Tab.7.12). Geometry parameters are summarized in Facchini and Tarchi (2008). Figure 7.43 is a
Table 7.12: Geometries
Geometry
G1.0 G1.1A G1.2A
Characteristics
not ribbed square ribs circular ribs
sketch of the computational domain together with boundary condition type used. The choice of a symmetry plane also avoided convergence problems due
to numerical jet shedding and oscillating phenomena that prevent a proper convergence of the RANS analysis. A summary of the boundary conditions is reported in Tab.7.13 Constant uid properties were considered. For geometry G1.1A unstructured grid was generated while for geometries G1.0 and G1.2A only the structured multi-block grid one. Table 7.14 reports grid characteristics in terms of mesh type and number of elements. Grids were clustered near the walls so that y + << 1: gure 7.44 shows details of the computational grids. Comparisons with experimental simulations were performed for Re = 18000 and M a = 0.3 in the throat section. Computational results are compared with experimental data in terms of HTC distribution and streamwise pressure drop. The maximum absolute error in measuring the experimental data is 12.2% for the HTC and 0.2% for the static pressure.
154
Chapter 7.
Location
Inlet Inlet Inlet Inlet Wall Outlet
Variable
m T Tu l Tw p
Units
kg s K m K Pa
1
Value
0.00231528 326.24 0.01 0.05 0.001 273 293 313 43000
Name
A D E
Geometry
G1.1A G1.0 G1.2A
Type
unstructured structured structured
Elements
1800000 1100000 1500000
(a)
(b)
(c)
(d)
Figure 7.44: Computational grids: pedestal and rib details for unstructured and structured
meshes
For geometry G1.1A rst a turbulence model sensitivity analysis was performed with mesh A and Tw = 313K . Figure 7.45 shows the heat transfer coecient maps over the endwall while Fig. 7.46 reports the HTC distribution on the centerline. All turbulence models predict some spurious spots in the heat transfer coecient due to the eect of tetrahedral mesh. We can see an over prediction of the heat transfer coecient in the leading edge zone of the rst enlarged pedestals and upstream the rst rib just near the lateral pedestal wall Bacci et al. (2005). These over predictions can be attributed to the excess of turbulence production in the stagnation regions typical of the two equations eddy viscosity models Durbin (1996). Downstream, this highly turbulent ow collides on the frontal surface of the rst rib. The ow in this region is sensibly accelerated because of the curvature of the leading edge of the enlarged pedestal and because of the reduction of the channel passage section due to the blockage
155
Chapter 7.
CFX SST
FOAM SST
FOAM RLR
FOAM TL
HTC [W/m2K]
EXP
400
200
150
100
50
0 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
x/D [-]
eect of the pedestal. When this accelerated and turbulent ow impinges on the rst rib we have the formation of a new recirculating ow zone and again a lot of turbulence production. This leads to a sensibly higher value of the HTC with respect to that resulting from the experimentations. Observing the HTC behavior near the second enlarged pedestal, we can see typical peak values due to impinging ow and vorticity produced by the horse-shoe vortex Ireland and Jones (1986). The exceeding turbulence produced in the stagnation region in front of the pedestal is then convected along the pedestal surface and lead to the overestimation of the HTC in the whole leading edge region. This disagreement with experimental results is made larger by the experimental setup. In fact, it
156
Chapter 7.
FOAM TL
FOAM LR
FOAM RLR
HTC [W/m2K]
EXP
must be pointed out that the presence of aluminum pedestals leads to an underestimation of the heat transfer coecient around them, where an appreciable air temperature drop occurs because the pedestal temperature is, on the average, lower than the endwall one. Such temperature decrease is not considered in the experimental data reduction (see Facchini and Tarchi (2008)). The comparison between numerical and experimental data shows that the SST turbulence model of both codes doesn't capture the rate of growth of the HTC peaks after the rst two ribs and underestimates it in the sloping at plate after the rib region. Furthermore the CFX SST overestimates the heat transfer after every rib while FOAM SST has a better accuracy. The discrepancy between the two codes can be explained by the universal temperature law used in CFX code Menter et al. (2003) while FOAM uses the standard low Reynolds formulation. The FOAM RLR doesn't predicts the rate of growth of the HTC peaks after the rst two ribs and underestimates the heat transfer downstream the third rib and in the sloping at plate after the rib region. The FOAM RLR is the only turbulence model that correctly predicts the heat transfer on the at plate in the inlet. The FOAM TL shows the better agreement with experimental data even if over predicts the heat transfer coecient in the rst part of the channel upstream the ribs. In fact HTC is well predicted compared to experimental data in the rib region as we can see from HTC map (see Fig.7.45) and HTC centerline prole (see Fig.7.46). Furthermore the FOAM TL is the only model that correctly evaluates the rate of growth in the ribbed channel.
157
Chapter 7.
Slope change
Section change
200
150
100
50
0 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
x/D [-]
158
Chapter 7.
Wx [1/s]
Figure 7.49: Geometry G1.0: contour of x vorticity component on six sections normal to x
axis
FOAM TL
EXP
HTC [W/m2K]
we can appreciate secondary ow intensity reduction from pedestal leading edge to its trailing edge: this causes HTC streamwise reduction.
200
150
100
50
0 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
x/d [-]
159
Chapter 7.
heat transfer coecient values then geometry G1.1A, in the numerical results this geometry shows higher values. This behavior can be explained looking at Fig.7.52, that shows a comparison between geometry G1.1A and G1.2A in term of velocity vector plot on the symmetry plane. The conventional ow pattern
Velocity peak
Reattachment point
v [m/s]
is established between two ribs. A primary recirculation zone is formed just downstream the rst turbulator with much more little secondary recirculation in the corner created by the side of the rib and the endwall. Downstream we have the ow reattachment and a new little recirculation zone arise in front of the following rib. Geometry G1.1A presents bigger vorticity zones downstream and upstream the ribs. Instead geometry G1.2A presents a higher velocity peak near the wall, where there is the heat transfer coecient higher value. So the ow eld is coherent with the heat transfer distribution. We also notice that little recirculation zones are present on the top of both rib type. Figure 7.53 shows the static pressure distribution on the centerline of the upper wall. Clearly geometry G1.1A presents more pressure losses then geometry G1.2A and
Static Pressure centerline up
1.02 G1.0-RLR 1.00 0.98 G1.0-EXP 0.96 0.94 0.92 0.90 0.88 0.86 0.84 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 G1.1A -EXP G1.2A -EXP G1.1A -TL G1.2A -TL
p/p in [-]
x/d [-]
geometry G1.2A presents more pressure losses then geometry G1.0. There is also a quite good agreement with experimental data. All the turbulence models used show the same trend.
160
Chapter 8
Contents 8.1
8.2 8.3
Numerical Modeling . . . . . . . . . . . . . . . . . . 176 Model Validation: Lean Premixed Test Rig . . . . . 177 Numerical Modeling . . . . . . . . . . . . . . . . . . 180 Model Validation: GE10 Burner . . . . . . . . . . . 181
. . . 180
161
Chapter 8.
Yk = Yk (Z, K)
T = T (Z, K)
(8.1)
where Yk is the generic mass fraction of the k species, T the temperature and Z the mixture fraction Anyway in the present work the strain rate K is negligible and hence is not considered. The quasi-1D laminar ames solution output (Eq. 8.1) are then integrated by:
1
Q=
0
Q(Z)p(Z)dZ
(8.2)
where p(Z) is the Z-pdf. The integration of Eq. 8.2, performed during preprocessing by the in house FORTRAN code prePdf, gives libraries of the generic variable Q as:
Q = Q Z, Z
Equations 8.1 tighter with equations 5.7, 5.8 and 5.9 allow to calculate the ow density (that becomes function of only Z and Z 2 ) needed to solve the RANS equations, anyway if the ame is not adiabatic it is necessary to consider the heat exchange eects. To make this, the conservation of thermochemical enthalpy is enforced (Eq. 8.3) through a proper transport equation (5.3) solved by the CFD code and is
162
Chapter 8.
explicitly considered the dependence of mean specic heat on temperature (Eq. 8.4).
K
h=
hk = Cp (T )(T Tref ) +
k=1
(8.3)
=
k=1
Yk ho k
Cp (T ) =
1 T Tref
K
Cp (T )dT
Tref
(8.4)
Cp (T ) =
The libraries are:
Yk Cpk (T )
k=1
Formation enthalpy
= Z, Z
2
(8.6)
(8.7)
(8.8)
pf lamelet W RT
where the pressure pf lamelet is an input (which is the same used to solve the laminar ames, being usually nearly constant in combustion systems) while both molecular weight W and temperature T are given after interrogating the libraries:
163
Chapter 8.
1. Z and Z 2 obtained by the RANS solution are used to read the libraries to obtain the values of W , Yk and and the vector Cp T , and usually bilinear interpolations are required; 2. a guessed temperature is used to calculate a Cp (again interpolations are usually needed) that together with the formation enthalpy is used to invert Eq. 8.3 to obtain again the temperature value; 3. guessed and calculated temperature are compared and if tolerance requirement is met, follow point 4 otherwise steps back to point 2; 4. the calculated temperature and the mean molecular weight are used in perfect gases law.
164
Chapter 8.
The burned is fed with a mixture of CH4 /H2 (1:1 in volume), having Zst = 0.04997. This ame represents a suciently complex testcase. Just the prediction of the recirculation zone extension is hardly achievable with two equation turbulence models, but it is key of the estimation of ame stabilization. Low value of stoichiometric mixture fraction (Zst ) may cause strong discrepancies on temperature eld even if small errors on mixture fraction occurred ((Andreini, 2004)). In (Hossain et al., 2001), it has been shown that the radiative transfer does not induce sensible eects on the temperature eld (T < 50 K ) and simulations performed with radiative exchange model (for implementation details see Chapt. 10), whose results are not reported here, conrm it. The simulated ame
Figure 8.3: BBody burner: schematic representation of the ame zones with some sampling
sections
is divided into three dierent zones (Fig. 8.3): 1. The recirculation zone where are overriding the turbulent mixing eects 2. An extinction zone (neck zone) just downstream the recirculation zone, where the ux regains its uniform jet-like structure: the chemical reactivity is in this case the limiting phenomenon.
165
Chapter 8.
3. Farther from injection orice the ame is stabilized (somehow it is reignited) due to the weakening of turbulent mixing allowing a real jet like propagation.
The mesh used in simulations is shown in gure 8.4. It is made of 24948 hexaedrons, and one in circumferential direction due to the axial-symmetry. The inlet boundary conditions are the averaged experimental values of temperature, velocity and turbulent kinetic energy (table 8.1) while the dissipation has been chosen to comply with:
k 3/2 Ldissipation = 0.07 Rjet = 0.13 [mm] while all wall are adiabatic and no-slip and at the outlet has been imposed the atmospheric pressure. The laminar amelets are built with the detailed Inlet F uel Air v[m/s]
118 40
k[m2 /s2 ]
208.88 1.8950
T [K]
300 300
mechanism of Gas Research Institute. It contemplates 32 chemical species and 177 reactions. It is capable to manage steady combustion with high temperature of methane and hydrogen but it neglects nitrogen oxides formation mechanisms (anyway not interesting since these have small inuences on energy balances). Simulations have been performed with the standard high-Reynolds turbulence model ((OpenCFD, 2006b)) where the coecient C1 has been changed from the usual value of 1.44 to 1.6 in order to better estimate jet penetration as suggested in (Masri, 1998). The obtained results, reported in the plots in gures 8.9-8.17, are compared to results from experiments, and from a amelet model implemented in ST AR CDT M (by eective members of the group in which present work has been carried out). The model implemented in ST AR CDT M used the same chemical mechanism (GRI) and the same combustion model of the present work but neglects the dependence on mixture fraction variance of mean specic heat (Eq. 8.5) i.e. the Z pdf for Cp is a Dirac-delta. Calculation results for mixture fraction Favre average (g. 8.9) match experimental results in each section, except for an overestimation close the burner axis in the recirculation zone (x < 65 mm), this may
166
Chapter 8.
Figure 8.5: BBody burner: mixture fraction (left) and temperature contours (right)
be explained by an excessive penetration of the fuel stream. ST AR CDT M correctly evaluates the axis value of Z together with a global underestimation that grows with the distance from the burner contrarily to F OAM that, on the contrary, improves moving away from the burner. The implemented model correctly spots the positions of stoichiometric condition even if a very small error may cause much larger error on temperature and species. The shape of mixture fraction variance (g. 8.10) is well reproduced for the rst two sections (x = 13 [mm] and x = 30 [mm]), with an higher peak value. Moving downstream, results become worse for the last two sections in the recirculation zone (x = 45 [mm] and x = 65 [mm]) while they improve in next extinction and jet zones (x = 90 [mm] and x = 120 [mm]). A similar, but stronger, behaviour is shown by ST AR CDT M . The good prediction of mixture fraction moments for the rst two sections implies the matching of temperature eld (g. 8.11) even if the reactivity is underestimated, how can be spotted by the CO and CO2 plots (g. 8.12, 8.13), where CO has higher value than experimental results (at second section g. 8.13) while CO2 has a lower mass fraction than experiments, except where a local high reactivity is present. This localized higher reactivity maybe due to the neglection of strain in a zone where the ow is highly curved is also spotted by the high OH concentration, that is much lower elsewhere. Hydroxide is produced where the CO is oxidized to CO2 where the ame is far from extinction; this can also be noticed in gure 8.6, that shows higher OH concentration where is present a gradient of CO mass fraction on the air side. ST AR CDT M shows a similar behaviour for the rst section (g. 8.12) and the temperature peak is lower aside with less steep shapes of CO2 and OH plots, that anyway have higher concentrations resulting in an overall higher temperature. In the second section (g. 8.13) the reactivity is largely overestimated by ST AR CDT M .
167
Chapter 8.
The H2 O mass fraction matches experimental and ST AR CDT M results at both sections. The section in the recirculation zone (clearly visible in g. 8.7) at x = 45 [mm] (g. 8.14) shows lower temperature close to the axis, this is provided by either ST AR CDT M or present work models, so it probably cannot be attributed to the local overestimation of Z since ST AR well predicts it on the axis, instead it maybe due the local overestimation of the variance that gives local higher concentration of CO and the lower one of CO2 . Moving away from the axis temperature eld is well predicted until the zone near to the stoichiometric point that is better localized by ST AR CDT M that indeed a more precise estimation of the peak value. F OAM much better predicted the CO while ST AR CDT M shows a large decrease, for radial distances greater than 10 mm, of its concentration sided by the matching of CO2 values that anyway seems to be met by a larger reactivity demonstrated by an higher concentration of OH that is better predicted by the present work model that also gives lower mass fraction of carbon dioxide. The lack of a very good prediction of CO may seem in contrast with the good prediction of temperature, especially by ST AR CDT M , but it has to be considered that this is a section strongly aected by recirculation and then by the mixing with hot burnt species from other zones. For the right downstream section (x = 65 [mm], g. 8.15) can be done similar considerations although ST ARCDT M works much better on CO and for zone close to the stoichiometric point it matches experimental results, and now it is F OAM that overestimate the reactivity. For both sections, both codes can predict H2 O concentration with a high precision. In the extinction and jet-like propagation zones (x = 90 mm, g. 8.16 and x = 120 mm, g. 8.17) ST AR CDT M matches the temperature eld while
168
Chapter 8.
Figure 8.7: BBody burner: details of velocity eld contour in the recirculation zone with
streamlines
Figure 8.8: BBody burner: mixture fraction variance contour near blu body.
F OAM strongly underestimates it closer to the axis and overestimates it near the stoichiometric zone. These statements strongly disagree with the prediction of mixture fraction moments so that the amelet model may be inadequate in these zones indeed either (Hossain et al., 2001) or (Masri, 1998) fails here, while (Andreini, 2004) matches experimental results with an EDC model with detailed chemistry. Moreover, aside with a good prediction of mixture fraction moments, and with much dierent values than ST AR CDT M ones, CO is strongly overestimated by both codes (calculated concentrations are nearly two folded than the experimental ones). CO2 concentration is well predicted by ST AR CDT M while F OAM strongly overestimates it near the stoichiometric value that is better localized by ST AR CDT M that also work better on OH . H2 O mass fraction is matched by both codes. The analysis of results shows the good behaviour of the implemented model. Previous simulations, whose results are not reported here, where conducted with the same model implemented in ST AR CDT M and the obtained results were compared the above reported ones and the improvement in temperature eld calculation has allowed to demonstrate that the specic heat dependence on mixture fraction variance cannot be neglected.
169
Chapter 8.
1,0
1,0
exp
exp
0,8
HTC STAR-CD
0,8
st
0,6
HTC FOAM
0,6
0,4
0,4
Y=13 [mm]
0,2
Y=30 [mm]
0,2
0,0 0 5 10 15 20 25 30
0,0 0 5 10 15 20 25 30
0,6
exp
exp
0,6
0,5
0,4
st
0,4
0,3
Y=45 [mm]
0,2
Y=65 [mm]
0,2
0,1
0,0 0 5 10 15 20 25 30
0,0 0 5 10 15 20 25 30
0,4
0,30
0,25
HTC FOAM Z
st
0,20
0,2
0,15
Y=90 [mm]
0,1
0,10
Y=120 [mm]
0,05
0,0 0 5 10 15 20 25 30
0,00 0 5 10 15 20 25 30
170
Chapter 8.
0,020
0,016
0,018
0,014
0,016
exp
0,012
HTC FOAM
0,014
0,012
0,010
0,010
0,008
0,008
0,006
Y=13 [mm]
0,006
Y=30 [mm]
0,004
0,004
0,002
0,002
0,000 0 5 10 15 20 25 30
0,000 0 5 10 15 20 25 30
0,010 0,012
exp
mixture fraction variance mixture fraction variance
HTC STAR-CD
0,010
exp
0,008
HTC FOAM
0,008
0,006
0,006
0,004
0,004
0,002
Y=45 [mm]
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30
0,002
Y=65 [mm]
0 5 10 15 20 25 30
0,000
0,000
0,012
0,010
exp
mixture fraction variance
0,010
exp
mixture fraction variance
HTC STAR-CD
0,008
HTC FOAM
0,008
0,006
0,006
0,004
0,004
0,002
Y=90 [mm]
0 5 10 15 20 25 30
0,002
Y=120 [mm]
0 5 10 15 20 25 30
0,000
0,000
171
Chapter 8.
2000 1800 1600 1400 1200 1000 800 600 400 20 25 30 200 0 5 10
Y=
1400 1200 1000 800 600 400 200 2000 1800 1600 0 5 10
Y=
13 mm
30 mm
15
15
20
25
30
1800
Temperature [K]
Temperature [K]
Y=
45 mm
Y=
65 mm
15
20
25
30 1600
Temperature [K]
10
15
20
25
30
Temperature [K]
120 mm
10
15
20
25
30
200
10
15
20
25
30
172
Chapter 8.
0,06
0,12
H O [mass fraction]
CO [mass fraction]
0,05
0,10
0,04
0,08
0,06
0,03
exp
0,04
0,02
exp
0,02
0,01
0,00
0,00 0 5 10 15 20 25 30
-0,02 0 5 10 15 20 25 30
[mass fraction]
0,05
OH [mass fraction]
exp HTC STAR-CD HTC FOAM
0 5 10 15 20 25 30
0,0010 0,0009 0,0008 0,0007 0,0006 0,0005 0,0004 0,0003 0,0002 0,0001 0,0000 -0,0001 0 5
0,04
0,03
CO
0,02
0,01
0,00
-0,01
10
15
20
25
30
0,12
0,08
0,06
EXP
0,04
CO [mass fraction]
H O [mass fraction]
0,10
0,02
0,00 0
0,07
-0,01
5 10 15 20 25 30
0,0020
10
15
20
25
30
0,0018
0,06
0,0016
EXP
0,05
0,0014
CO [mass fraction]
O H [mass fraction]
0,04
0,0012
0,0010
0,03
0,0008
EXP
0,02
0,0006
0,0004
0,01
0,0002
0,00 0 5 10 15 20 25 30
0,0000 0 5 10 15 20 25 30
173
Chapter 8.
0,12
H O [mass fraction]
CO [mass fraction]
0,04
0,10
0,03
0,08
0,02
0,00 0 5 10 15 20 25 30
0,00 0 5 10 15 20 25 30
0,08
2,0x10
[mass fraction]
OH [mass fraction]
0,06
1,5x10
-3
HTC FOAM
0,04
CO
5,0x10
-4
0,0
0,00 0 5 10 15 20 25 30
0 5 10 15 20 25 30
0,05
0,12
[mass fraction]
[mass fraction]
2
0,10
0,04
0,08
0,03
0,06
CO
0,02
0,04
exp HT
exp
0,01
HT HT
C C O
F
STARAM
CD
15 20 25 30
0,02
HT
C C O
F
10
STARAM
CD
0,00 0 5 10
0,00 0
-3
15
20
25
30
2,0x10
1,8x10
-3
[mass fraction]
0,07
H [mass fraction]
1,6x10
-3
0,06
1,4x10
-3
exp HT HT
1,2x10
-3
0,05
C C O
F
STARAM
CD
1,0x10
-3
0,04
8,0x10
-4
CO
O
exp HT HT
0,03
6,0x10
-4
0,02
0,01
C C O
F
STARAM
CD
20 25
4,0x10
-4
2,0x10
-4
0,0
-4
0,00 0 5 10 15
-2,0x10
30
10
15
20
25
30
174
Chapter 8.
0,06
0,14
0,05
0,12
0,10
CO [mass fraction]
0,04
H O [mass fraction]
0,08
0,03
0,06
0,02
0,04
exp
0,02
0,01
0,00
0,00 0 5 10 15 20 25 0 5 10 15 20 25 30
0,06
[mass fraction]
0,04
OH [mass fraction]
0,0008 0,0007 0,0006 0,0005 0,0004 0,0003 0,0002 0,0001 0,0000 -0,0001
0,02
CO
HTC FOAM
10
15
20
25
30
10
15
20
25
30
0,14
0,06
0,12
0,05
CO [mass fraction]
H O [mass fraction]
0,10
0,04
0,08
0,03
0,06
exp
0,04
0,02
0,01
exp
0,02
HTC STAR-CD
0,00
HTC FOAM
0,05
O H [mass fraction]
CO [mass fraction]
0,04
0,0006
0,03
0,0004
0,02
exp
HTC STAR-CD
0,01
HTC FOAM
0,0002
0,00 0 5 10 15 20 25
0,0000 30 0 20
175
Chapter 8.
3. Evaluation of minimum distance dmin from the mean ame front for all cell centers 4. Re-initialization of G eld (G = dmin , G = G0 ) 5. Re-distribution of mean ame front G front normal direction
2
Referring to Fig. 8.18, in which A and B denote two adjacent volume cells, such as to satisfy Eqn. (8.9), mean ame front Cartesian coordinates X = (x1 , x2 , x3 ) may be found by rst order interpolation algorithm of Eqn. (8.10)
(8.9)
176
Chapter 8.
XG=0 = XA + e
(XB XA ).
(8.10)
2
A G weighted average is performed to evaluate mean ame front values for G and T (Eqn. (8.11)).
G=0 = e
(A) +
(B),
(8.11)
= G 2 , T
As mentioned before, in this Level-set method the characterization of thermochemical properties of reacting mixture is determined following a laminar amelet approach. In particular laminar back-to-back amelets are solved externally from CFD computation. This type of amelets allows to set aerodynamic stretch as a function of strain K, which only depends on inlet conditions. Laminar amelet data are integrated before the CFD computation and then stored into external look-up tables. The generic mixture averaged scalar is given by Eqns. (8.12) and (8.13).
+
Q(G, G 2 , K) =
Q(G G0 , K) P (G)dG,
(8.12)
P (G) =
1 (2 G 2 )0.5
exp
(G G)2 2G
2
(8.13)
To limit the size of look-up tables, the minimal set of data is stored, which is constituted by laminar ame speed, mean molecular weight, enthalpy of formation and averaged mixture specic heat like in the previous section. Because of PDF integration, table data set is stored as a function of G, its variance G 2 and the turbulent parameter stretch K , dened in chapter 5.4: this last, during the CFD calculation is approximated by integral turbulent frequency /k and during calculations the value of each scalar is evaluated by a tri-linear interpolation.
177
Chapter 8.
Calculations were performed with the two selected models Level-Set 5.6.5 (GEq) and Turbulent Flame Closure (TFC) 5.54: laminar amelets for LevelSet model have been prepared using the reaction mechanisms of Gas Research Institute (GRIMech 2.11). For TFC model the unstretched laminar ame speed was assumed equal to 0.108[m/s]. Computational mesh was prepared assuming the axis symmetry of the ame: about ten thousands hexahedral elements had been proved to satisfy mesh independence. Fig. 8.20 reports a direct comparison
Figure 8.20: Contour plots of temperature and progress variable source for tfc model (left)
and of temperature and iso-g for level-set model (right).
178
Chapter 8.
between the two models. A qualitative comparison of the temperature distributions points out a very similar behavior of the two models, with a comparable ame length and brush thickness. Interesting to observe the contour plot of the progress variable source term for the TFC model: the maximum values are located in the shear layer region where local increasing of turbulent kinetic energy implies high values of turbulent ame speed (see Eqn. 5.63). Iso-lines plot of scalar G allows to properly point out the location of averaged ame front as predicted by Geq model. Fig. 8.21 reports a comparison among temperature radial proles predicted by the two models and measured on test rig: three dierent axial locations were considered, being D the diameter of blu-body. An overall agreement is observed even if, locally, some discrepancies arise. Both models underpredict hot gas temperature in the central region for the rst axial section. A possible explanation is a local heat loss in the actual ame due to the cooling of blu-body plate which was not modeled in the computations. A wrong modelization of aerodynamic stretch may also locally inuence the accuracy: this is conrmed by recalling that implemented TFC model do not account for ame stretch giving an explanation to the underestimated ame brush thickness and to the higher values of ame temperature predicted by this model. Moving downstream the blu-body, the Geq model seems to slightly overpredict ame brush thickness: also in this case it is possible to explain the issue with an overprediction of ame stretch which it is probably due to an inaccurate prediction of turbulent eld. Nevertheless, Geq model appears to more properly predicts the real length of ame with respect to measured data. The real value added
1700 1600 1500 1400 1300 1200 1100
T [K]
1000 900 800 700 600 500 400 300 200 0.00 0.05 0.10 0.15 0.20
x/D = 1.0
0.25
0.30
0.35
0.40
0.45
0.50
0.55
0.60
0.65
r/D [-]
1700 1600 1500 1400 1300 1200
T [K]
1100 1000 900 800 700 600 500 400 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50 0.55 0.60 0.65
x/D = 0.8
r/D [-]
1700 1600 1500 1400 1300 1200 1100 1000 900
T [K]
x/D = 0.1
EXP
800 700 600 500 400 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50 0.55 0.60 0.65
GEq TFC
r/D [-]
Figure 8.21: Averaged temperature proles at three dierent locations x/d = 0.1, x/d = 0.8
and x/d = 1.0.
179
Chapter 8.
OF
10000
EXP -GEq
[ppm]
CO
1000
x/D = 1.0
r/D [-]
OF
10000
EXP -GEq
CO
[ppm]
x/D = 0.3
1000
r/D [-]
by the use of Level-Set approach is the opportunity to predict non-equilibrium eects on main chemical species due to the interaction with turbulent ow eld. Fig. 8.22 reports a comparison between measured and computed CO mole fraction. A noticeable overprediction is observed even when the model is able to catch the location of local maxima in radial proles. The inaccurate quantitative prediction of CO distribution is partly due to the cited issues in the computation of stretch eld, but it aso depend on the use of scalar G as abscissa in laminar amelet without the suggested normalization proposed by Herrmann Hermann (2006). A detailed investigation on this topic is beyond the scope of this work. In fact it is important to recall that actual combustors are usually operated with partially premixed ame that will be modeled with non-premixed laminar amelets based on mixture fraction as primitive scalar.
Progress variable equation steady formulation of Eq. 5.51, with the turbulent transport term modelled with Eq. 5.52, is solved aside RANS.
180
Chapter 8.
Flamelet libraries for unburnt state are built in the same way of the diusion ames ones but without the dependence on mixture fraction variance (i.e. the pdf is a Dirac's Delta) since its eect is negligible: Qu = Qu Z Interpolation performed on burnt state amelets libraries gives: Cpb , b , Wb . Interpolation performed on unburnt state amelets libraries gives: Cpu , u , Wu .
Unburnt state values are used to invert equation 8.3 to obtain Tu that together with Wu , is used to calculate, by perfect gases law (Eq. 5.9), the unburnt gases density u needed by c equation (5.51) source term.
By burnt and unburnt values, through Eq. 5.66, the ow values of Cp , , W are obtained. The resulting specic heat is used to calculate the ow temperature by inverting Eq. 8.3, and then, by perfect gases law (Eq. 5.9), the ow density needed by the the RANS code.
cequation source term both Eq. 5.54 and 5.56 models, presented in chapter 5.6.5, have been implemented.
The laminar ame speed (SL ) is obtained by interpolating on libraries function of mixture fraction Favre's average, built with and experimental correlation from (Elia et al., 2001). The velocity uctuation, needed by equations 5.55 and 5.57 is dened as
181
Chapter 8.
Nuovo Pignone (GE-NP). It is a single tubular combustor, settled with its axis orthogonal to the turbine one (g. 8.24) and burns an air-methane premixed ame, that is stabilized by a diusion pilot jet. The device then belongs to the Dry-Low NOx combustors category based on the lean premixed technology. The combustor has been analysed by a simplied geometrical representation allowing to treat it as axial symmetrical. The used geometrical model, even when simplied, can analyse every most important combustor subsystems (g. 8.25), keeping in consideration the reactants mixing processes and considering the air injection needed for wall cooling and for combustion products dilution, performed to reduce the temperature to values allowed in turbine rst stages. The combustor is made of a cylindrical
182
Chapter 8.
enclosure having, on its top, a conical cover (g. 8.25). The main stream reaches the reaction zone through an annular duct, that connect the so-called burner cap to the cylindrical enclosure. The pilot ame is realized by the injection of methane through 32 nozzles disposed on the cylindrical cover around the already described duct. Burner cap is made of an annular duct where reactant premixing occurs while the stream has got a centripetal motion (g. 8.25). At the duct inlet are disposed the fuel injection devices, which are 24 small pipes, with their axis in radial direction, equally distributed around the burner axis. Each of these pipes has got 7 holes through which the fuel is injected. In this premixing zone are also present some vanes, with variable stagger angle making them capable of creating a swirl of the reacting mixture. In design condition, at which simulations have been performed, the vane are setted in other to remove the swirl, induced by the non uniform incoming air velocity: the reacting ow get into the main reaction zone in, nearby, axial direction. The cylindrical liner is cooled, mainly, by lm cooling even if, in the cold air side, turbulence promoters are present in order to enhance the heat exchange. The exceeding air dilution is performed thanks to 4 holes that inject it in correspondence of the combustor outlet. The mesh on which runs have
T CD [K] 686
P CD [bar] 15.7
Table 8.2: GE10: inlet air condition, TCD and PCD stand for Temperature and Pressure
at Compressor Discharge
been performed has been made available by GE-NP for (Andreini, 2004). The model has a single layer of hexaedrical cells representing a 5 angular portion, for a total of 80000 elements (g.8.26). In order to represent the whole combustor geometry, by a bidimensional mesh, each cooling and dilution air holes and the 32 fuel injector have been represented by slots having equivalent eective areas. Fuel injection nozzles, on the 24 pipes, has been represented as injection points equally distributed in circumferential direction, keeping in consideration their real radial position: the fuel injection has been simulated by local mass source on single cells (see g. 8.26). The combustor treats an air mass ow of about 37 kg/s shared at 70% by the main inlet (combustion air), at 20% by the cooling holes and the remaining 10% by the dilution holes. The fuel mass ow is about 0.728 [kg/s]: 5% for the pilots and 95% for the main ame. Fuel temperature has been assumed equal to 293 K . All simulations have been carried out with the standard k high-Reynolds turbulence model (see g. 8.27): for the main inlet has been considered a turbulence intensity of 15% and 10% for all other inlets, while for all inlet has been considered a dissipation length of 7% of the hydraulic radius.
183
Chapter 8.
184
Chapter 8.
Simulations results have been compared to those by (Andreini, 2004) coming from an implementation of the KPP-BML model in ST AR CDT M , with a slightly dierent approach: the model is adiabatic and temperature and density, in both burnt and unburnt states, are adiabatic equilibrium values obtained as function of only mixture fraction, for further details see (Andreini, 2004). In gure 8.27 is shown the schematic representation of expected K-ONE ame. Due to the lack of reactant stream swirl, it assumes a conical like shape, extending in axial direction up to far sections from the inlet. The ame is held in correspondence of the diusion pilot ame. It has also been shown that the holding may occur in correspondence of the nose (see g. 8.26). (Andreini, 2004) places the combustion regime that interests the K-ONE, in the thin reaction zone with huge Karlovitz's numbers that place the ame very close to the distributed reaction zone. The BML model assumes high value of Damkhler's number being formally correct in the corrugated ame zone. Then it has to be expected an underestimation of the overall reactivity.In reported plots KPP stands for KolmogorovPetrovski-Piskunov closure, TFC stands for Turbulent Flame Closure and ST AR CDT M stands for result from (Andreini, 2004). Considering the mixture fraction plots (Fig. 8.28) at the end of the premixing duct, results from implemented models are identical and nearly match the results by ST AR CDT M . F OAM predict a slightly more uniform mixing while the manufacturer declares that the mixing is uniform. The velocity elds by
0,030
HTC-STAR-CD
0,028
TM
HTC-FOAM
0,026
0,024
0,022
0,020
0,018
0,00
0,02
0,04
0,06
0,08
0,10
Radial distance
the implemented models are identical and nearly equal to the ST AR CDT M one (g 8.29): the main stream from inlet and the lower speed beneath the nose and in correspondence of the recirculation zones are clearly visible. Comparing g. 8.30 and g. 8.31, reporting the vector plot of velocity, is noticeable the agreement in the prediction of the recirculation zone that (Andreini, 2004) has demonstrated to be unable to induce a ame holding, instead assured by the pilot ame. Figure 8.32 reporting the progress variable contours and gure 8.33
185
Chapter 8.
Figure 8.30: GE 10: vector plot of velocity eld near combustor cap by (Andreini, 2004)
Figure 8.31: GE 10: vector plot of velocity eld near combustor cap by the implement
model
reporting temperature contours show how the ame structures predicted by the implemented model are very similar to (Andreini, 2004) results. In particular, TFC closure gives a longer ame respect to ST AR CDT M results while the KPP gives a shorter ame and is the only one that achieve to complete the reaction before the end of the combustion dedicated zone localized in correspondence of dilution air inlets. ST AR CDT M gives a maximum pilot ame temperature of 2476 K , the KPP model 2320 K and the TFC model 2307 K these dierences will be discussed later. Discrepancies between the elds of progress variable, and consequently of temperature, provided by the implemented models have to be attributed mainly to the progress variable source terms, whose contours are reported in g. 8.37 where is clearly noticeable the links between the source term and the progress variable itself: as the source term lm is longer the c will have a longer zone with values greater than 0. Indeed, the implemented models dier just by the source term denition (Eq. 5.54-5.56) while dierences with ST AR CDT M are also due to discrepancies between turbulent kinetic energy
186
Chapter 8.
Figure 8.32: GE 10: progress variable contours from right to left: KPP closure, TFC closure and ST AR CDT M KPP closure
tours from right to left: KPP closure, TFC closure and ST AR CDT M KPP closure
and dissipation elds. Progress variable source terms by implemented models also dier because of the dierent turbulent ame speeds by which they are calculated (Eq. 5.555.57). Contours of these two ame speeds are reported in g. 8.38 that shows the agreement, in shapes, among the runs results, anyway are also noticeable dierences in magnitude, with the higher provide by the KPP model and the lower by ST AR CDT M , again the turbulence elds play a key role. Moreover is noticeable that there are no reaction beneath the nose, where the presence of a shear layer would justify the reacting mixture ignition that is prevented by the there occurring strong dilution. The manufacturer conrmes these considerations also informing about the necessity of injecting a small air ow in correspondence of the nose produced recirculation zone (purge air ) in order to avoid any reaction. The interaction between the cooling air stream, the
187
Chapter 8.
source term from right to left: KPP closure, TFC closure and ST AR CDT M KPP closure
source term from right to left: KPP closure, TFC closure and ST AR CDT M KPP closure
main reactant stream and the pilot jet produces a lifted ame (see chapter 5.7) that can be spotted in g 8.34-8.35-8.36 that show how the implemented models gives result very close to those coming from ST AR CDT M . The here described implementation of the partially premixed ames model is part of a larger work not yet completed. The analysis performed on the GE-10 combustor has shown the good capability of the implemented models moreover the non-adiabiticity of the model, aside with the complete analysis of the mixing process performed by a amelet approach (considering also mixture fraction variance) allows the prediction of more plausible pilot ame temperature that indeed where surely overestimated by (Andreini, 2004).
188
Chapter 9
Results: P1 Model
9.1.1 9.1.2 9.2.1 9.2.2 9.2.3
. . . . . . . . . . . . . . . . . . 190 . . . . . . . . . . . . . . . 194
. . . . . . . . . . . .
RTE Numerical Modeling . . . . . . . . . . . . . . . 190 Model Validation: Concentrical Cylinders Testcase . 191 RTE Numerical Modeling . . . . . . . . . . . . . False Scattering . . . . . . . . . . . . . . . . . . . Model Validation: RADIARE . . . . . . . . . . . 9.2.3.1 Case 1 . . . . . . . . . . . . . . . . . . . 9.2.3.2 Case 2 . . . . . . . . . . . . . . . . . . . Model Validation: Spherical Isothermal Domain 194 196 197 197 199 202
9.2.4
189
Chapter 9.
>0 k>0
because otherwise we would have singularities, and moreover there would be no need to solve Eq. 6.34 since it would be simply expressed as:
G = 0,
where n is the normal to the wall with a domain incoming direction. Even if sometimes Marshak 's boundary condition can be simplied, when it has to be solved it gives many problems. Being an implicit equation in G it requires an iterative procedure, in the present work it has been chosen to rearrange it in order to calculate the Gi value as a function of the gradient of its previous iteration value Gi1 and of wall black body emission:
Gi = 4Ib (Tw ) + 2
1 3k
Gi1 .
Unfortunately the term Gi1 shows an highly instable behaviour, so a strong explicit under relaxation is needed to obtain its convergence. The solution procedure is reported in Fig. 9.1 where it can be easily spotted the presence of two iterative loops, one to solve the Marshak 's equation with G eld from P1 last solution, and the other to solve the RTE with the last calculated boundary conditions. Marshak 's condition is suitable for diusive surfaces, but P1 equation can be solved together other kind of boundary conditions such as inlet, outlet, symmetry and cyclic that can be used in their normal form. The wall heat ux, that exists only when emissivity is higher than zero, is not directly calculated as the wall normal component of Eq. 6.35 again due to the singularities occurring as k 0. To avoid singularities Eq. 6.35 has been
190
Chapter 9.
rearranged by the use of Marshak 's boundary condition (Eq. 6.34), that has the RHS written in term of n G as well as the wall heat ux that becomes:
qw = n qwR =
1 n 3k
G|w =
2 (2 )
(4Ib (Tw ) G)
191
Chapter 9.
4 1 + 3 2 ln 2 T 4 T2 2 = 4 4 T1 T2 1 + 2 + 3 2 ln 2 1 2 1
as function of radial coordinate r ( = kr) once with the values 1 = kR1 and 2 = kR2 . The testcase presented by (Modest, 2003) gives results in adimensional form. The present validation procedure has been performed with cylinders having R1 = 0.25 m, R2 = 0.5 m and a height of 3 m. The mesh is a multiblock structured one with 10 equally spaced elements in radial direction, 36 equally spaced elements in circumferential direction and 29 equally spaced elements in axial direction. Four sets of simulations has been carried out with dierent cylinders temperatures (table 9.1), and the same range of optical thicknesses. The two cylinders case a) b) c) d)
have been represented by two black diusely emitting-absorbing surfaces then Marshak 's boundary condition is there imposed, while top and bottom are represented by symmetry-planes in order to simulate innite cylinders. The temperature elds have been dened by a properly realized utility. The results, represented by the nondimensional radiative heat ux at inner cylinder
simulation =
qw (r = R1 ) 4 4 (T1 T2 )
analytical =
1+
2 1
2 + 3 2 ln 2 2 1
where = 1 = kR1 . The percentage errors versus the optical thickness = k(R1 R2 ) = 0.25k are reported in table 9.2 . The error is expressed as:
%Err =
Simulation results are characterized by small errors (less than 3%) that increase with optical thickness while the dierence between the temperatures of the two cylinders compensates the error.
192
Chapter 9.
case a) b) c) d)
Table 9.2: P1 cylinders: %Error on nondimensional heat ux P si versus optical thickness
The matching of analytical solutions implies the good behaviour of the numerical model but it is important to state that this does not imply the good prediction of an eventual analytical solution of the exact form of the RTE.
193
Chapter 9.
4n n
The LHS requires more attention: the divergence operator integration is performed by FOAM, but again the directional treatment has to be explicitly done. Directional treatment requires the computing of Eq. 6.42 on every faces of every mesh cells, that is done by considering again the division setted by the user. Boundary conditions for diusely absorbing-emitting opaque surfaces are imposed through Eq. 6.45 that also requires the computation of Eq. 6.42 and furthermore its component normal to boundary surface Di nq . Every direction contribution to the second term of RHS is added to the stored previouses directions value, as well as is done for heat ux (Eq. 6.6) and enthalpy heat sink (Eq. 6.48). The condition for an opaque surface is not completely exhaustive: aim of the present work is to develop radiative heat transfer modelling coupled with combustion modelling; hence conditions, such as inlet, outlet, symmetry and cyclic, may be needed. Literature usually deals with only diusive emitting-absorbing surfaces, and it is common use to represent inlets and outlets as diusive surfaces with zero emissivity since neither absorption nor emission occur on these patches, except for those of participating media that is considered by RTE iteself. Instead symmetry and cyclic conditions, in their rigorous formulations, are hardly implemented due to the directional nature of radiative exchange: - symmetry plane is the same of a mirror so every beam is specular reected then, on symmetry planes, directions result coupled. And worse a beam
194
Chapter 9.
impinging on a patch might not have a specular outgoing direction among those considered, then interpolations among directional radiative intensities are needed: the algorithm is memory consuming and convergence could be hardly achieved. - a cyclic condition can be represented by a transmissive surface and same drawbacks of symmetry have to be faced: directions are coupled, interpolations are needed. These considerations, at least for symmetry conditions, give the necessity and the possibility to simplify. Indeed, if a symmetry is a reecting surface it could be represented by a diusive surface having null emissivity, obviously the neglection of directionality would give some error but we have demonstrated that this approximation does not give poor results (see subsection 9.2.3.2). If combustion processes would be taken into account, other sources of errors would be introduced such as absorption coecients calculations and neglection of temperature oscillation due to RANS approach, each of them of big inuence, so the error due to symmetry treatment as diusive surface might be one of the less important. In present work cyclic condition treatment has not been handled.
195
Chapter 9.
If,total =
S
If dS = If S Ib es ds = Ib,w e + Ib (1 e )
(9.1) (9.2)
with
If = Ib,w eh +
0
196
Chapter 9.
where Ib,w is the lower surface black body emission, Ib is the media absorption coecient and is the media optical thickness (Eq. 6.20). The nite volume discretization assumes that the radiative intensity is constant across the whole AB surface but it can be easily demonstrated that radiative intensity at point A is equal to Ib,w while at point B is equal to the value calculated with Eq. 9.2, and the dierence between them grows aside the optical thickness of the media (Eq. 6.20). This error is surely increased by the step (upwind) scheme, since it simply puts on the AB face value the C1 one, however (Joseph et al., 2005) this error was demonstrated to be present even when using the exponential scheme that exactly solves the RTE for each cell, while hybrid interpolation scheme able to solve the problem (Hong-Shung, 2004) has been developed. This simple example allows to understand that the solution to false scattering comes from a mesh renement and after by the use of a more sophisticated exponential scheme, but this is not always feasible, due to oscillations and the usual heaviness of schemes such as the exponential one. While dealing with non isothermal media characterized by a variable absorption coecient, that is the usual case of combustion systems, the radiative exchange take place at shorter distances than those between surrounding surfaces so the eective optical thickness is reduced and false scattering phenomenon is mitigated. This, together with the advanced schemes, induces the use of only the step-upwind scheme in present work.
ENEL Produzione - Research (ENEL), Pisa, Italy; Discrete Transfer Method (DTM) and Discrete Ordinates Method (DOM) Lehrsthul fuer Stroemungsmechanik (LSTM), Erlangen, Germany; MonteCarlo Method (MC) Instituto Superior Tecnico (IST), Lisbon, Portugal; Discrete Transfer Method (DTM) and Discrete Ordinates Method (DOM) International Flame Research Foundation (IFRF), IJmuiden, Netherland; Boundary Element Method (BEM).
The present testcase represents a well-stirred combustion chamber. Its geometry is quite simple (Fig. 9.4): it is a rectangular combustion chamber with size 1m x 1m x 3m. Participating media temperature is uniform with a value of 1773
197
Chapter 9.
K while its absorption coecient is equal to 0.1 m1 , wall temperature is 1273 K everywhere and wall emissivities are equal to 0.5 and 0.8 for two dierent simulations. The burner opening and furnace outlet shown in Fig. 9.4 are ignored in the calculations. Simulations have been performed on a structured mesh made of 30 x 30 x 90 uniformly distributed cubic control volumes. Both runs were performed with 16 directions (n = 2 and n = 2) and upwind-STEP was the interpolation scheme. The sampled results are the wall
sivity=0.5
Figure 9.5: RADIARE perfectly stirred combustor wall radiative heat ux contourt. Emis-
radiative uxes on one of the side-walls (all four having the same values) centerline parallel to combustor axis (Fig. 9.4), with a pitch of 0.25 m. Wall radiative ux represents the wall-normal component of radiative heat ux vector (Eq. 6.6). Obtained results are compared to those by DOM and DTM models, both implemented by either ENEL or Instituto Superior Tecnico (IST). DOM model has been chosen because is very similar to the used FVM (see subsec. 6.5.5), while DTM is very accurate and is often used to validate other solution methods. In Fig. 9.6 and Fig. 9.7 are plotted the results of the implemented model (HTC-FVM) and of benchmark simulations. For sections closer to the combustor inlet and outlet results from every solution method are more or less identical with both emissivities. At intermediate distances the wall heat ux is underestimated in magnitude for both emissivities, this is an eect of the false scattering and of the diusivity of upwind, the ENEL-DOM shows a similar but stronger trend enhanced by the higher emissivity, anyway the lack of details on simulation settings used for benchmark models makes impossible
198
Chapter 9.
-30
-31
-32
-33
-34
-35 0,25 0,50 0,75 1,00 1,25 1,50 1,75 2,00 2,25 2,50 2,75
Figure 9.6: RADIARE perfectly stirred combustor wall radiative heat ux vs z coordinate.
Emissivity=0.5
-32
-33
-34
-35
HTC-FVM
-36
-37
-38
-39 0,25 0,50 0,75 1,00 1,25 1,50 1,75 2,00 2,25 2,50 2,75
Figure 9.7: RADIARE perfectly stirred combustor wall radiative heat ux vs z coordinate.
Emissivity=0.8
deeper analyses.
9.2.3.2 Case 2
The second testcase also belongs to RADIARE programme and concerns a cylindrical combustion chamber. The dimensions of the furnace are 1 m in diameter and 3 m in length (Figure 9.8). The furnace is axially red and symmetrically cooled. The constant wall emissivity is equal to 0.7 and absorption coecient of the medium is equal 0.1 m1 . The temperature distribution of the wall and the medium temperatures is known and dened (T (x, y, z)), by keeping in consideration gure 9.8, by the
199
Chapter 9.
following equations:
g(x, y, z) =
L 1 exp m0 z 2
ln(z/L) + m0 s
x2 + y 2 (R)
2
where R is the radius of the chamber, L is the Lent of the furnace, m0 and s are the parameters determining the the position of the maximum temperature along the axis Z and is a parameter determining the jet width and the spread of the jet. All parameters assumes the values reported in table 9.3, it should be noticed that two ame lengths are considered by parameter m0 . In the above
T1 Ta m0 s
Figure 9.9: RADIARE case 2 temperature distribution for short-ame (upper) and longame (lower)
equations temperatures are expressed in degrees Celsius then, in order to be used in RTE code, they have to be translated in the Kelvin scale.
200
Chapter 9.
Four simulations were performed: we have considered both ame lengths on two meshes, both structured. One mesh is representative of the whole combustion chamber and is made of 9 control volumes in radial direction, 72 on circumferential direction and 49 in axial direction. The second one is representative of a quarter of the domain and obtained by splitting the combustion chamber, and the mesh, with two planes intersecting each other on the chamber axis (Fig. 9.10). These two meshes have been used in order to evaluate the eect of approximations made for symmetry planes (see sec. 9.2.1).
For all simulations walls emissivity was uniform and equal to 0.7. Simulations were performed on 64 directions (n = 4 and n = 4) and with upwind-STEP as interpolation scheme. The ame structure, i.e. the temperature eld, has been dened by using a properly realized utility. In this case are considered the irradiances (Eq. 6.4) on a generating line of the cylinder with a pitch on 0.25 m on both sides. Obtained results are compared to DOM and DTM from RADIARE even if the ENEL-DOM is not available. Both Fig. 9.11 and Fig. 9.12 show the good agreement with benchmark models of the simulations performed on the whole mesh. Results are better than those obtained through the RADIARE testcase 1, this can surely be partially attributed to the false scattering indeed, since the temperature eld is not uniform, radiative exchange also takes place between zones with dierent gases temperature and not only between gases and surfaces as happened in the isothermal domain, so the eective optical thickness is reduced implying a false scattering softening. This is conrmed by the better performances while analysing the longer ame where also the distances between ame and surfaces are reduces, while for the shorter ame the error increases farther from the ame. Greater discrepancies between the two meshes were found in those zone closer to the ame where walls reect more being aected by a higher irradiance, this explain the larger dierences for the longer ame. Anyway the adopted approximation for symmetry-plane (see sec. 9.2.1) does not give unacceptable errors.
201
Chapter 9.
150
IST DOM
145
140
135
130
125
120
115
110 0,25 0,50 0,75 1,00 1,25 1,50 1,75 2,00 2,25 2,50 2,75
As well as for case 1 the lack of further details on benchmark models simulations does not allow to go deeper in the analysis, so it cannot be established how discrepancies among models are aected by the denition of the temperature elds, that may change with mesh renement level. After this analysis it can be stated that both RADIARE testcases have shown the good behaviour of the implemented model while used to analyze optically thin systems, clearly showing the eect of false scattering moreover the representation of symmetry-planes with a diusive absorbing reecting emitting wall has been shown not to compromise the results.
q(r = 0) = 4k Ib (r = 0)
Ib,w ekR + k
0
Ib (r)ekr dr
(9.3)
4kekR Tm 4 Tw 4
Furthermore in order to study the eect of false scattering this testcase is intented to show the behaviour of the implemented RTE code while used on
202
Chapter 9.
160
155
150
145
140
IST DOM
135
130
125 0,25 0,50 0,75 1,00 1,25 1,50 1,75 2,00 2,25 2,50 2,75
unstructured meshes like the one on which simulations have been carried out that is made of 130088 tetrahedral. The used interpolation scheme is upwind-STEP. Either or are divided in 20 parts for a total of 1600 directions, that large number of direction were required to obtain as uniform as possible elds value on the wall. Runs were performed with a wide range of absorption coecient values to simulate dierent optical thickness (Eq. 6.20). In this testcase the radiative enthalpy sinks obtained by simulations (Eq. 6.48) are compared to those obtained by analytical solutions (Eq. 9.3). The comparison is performed by calculating the error as:
%Err =
q(r = 0)simulation
q(r = 0)exact
q(r = 0)exact
100
whose results are plotted in Fig. 9.14. The error exponentially grows aside optical thickness, evidencing a strong eect of false scattering. For kabs < 3, that is representative of case involving combustion, error is below 10%.
203
Chapter 9.
100 90 80 70 60 50 40 30 20 10 0
2005]
Error
-1
4
k
5
[
1/m]
10 11
Same trends, or worse, were obtained by (Joseph et al., 2005). Moreover this testcase evidenced the need of large number of discretization directions with unstructured meshes. This testcase beside having shown the behaviour of the implemented model while analysing optically thick systems, has demonstrated the capability to use unstructured meshes, anyway sided by the requirement of a great number of directions.
204
Chapter 10
205
Chapter 10.
k=
j=1
pj aj
(10.1)
where:
I
aj =
i=0
Cji
1000 T
specie CO2 H2 O
206
Chapter 10.
made available by Mauss. Here is reported the description of the performed implementation procedures. It has to be stated that soot is not considered in amelet libraries so its eect on combustion processes is taken in account only through its enhancing eect on radiative heat exchange. The soot mass fraction (Ys ) is given by a proper conservation equation:
ui Ys Ys + t xi
xi
T Ys Scs xi
+ s
dfv dt
(10.2)
that has got similar features of other species conservation equations (Eq. 5.1) f f with the term ddtv representing the reaction rate. This equation is not correct from a theoretical point of view, as conservation equations are derived from the Eulerian analysis of continuum systems, and if this approach is practically correct for uids, its use to represent a particulate matter cannot be completely justied. The reaction rate is expressed as:
fv =
Ys s
(10.3)
In Mauss utility the soot reaction rate terms are given by a pdf of mixture fraction and scalar dissipation rate (Eq. 5.27), as is done for diusion ames (section 5.5.6).
207
Chapter 10.
Figure 10.1: Diusion ame combined with soot and radiative heat exchange modelling
owchart
The global absorption coecient and temperature are given to the RTE solver (see chap. 9.2 and 9.1) that gives back the radiative enthalpy sink (Eq. 6.48) that is used, by the CFD solver, in the enthalpy equation (5.3).
208
Chapter 10.
of predicting quenching that makes the pilot necessary in the experimental apparatus. As well as has been done by (Da Soghe, 2006) the fuel is represented
by the civil aeronautical Jet-A, the chemical mechanism is the 16 steps by the NASA and the Mauss' utility represents it as the normal-Decane (C10 H22 ). The mesh is the same used by (Da Soghe, 2006) and is made of a 10 portion of a cylinder (bounded by two symmetry planes) and is made of 29000 hexaedrons. All wall are considered adiabatic from the convection point of view. While the pipe is a black wall for RTE solution. The outlet pressure is atmospheric. In table 10.2 are reported the boundary conditions applied considering that (Z. Wena et al., 2003) has shown that proles of inlets velocities can be considered uniform. The second coecient of turbulent energy dissipation equation
U [m/s]
22.28 0.234
k [m2 /s2 ]
0.6701 7.3920 105
[m2 /s3 ]
4.507 5.2220 106
T [K]
598 288 0 0
C2 (relative to the dissipation term) is changed from 1.92 to 1.84 in order to take in account the jet spreading. While comparing with ST AR CDT M results it is important to consider that (Da Soghe, 2006) did not perform this modication since that work was aimed to evaluate the dierent behaviour of combustion models, chemical mechanism and radiative exchange model rather than matching experimental results. Two runs where performed, one with the Finite Volume Method (with n = 2 and n = 6 for a total of 48 directions) and the other with the P1 method. Experimental values (J.B Moss et al., 1994), are available for: Mixture fraction Temperature
209
Chapter 10.
210
Chapter 10.
1,1 1,0 0,9 0,8 0,7 0,6 0,5 0,4 0,3 0,2 0,1 0,0 -0,1 0 100 200 300 400
EXP STAR-CD DOM OpenFOAM FVM OpenFOAM P1
Mixture Fraction Variance [-]
0,018
0,014
0,010
0,008
0,006
0,004
0,002
0,000 0 50 100 150 200 250 300 350 400 450 500 550 600
0,15 0,14 0,13 0,12 0,11 0,10 0,09 0,08 0,07 0,06 0,05 0,04 0,03 0,02 0,01 0,00 HTC FOAM FVM HTC FOAM P1 HTC STAR-CD
1400
Temperature [K]
1300 1200 1100 1000 900 800 700 600 500 0 50 100 150 200 250 300
EXP STAR-CD DOM OpenFOAM FVM OpenFOAM P
350
400
50
100
150
200
250
300
350
400
450
500
550
600
0,08
1,2x10
-5
0,07
EXP
[-]
1,0x10
-5
0,06
O [mass fraction]
8,0x10
-6
0,05
0,04
6,0x10
-6
0,03
0,02
4,0x10
-6
2,0x10
0,01
-6
0,00 0 50 100 150 200 250 300 350 400 450 500 550 600
0,5
6
0,4
-1
0,3
0,2
0,1
-0,2
100
200
300
400
500
600
100
200
300
400
500
600
211
Chapter 10.
0,40
0,007
0,35
0,006
0,30
0,005
OpenFOAM P1
0,004
0,20
0,003
0,15
0,002
x=100 [mm]
0,10
0,001
x=100 [mm]
0,05
HTC FOAM P1
0,00 0 5 10 15 20 25
-0,001 0 5 10 15 20 25 30
1800 1600
Temperature [K]
[mass fraction]
0,10
x=100 [mm]
0,08
0,06
x=100 [mm]
0,04
CO
HTC FOAM-FVM
0,02
0,00
10
15
20
25
30
10
15
20
25
30
0,06
2,5x10
-6
0,05
2,0x10
-6
H O [mass fraction]
0,04
[-]
1,5x10
-6
OpenFOAM P1
0,03
x=100 [mm]
1,0x10
-6
0,02
x=100 [mm]
HTC FOAM-FVM HTC FOAM-P1 HTC STAR-CD
0,01
5,0x10
-7
0,00
0,0 0 5 10 15 20 25
10
15
20
25
30
0,30
0,25
-1
0,20
HTC STAR-CD
x=100 [mm]
0,15
x=100 [mm]
0,10
0,05
0,00
10
15
20
25
30
10
15
20
25
30
sink that become negative on ame-air border, and being very close either the
212
Chapter 10.
0,14
1,1x10 1,0x10
-3
-3
0,12
ce [-]
-4
-4
Var a
in ion
xture Fract
0,10
-4
0,08
-4
-4
0,06
x=300 [mm]
-4
0,04
-4
Mi
x=300 [mm]
0,02
-4
M M M
FV P1
-4
HTC STAR-CD
0,00
0,0
0 10 20 30 40 50 60
10
20
Rad al d
i is n
ta
30
40 ce [mm]
50
60
0,14
[mass fraction]
0,10 0,09 0,08 0,07 0,06 0,05 0,04 0,03 0,02 0,01 0,00 -0,01
x=300 [mm]
HTC FOAM-FVM HTC FOAM-P1 HTC STAR-CD
x=300 [mm]
CO
10
20
30
40
50
60
70
80
1,0x10
0,07
-5
9,0x10
0,06
-6
8,0x10
-6
[-]
H O [mass fraction]
0,05
-6
x=300 [mm]
-6
0,04
-6
0,03
x=300 [mm]
HTC FOAM-FVM
-6
-6
0,02
-6
0,01
HTC FOAM-P1
-6
HTC STAR-CD
0,00 0 10 20 30 40 50 60 70 80
OpenFOAM P1
0,0 0 5 10 15 20 25
6 6 6 6 6
1,2x10 1,1x10 1,0x10 9,0x10 8,0x10 7,0x10 6,0x10 5,0x10 4,0x10 3,0x10 2,0x10 1,0x10
-1
6 6 5 5 5 5 5 5 5 5 5
0,35 0,30 0,25 0,20 0,15 0,10 0,05 0,00 -0,05 -0,10
x=300 [mm]
10
15
20
25
30
35
40
45
50
55
60
10
20
30
40
50
60
213
Chapter 10.
1,2
1,4
Absorption coefficient [m ]
1,0
-1
1,2
Absorption coefficient [m ]
-1
0,8
1,0
0,8
HTC FOAM-FVM HTC FOAM-P1
0,6
0,6
0,4
x=100 [mm]
0,4
0,2
0,2
x=300 [mm]
0,0 0 5 10 15 20 25 30
0,0 0 10 20 30 40 50 60 70
2,0
HTC FOAM-FVM
Absorption coefficient [m ]
HTC FOAM-P1
-1
1,5
1,0
0,5
axial trend
temperatures or the absorption coecients, predicted by the two RTE models, these dierences have to be attributed to RTE models only. It can also be spotted a slight shifting toward the axis of mixture fraction variance, carbon dioxide and water vapour of results by FVM that induces a slight shifting of the absorption coecient (g. 10.6). P1 model gives lower values of soot reaction rate that has been found to be due to an higher value of scalar dissipation rate. Bigger dierences can be found at radius 300 mm far from burner (g 10.5), here mixture fraction has greater values than experiments and again this is a clear eect of enhanced jet penetration. The shapes of obtained results, but not the values, are closer to experiments than ST AR CDT M ones. The same thing has to be said for temperature that is higher with P1. Mixture fraction variance is higher than in ST ARCDT M and also has got steeper shapes. CO2 and H2 O have got lower concentrations than in (Da Soghe, 2006) that also had obtained more diusive shapes. Soot volume fraction is closer to experimental results than ST AR CDT M value, soot is oxidised and still produced while for ST AR CDT M it is only oxidised, this is due to the still high value of mixture fraction. Again P1 gives lower value of radiative enthalpy sink, enhanced by a smaller absorption coecient (g. 10.6), due to higher temperatures in those radial distances where soot volume fraction is negligible (r > 30 mm). By an overall analysis of results it can be stated that the spreading of the ame is lower than predicted by ST AR CDT M this can be noticed by the overprediction of mixture fraction on the axis and with the tighter shapes of mixture fraction variance, species mass fractions and temperature plots in radial distances, and has several times stated, this is an eect of the turbulence model modication. Figure 10.2 clearly shows that P1 model gives lower enthalpy sink and consequently larger temperature, in gure 10.9 are reported the contours
214
Chapter 10.
Figure 10.8: Jet-K burner : enthalpy radiative sink contours left P1, right FVM
Figure 10.9: Jet-K. From left to right contours of: mixture fraction Favre's average, hydroxide mass fraction, soot source term [s1 ] and soot mass fraction
of mixture fraction Favre average, hydroxide and soot mass fraction and soot reaction rate that clearly shows formation and oxidation zones. The results show that the implemented models behave well better than can be stated by simply looking at the reported graphs. Indeed many levels of approximation are present, and it could not be otherwise because of the made simplications. The fuel is a mixture so a complete modelling would be unaordable, hence the combustion mechanism is a semi-detailed one, the lack of species experimental concentrations impedes to directly evaluate its eects. The soot modelling is simplied since a detailed modelling would be heavy, and soot formation has not been completely understood yet.
215
Chapter 10.
The RTE models are both aected by approximation by their nature: the P1 is suitable to be used with optically thick systems and Jet-K is not one of these while the FVM is solved with an approximated form of symmetry conditions, the gases absorption coecient is calculated with a global model. The jet-like nature of the ame cannot be completely reproduced by the two equations turbulence models although some coecients modications improves results.
216
Conclusions
A RANS CFD multipurpose code was developed for heat transfer and combustion modeling. Object oriented continuum mechanics libraries based on OpenFOAM has been improved to predict heat transfer and turbulent combustion phenomena in gas turbine, many dierent turbulence and combustion models has been developed and calculations were performed on representative tests. Numerical results have been compared mainly with experimental data obtaining in some cases good agreement. Disagreements might be explain to a poor adherence of the model to the physics, typical of RANS simulations. An a priori knowledge of the models behavior in the dierent situations is thus required to decouple these eects. That is why models have always been tested against cases, representing the state of art for turbomachinery cooling and combustion modeling, known in literature. The real aim of this investigation was to render OpenFOAM as reliable as commercial codes in the numerical predictions for heat transfer simulations. So of primary importance was the adherence to experimental data in the cases where commercial solvers are in good agreement too. In fact for the in-house cases, a quite wide database of previous simulations, done with commercial solvers (CFX (2004); Sta (2004)), was available. Results of commercial and OpenFOAM solvers are in substantial agreement, with major disagreements concentrated in zones where trusting commercial solvers is questionable, meaning that more evident failures can be attributed to the lack of modeling more than incapacity in solving the model equations. In research and industrial environment, it is fundamental to correctly predict the eorts necessary to obtain a goal, in order to establish which are the most convenient developments to invest on. From this point of view, OpenFOAM surely oers great advantages to the programmers in comparison with other codes. With the acquired knowledge of language structure, programming is becoming faster and major objectives can be met. In implementing such models the adopted strategy has been mainly to separate the activity of programming, namely developing new models, and running, that is solving cases of interest. After this long testing OpenFOAM is proven very exible for the implementation of new, solver algorithms, boundary condition types and physical models.
217
Chapter 10.
Conclusions
Furthermore it shows a quite good stability and precision both for single processor and multi processor parallel calculations. Nowadays however OpenFOAM standard release is not ready to be treated as a ready to run code, its selection of models is too small and only rough estimates can be done. For complex cases such the ones presented in this thesis, specic model implementation is necessary. Theoretically there are no limitations on changes to be done in the code, all les are opened and built in a way to facilitate the necessary additions and modications. The problem stands in the complexity of the code itself: even if rough, it has a wide variety of dierent tools, and in the programming language often quite dicult to interpret. The used object oriented language results to be very exible for implementing new turbulence and combustion models, solver algorithms, boundary condition types and physical combustion models. The modied version of OpenFOAM was an ecient solution framework for CFD calculation and it take place as a real alternative to the CFD commercial software.
218
Bibliography
(2004). CFX Manual - version 10. ANSYS. (2004). StarCD Methodology - version 3.24. CD adapco Group. A. Azzi and B.A. Jubran (2001). Numerical modeling of lm cooling from short length strem-wise injection holes. Heat and Mass Transfer, 39, 345353. A. Azzi and D. Lakehal (2001). Perspectives in modeling lm cooling of turbine blades by transcending conventional two-equation turbulence models. In Trans. ASME. ASME Paper GT2005-68155. A. Quarmby and R. Quirk (1974). Axisymmetric and non-axisymmetric turbulent diusion in a plain circular tube at high schmidt number. Int. J. Heat Flui Flow, 17, 143. A.Azzi and D.Lakehal (2002). Perspectives in modeling lm cooling of turbine blades by transcending conventional two-equation turbulence models. ASME J. Turbomach., 124, 472484. Abe, K., Kondoh, T., and Nagano, Y. (1994). A new turbulence models for predicting uid ow and heat transfer in separating and reattaching owsi. ow eld calculations. International Journal of Heat Mass Transfer, 37, 139151. Acharya, S. (2007). Film cooling sciences and technology for gas turbines. In Steady and Unsteady RANS Film Cooling Predictions., Lecture Series 200607. von Karman Institute for Fluid Dynamics. A.Hoda and S.Acharya (2000). Predictions of a lm cooling jet in crossow with dierent turbulence models. ASME J. Turbomach., 122, 558569. Anderson, jr., J. D. (1995). Computational Fluid Dynamics, the basic with applications. McGraw-Hill, US. Andreini, A. (2004). Sviluppo di Modelli Numerici per L'Analisi Della Combustione Turbolenta Premiscelata Nelle Turbine a Gas. Ph.D. thesis, Universit degli Studi di Firenze, Dipartimento di Energetica.
219
Chapter 10.
BIBLIOGRAPHY
Andreini, A., Carcasci, C., and Magi, A. (2004). Heat transfer analysis of a wedge shaped duct with pin n and pedestal arrays: a comparison between numerical and experimental results. ASME Turbo Expo, (GT2004-53319). Andreini, A., Carcasci, C., Gori, S., and Surace, M. (2005). Film cooling system numerical design: adiabatic and conjugate analysis. ASME paper, (HT200572042). Andreini, A., Champion, J., Facchini, B., Mercier, E., and Surace, M. (2006). Advanced liner cooling numerical analysis for low emission combustor. 25T H International Congress Of The Aeronautical Sciences. Apsley, D. (2006). Computational hydrolics. Technical report, University of Manchester. Arcangeli, L., Surace, M., Tarchi, L., Coutandin, D., and Zecchi, S. (2006). Correlative analysis of eusion cooling systems. ASME Turbo Expo, (GT200690405). B., A., K. (1982). Temperature and concentration proles in fully turbulent boundary layers. International Journal of Heat and Mass Transfer, 24(2), 15411544. Bacci, A. and Facchini, B. (2007). Turbulence modeling for the numerical simulation of lm and eusion cooling ows. ASME Turbo Expo, (GT2007-27182). Bacci, A., Facchini, B., Innocenti, L., and Tarchi, L. (2005). Combined use of turbulators and enlarged pedestals in trailing edge cooling systems: an experimental and numerical analysis. 6th ETC conference, 7-11 march 2005, Lille, France, Paper 078_04/94. Borghi, R. (1990). Gas turbine combustion. Lecture Notes, Von Karman Institute for Fluid Dynamics. Bredberg, J. (2001). On two equation eddy-viscosity models. Technical Report Internal report 01/8, Chalmers University of Technology. Briggs, L., Miller, W., and Lewis, E. (1975). Ray-eect mitigation in discrete ordinate-like angular nite element approximations in neutron transport. Nuclear Science and Engineering, 57, 205 217. Carmine, E. D., Facchini, B., Mangani, L., Abba, L., Arcangeli, L., and Traverso, S. (2007). Dierent manufacturing solution of fan-shaped cooling holes part ii: Numerical analysis. IGTC 2007, (ABS-54). Carmine, E. D., Facchini, B., and Mangani, L. (2008). Investigation of innovative trailing edge cooling congurations with enlarged pedestals and square or circular ribs. part ii - numerical results. ASME Paper, (GT2008/51047). Chai, J., Lee, H., and Patankar, S. (1993). Ray eect and false scattering in the discrete ordinates method. Numerical Heat Transfer Part B, (24), 373 389.
220
BIBLIOGRAPHY
Chapter 10.
Chen, W. L., Lien, F. S., and Leschziner, M. A. (1996). Low-reynolds-number eddy-viscosity modelling based on non-linear stress-strain/vorticity relations. Engineering turbulence modelling and experiments, 3, 91100. Chien, K. Y. (1982). Predictions of channel and boundary-layer ows with a low reynolds number turbulence model. AIAA Journal, 20(1), 3338. Coelho, P. (2002). The role of ray eects and false scattering on the accuracy of the standard and modied discrete ordinates methods. Journal of Quantitative Spectroscopy & Radiative Transfer, (73), 231 238. Coles, D. E. and Hirst, E. A. (1969). Computation of turbulent boundary layers. In AFOSRIFPStanford Conference, volume 2. Stanford University. Cooper, D., Jackson, D. C., Launder, B. E., and Liao, G. X. (1993). Impinging jet studies for turbulence model assessment. part i: Flow-eld experiments. Int. J. Heat Mass Transfer, (36), 26752684. Da Soghe, R. (2005-2006). Studio della Combustione Turbolenta nell'ambito dei Combustori Aeronautici. Master's thesis, Universit degli Studi di Firenze, Facolt di Ingegneria. Dally, B. B., Masri, A. R., Barlow, R. S., and Fiechtner, G. J. (1998). Instantaneous and mean compositional structure of blu-body stabilized nonpremixed ames. Combustion and Flame, 114, 119148. Darbandi, M. and Schneider, G. E. (1999). Application of an all-speed ow algorithm to heat transfer problems. Numerical heat transfer, 35, 695715. Davenport, W. and Simpson, R. (1982). The ow past a wing-body junction - an experimental evaluation of turbulence models. International Journal of Heat and Mass Transfer, 24(2), 15411544. Davidson, L. (2003). An introduction to turbulence models. Technical Report Publication 97/2, Chalmers University of Technology. Davidson, L. (2005). Numerical methods for turbulent ows. Technical Report MTF071, Chalmers University of Technology. Davidson, L. (2006). Turbulence modeling. Chalmers University of Technology. Davis, S. R. (2005). C++. Wiley Publishing, Inc. Di Carmine, E. (2004). Studio Numerico del Rareddamento per Impingement nelle Turbomacchine: Simulazione di Geometrie Reali. Master's thesis, Universit degli Studi di Firenze, Facolt di Ingegneria. D.J.Ferguson and D.K. Walters and J.H. Leylek (1998). Performance of turbulence models and near-wall treatments in discrete jet lm cooling simulations. In Trans. ASME. ASME Paper GT2005-68155. Technical Report MTF270,
221
Chapter 10.
BIBLIOGRAPHY
D.Lakehal (2002). Near-wall modeling of turbulent convective heat transport in lm cooling of turbine blades with the aid of direct numerical simulation data. ASME J. Turbomach., 124, 485498. D.Lakehal and C.S. Theodoridis and W.Rodi (2001). Three-dimensional ow and heat transfer calculations of lm cooling at leading edge of a symmetrical turbine blade model. Int. J. Heat Flui Flow, 22, 113122. Doormal, J. P. V. and Raithby, G. D. (1984). Enhancements of the simple method for predicting incompressible uid ows. Numerical Heat Transfer, 7, 147163. Durbin, P. A. (1996). On the k stagnation point anomaly. International journal of heat and uid ow, 17(1), 8990. Durbin, P. A. and M. Benhia, S. P. (1998). Prediction of heat transfer in an axisymmetric turbulent jet impinging on a at plate. International journal of heat mass tranfer, 41(12), 18451855. D.Veynante and L.Vervisch (2002). Turbulent combustion modelling. Progress In Energy and Combustion Science, 28, 193266. Echekki, T. and J. H. Chen (1998). Structure and propagation of methanol-air triple ame. Combustion and Flame, 114, 231245. Elia, M., Ulinski, M., and Metghalchi, M. (2001). Laminar burning velocity of methane-air-diluent mixtures. Journal of Engineering for Gas Turbines and Power, 124(1), 190196. F. Mauss et. al. (2002). Implementation and validation of a new soot model and application to aeroegine combustors. Transaction of the ASME, 124. Facchini, B. and Surace, M. (2006). Impingement cooling for modern combustors: experimental analysis of heat transfer and eectiveness. Experiments in Fluids, (40), 601611. Facchini, B. and Tarchi, L. (2008). Investigation of innovative trailing edge cooling congurations with enlarged pedestals and square or circular ribs. part i - numerical results. ASME Paper, (GT2008/51047). Facchini, B., Surace, M., Tarchi, L., L.Toni, Abba, L., Arcangeli, L., and Traverso, S. (2007). Dierent manufacturing solution of fan-shaped cooling holes part i: Experimental analysis. IGTC 2007, (ABS-56). Ferziger, J. H. and Peric, M. (2002). Computational Methods for Fluid Dynamics. Springer, Germany. G. Bergeles and A. D. Gosman and B.E. Launder (1978). The turbulent jet in a cross stream at low injection rates: a three-dimensional numerical treatment. Num. Heat. Transf., 1, 217242.
222
BIBLIOGRAPHY
Chapter 10.
Gelbard, E. (1961). Simplied spheral armonics equation and their use in shielding problem. Technical report, Bettis Atomic Power Laboratory. Goldman, L. G. and McLallin, K. L. (1977). 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. Gritsch, M., Schulz, A., and Wittig, S. (2001). Eect of crossows on the discharge coecient of lm cooling holes with varying angles of inclination and orientation. ASME Turbo Expo, (2001-GT-0134). Gustafsson, K. M. B. (2001). Experimental studies of eusion cooling. Ph.D. thesis, Chalmers University of Technology, Department of Thermo and Fluid Dynamics. Hermann, M. (2006). Numerical simulation of turbulent bunsen ames with a level set amelet model. Combustion and Flame, 145, 357375. Hong-Shung, L. (2004). Reduction of false scattering in arbitrarily specied discrete directions of the discrete ordinates method. Journal of Quantitative Spectroscopy & Radiative Transfer, (86), 215 222. Hossain, M., J.C. Jones, and Walalasekera, W. (2001). Modelling of a blu-body nonpremixed ame using a coupled radiation/amelet combustion model. Flow, Turbulence and Combustion, 67, 217234. Hwang, C. B. and Lin, C. A. (1998). Improved low-reynolds-number k model based on direct numerical simulation data. AIAA Journal, 36(1), 3843. Iacovides, H. (2004). Current practice and recent developments in wall functions. Technical report, Department of mechanical Aerospace and Manufactoring Engineering UMIST. Incropera, F. P. and DeWitt, D. P. (2001). Fundamentals of Heat and Mass Transfer, 5th Edition. Ireland, P. T. and Jones, T. V. (1986). Detailed measurements of heat transfer on and around a pedestal in fully developed passage ow. 8th AIAA/ASME Joint Thermophysics and Heat Transfer Conference, 3, 975980. Issa, R. I. (1986). Solution of the implicitly discretized uid ow equations by operator splitting. Journal of Computational Physics, 62, 4065. Jasak, H. (1996). Error Analysis and Estimation for the Finite Volume Method with Applications to Fluid Flows. Ph.D. thesis, Imperial College of Science, Technology and Medicine. Jasak, H., Weller, H., and Nordin, N. (2004). In cylinder cfd simulation using a c++ object-oriented toolkit. SAE Technical Papers, (2004-01-0110).
223
Chapter 10.
BIBLIOGRAPHY
J.B Moss, K.J. Young, and C.D. Stewart (1994). Soot formation in turbulent nonpremixed kerosine-air ames burning at elevalte pressure: Experimental measurement. Twenty-fth Symposium (International) on Combustion/The Combustion Institute, pages 609617. Jeans, J. (1917). The equation of radiative transfer of energy. Monthly Notices Royal Astronomical Society, 78, 28 36. J.Kim and P.Moin and R.Moser (1987). Turbulence statistics in fully developed channel ow at low reynolds number. J. Fluid Mech, 177, 133166. Jones, W. P. and Launder, B. E. (1972). The prediction of laminarization with a two-equation model of turbulence. Int. J. Heat and Mass Transfer, (15), 301314. Joseph, D., Ha, M. E., Fournier, R., and Cuenot, B. (2005). Comparison of three spatial dierencing schemes in discrete ordinates method using threedimensional unstructured meshes. International Journal of Thermal Sciences, (44), 851 864. Juretic, F. (2004). Error Analysis in Finite Volume. Ph.D. thesis, Imperial College of Science, Technology and Medicine. Karki, K. (1986a). A calculation procedure for viscous ows at all speeds in complex geometries. Ph.D. thesis, University of Minnesota. Karki, K. C. (1986b). A Calculation Procedure for Viscous Flows at all Speed in complex Geometries. Ph.D. thesis, University of Minnesota, Faculty of the Graduated School. Karki, K. C. and Patankar, S. V. (1989). Pressure based calculation procedure for viscous ows at all speeds in arbitrary congurations. AIAA Journal, 27(9), 11671174. Knight, D. D. (1997). Numerical simulation of compressible turbulent ows using the reynolds-averaged navier-stokes equations. AGARD, R(819), 5.1 5.22. AGARD FDP Special Course on "Turbulence in Compressible Flows" at VKI. K.P. Helbig, Coelho, P., Chamio, J., Marakis, J., Mbiock, A., Weber, R., and Pasini, S. (2001). Comparison of radiative heat transfer methods using benchmark test case. In M. Meng and N. Seluk, editors, Radiative Transfer-III, pages 102110. Begell House Inc., New York. Lacor, C. (2006). Solution of time dependent reynolds averaged navier-stokes equations with the nite volume method. Technical report, Vrije Universitiet Brussel. Lam, C. K. G. and Bremhorst, K. (1981). A modied form of the k model for predicting wall turbulence. Journal of uids engineering, 103, 456460.
224
BIBLIOGRAPHY
Chapter 10.
Larsen, E., Thmmes, G., Klarc, A., Sead, M., and Gtz, T. (2002). Simplied pn approximations to the equations of radiative heat transfer and applications. Journal of Computational Physics, 183, 652 675. Lautenberger, C. W. (2002). Cfd Simulation of Soot Formation and Flame Radiation. Master's thesis, Worcester Polytechnic Institute. Lecturer, V. K. I. (2002). Introduction to Turbulence Modeling. Von Karman Institute Lecture Series, von karman institute lecture series edition. Lien, F. S. (1992). Computational modeling of 3-D ow in complex ducts and passages. Ph.D. thesis, University of Manchester, Institute of science and technology. Lien, F. S. and Leschziner, M. A. (1993). A pressure-velocity solution strategy for compressible ow and its application to shock/boundary-layer interaction using second-moment turbulence closure. Journal of uids engineering, 115, 717725. Magnussen, B. and Mjertag, B. (1976). On mathematical modelling of turbulent combustion. In 16th Symp. (Int.) on combustion, pages 719727. The Combustion Institute, Pittsburgh. Malalasekera, W. and Versteeg, H. K. (1995). Computational Fluid Dynamics. Longman Scientic, England. Mangani, L. (2002-2003). Sviluppo di un Modello per la Combustione Turbolenta Premiscelata basato su Level Set Flamelet Method. Master's thesis, Universit degli Studi di Firenze, Facolt di Ingegneria. Mangani, L. (2006). Openfoam: Basic theory and preliminary result. Technical report, University of Florence, Energy Engineering Department - Ansaldo Energia. Mangani, L., Bianchini, C., Andreini, A., and Facchini, B. (2007). Development and validation of a c++ object oriented cfd code for heat transfer analysis. ASME Summer Heat Transfer, (AJ-1266). Masri, A. (1998). Computation of blu-body stabilised jets and ames. In Third International Workshop on Measurement and Computation of Turbulent Nonpremixed Flames Boulder, Colorado. McGuirk, J. J. and Page, G. (1990). Shock capturing using a pressure-correction method. AIAA Journal, 28(10), 17511757. Medic, G. and Durbin, P. A. (2002). Towards improved prediction of heat transfer on turbine blades. Journal of turbomachinery, 124, 187192. Menter, F., Thomas, E., and Wolfgang, V. (16-20/03/2003). Heat transfer predictions based on two-equation turbulence models. ASME-JSME 2003 Thermal Engineering Joint Conference.
225
Chapter 10.
BIBLIOGRAPHY
Menter, F. R. (1993). Zonal two equation k turbulence models for aerodynamic ows. AIAA Paper, (93-2906). Menter, F. R. (1994). Two equation eddy viscosity turbulence model for engineering applications. AIAA Journal, 32, 15981604. Metzger, D. E., Shepard, W. B., and Haley, S. W. (1986). Row resolved heat transfer variations in pin-n arrays including eects of non-uniform arrays and ow convergence. ASME Paper, (86-GT-132). Mishra, S. and Roy, H. (2007). Solving transient conduction and radiation heat transfer problems using the lattice boltzmann method and the nite volume method. Journal of Computational Physics, 223, 89107. Modest, M. F. (2003). Radiative Heat Transfer. Academic Press, US, 2nd edition. Moukalled, F. and Darwish, M. (1994). Normalized variable and space formulation methodology for high-resolution schemes. Numerical heat transfer. Part B, fundamentals, 26, 7996. Moukalled, F. and Darwish, M. (1999). A high-resolution pressure-based algorithm for uid ow at all speeds. Technical Report SPC-99-4003, European Oce of Aerospace Research and Development. Moukalled, F. and Darwish, M. (2000). A unied formulation of the segregated class of algorithms for uid ow at all speeds. Numerical heat transfer. Part B, fundamentals, 37, 103139. Moukalled, F. and Darwish, M. (2003a). A robust pressure based algorithm for multiphase ow. International journal for numerical methods in uids, 41, 12211251. Moukalled, F. and Darwish, M. (2003b). Tvd schemes for unstructured grids. International Journal of heat and mass transfer, 46, 599611. Murthy, J. Y. and Mathur, S. R. (2002). Numerical methods in heat, mass and momentum tranfer. Technical Report ME 608, School of Mechanical Engineering Purdue University. Nandula, S. P. (2003). University of Vanderbilt, PhD Thesis. OpenCFD (2006a). OpenFOAM Programmers Guide. OpenCFD Limited. OpenCFD (2006b). OpenFOAM User Guide. OpenCFD Limited. OpenCFD (2006c). http: // www. opencfd. co. uk . OpenCFD Limited. Pasquini, M. (2005-2006). Soot modelling in jet engine combustors. Master's thesis, Universit degli Studi di Pisa, Facolt di Ingegneria.
226
BIBLIOGRAPHY
Chapter 10.
Patankar, S. V. (1980). Numerical Heat Transfer and Fluid Flow. Taylor & Francis, US. Patankar, S. V. and Spalding, D. B. (1972). A calculation procedure for heat, mass and momentum tranfer in three-dimensional parabolic ows. International Journal of Heat and Mass Tranfer, (15), 17871805. Patel, V. C., Rodi, W., and Sheuerer, G. (1985). Turbulence models for near wall and low reynolds number ows: a review. AIAA Journal, 23(9), 13081319. Peric, M. (2003). Numerical methods for computing turbulent ows. Technical Report VKI LS 2004-06. Peric, M., Lilek, Z., and Demirdzic, L. (1993). A collocated nite volume method for predicting ows at all speed. Journal of Numerical Methods in Fluids, 16, 10291050. Peters, N. (2000). Turbulent Combustion. Cambridge University Press, UK. Poinsot, T. and Veynante, D. (2001). Theoretical and Numerical Combustion. Edwards, US, 1st edition. Politis, E. S. and Giannakoglou, K. C. (1996). A pressure-based algorithm for high speed turbomachinery ows. International journal for numerical methods in uids, 25, 6380. Raithby, G. and Chui, E. (1990). A nite volume method for predicting radiant heat transfer enclosure with participating media. ASME Journal of Heat Transfer, 112(2), 415 423. Rhie, C. M. (1989). Pressure based navier-stokes solver using the multigrid method. AIAA Journal, 27(8), 10171018. Rhie, C. M. and Chow, W. L. (1983). Numerical study of the turbulent ow past an airfoil with trailing edge separation. AIAA Journal, 21, 15251532. Rincon, J. and Elder, R. (1997). A high resolution pressure based method for compressible ow. Journal of Computation and Physics, 3, 217231. Rodi, W. (1991). Experience with two-layer models combining the k model with a one-equation model near the wall. Technical Report 91-0216, AIAA. S. Acharya and M. Tyagi and A. Hoda and F.Muldoon (1999). From RANS to DNS: Application to Film Cooling. 26th National Conference on Fluid Mechanics & Fluid Power, 15 - 17th Dec 1999, IIT, Kharagpur, INDIA. Saimweber, C. and Schulz, A. (2007). Film cooling sciences and technology for gas turbines. In Eects of of Film Hole Shaping and Freestream Conditions on eat Transfer Augmentation and Discharge Coecients, Lecture Series 200607. von Karman Institute for Fluid Dynamics.
227
Chapter 10.
BIBLIOGRAPHY
Senocak, I. and Shyy, W. (2001). A pressure based method fot turbulent cavitating ow computations. Journal of Computation and Physics, 176, 363383. Shyy, W. and Braaten, M. E. (1988a). Adaptive grid computation for inviscid compressible ows using a pressure correction method. AIAA, ASME, SIAM, and APS, National Fluid Dynamics Congress, pages 112120. Shyy, W. and Braaten, M. E. (1988b). Applications of a generalized pressure correction algorith for ows in complicated geometries. Advances and Applications in Computational Fluid Dynamics - ASME Winter Annual Meeting, pages 109119. Sinha, A. K., Bogard, D. G., and Crawford, M. E. (1991). Film-cooling effectiveness downstream of a single row of holes with variable density ratio. ASME Journal of Turbomachinery, 113, 442449. Stourstroup, B. (1997). The C++ programming language. Addison Wesley. Surace, M. (2004). Investigation of impingement systems for gas turbine combustor cooling. Ph.D. thesis, University of Florence, Energy Engineering Department. Trivic, D. (2004). Modeling of 3-d non-gray gases radiation by coupling the nite volume method with weighted sum of gray gases model. International Journal of Heat and Mass Transfer, (47), 1367 1382. Viskanta, R. (2005). Radiative Transfer in Combustion Systems: Fundamental & Applications. Begel House, inc., US. White, F. M. (1991). Viscous Fluid Flow. McGraw-Hill, US. Wieghardt, K. and Tillman, W. (1951). On the turbulent friction layer for rising pressure. Technical Report TM-1314, NACA. Wilcox, D. C. (1998). Turbulence Modeling for CFD, 2nd Edition. DCW Industries, US. W.Rodi and N.N.Mansour and V.Michelassi (1993). One-equation near-wall turbulence modeling with the aid of direct simulation data. ASME J. Fluids Eng., 115, 196205. Yoder, D. A. and Georgiadis, N. J. (1999). Implementation and validation of the chien k turbulence model in the wind navier-stokes code. Technical Report 209080, NASA, Glenn Research Center, Cleveland Ohio. Z. Wena, S. Yuna, M.J. Thomson, and M.F. Lightstoneb (2003). Modeling soot formation in turbulent kerosene/air jet diusion ames. Combustion and Flame, 135, 323340. Zhanga, H. and Shib, G. (2005). A new approach to solve correlated kdistribution function. Journal of Quantitative Spectroscopy & Radiative Transfer, (96), 311 324.
228
BIBLIOGRAPHY
Chapter 10.
Zimont, V., Polifke, W., Bettelini, M., and Weisenstein, W. (1998). An ecient computational model for premixed turbulent combustion at high reynolds numbers based on a turbulent ame speed closure. ASME Journal of Engineering for Gas Turbines and Power, 120, 526532.
229