Академический Документы
Профессиональный Документы
Культура Документы
BIBLIOGRAPHIC REFERENCE
TOMITA, Jesuino Takachi. Three-Dimensional Flow Calculations of Axial
Compressors and Turbines Using CFD Techniques. 2009. 229f. Thesis of Doctor
of Science Technological Institute of Aeronautics, Sao Jose dos Campos.
SESSION OF RIGHTS
AUTHOR NAME:
PUBLICATION TITLE:
Presidente
ITA
Prof. Dr.
Advisor
ITA
Membro Interno
ITA
Membro Externo -
UNIFEI
Membro Externo -
IAE
ITA
Agradecimentos
Durante os anos em que estudei no ITA, pude aprender e fazer muito mais do que eu
esperava, quando nesse Instituto entrei pela primeira vez. Tambem aprendi bastante
com os amigos que fiz no CTA. A soma de tudo que ganhei durante esse tempo e muito
conhecimento de vida.
mais do que conhecimento tecnico-cientfico. E
Agradeco a Deus, por permitir a conclusao de mais uma etapa da minha vida.
Agradeco o apoio de minha mae Izabel Tiyoka Tomita, que desde o primeiro dia de aula
da minha vida, me incentivou e me deu suporte para que eu pudesse concluir mais um
trabalho.
Agradeco a minha linda esposa Thaisa Talarico Hyppolito Tomita, que durante esses
anos, tem compreendido todo o meu esforco, durante in
umeras noites, fins de semana e
feriados, na qual fiquei gerando geometrias de compressores e turbinas, malhas e
compilando programas.
Agradeco a todos os turbineiros do Grupo de Turbinas do ITA (Cleverson, Franco,
Santin, Luciano, Gustavo, Dulceneia, Renato, Dora, Di Fiori, Helder, Marcio Mendonca,
e Daniel) e nao poderia deixar de agradecer o Demerval, pela amizade e companheirismo.
Reconheco toda a ajuda e todo o conhecimento que recebi do Professor Joao Roberto
Barbosa, que, sem d
uvida, e uma referencia internacional em turbinas a gas. Agradeco
por ter me dado a oportunidade de mergulhar no mundo das turbomaquinas. Espero
poder fazer o mesmo com as proximas geracoes.
Obrigado!
A natureza,
e a arte de Deus.
Dante Alighieri).
Resumo
Com o advento de potentes computadores, a Dinamica dos Fluidos Computacional
(DFC) tem sido vastamente utilizada por pesquisadores e cientistas para investigar o
comportamento de escoamentos e a variacao das propriedades dos mesmos. O custo de
simulacao de DFC e muito pequeno comparado com o arsenal experimental como bancos
de ensaio e t
uneis de vento. Nos u
ltimos anos, muitos pacotes comerciais de DFC foram
desenvolvidos, alguns deles possuem proeminencia na ind
ustria e na academia. Porem,
alguns calculos especficos de DFC sao casos muito particulares e a`s vezes necessitam de
atencao especial devido a complexidade do escoamento. Nesses casos, uma pesquisa meticulosa torna-se necessaria. Este e o caso do calculo de escoamentos em turbomaquinas.
O desenvolvimento de codigos de DFC aplicados em simulacoes de escoamentos em turbomaquinas e os detalhes das implementacoes sao assuntos muito reservados. Um pequeno
n
umero de instituicoes possui esse tipo de conhecimento. Cada codigo de DFC possue sua
particularidade. Desenvolver um codigo fonte particular e um assunto muito interessante
no senso academico.
Nesse trabalho, um codigo computacional escrito em FORTRAN, foi desenvolvido
para calcular escoamentos internos em turbomaquinas usando tecnicas da DFC. O programa e capaz de calcular escoamentos tridimensionais nao somente em turbomaquinas.
Por exemplo, escoamentos internos e externos como bocais e aerofolios podem ser cal-
viii
culados. O tratamento dado no codigo permite o uso de malhas nao-estruturadas com
elementos hexaedricos. Escoamentos envolvendo as equacoes de Euler, Navier-Stokes e
escoamentos turbulentos podem ser calculados, dependendo da necessidade do usuario.
Diferentes esquemas numericos foram implementados para a integracao no tempo e no
espaco. Metodos numericos para melhorar a estabilidade e aumentar o passo no tempo
(passo no tempo variavel, suavizacao implcita do resduo) foram tambem implementados
e todos os detalhes estao descritos nesse trabalho.
A origem do codigo computacional e para simular escoamentos em compressores e
turbinas. Dessa forma, o sistema de referencia rotacional e nao-rotacional e calculado
simultaneamente. Dessa forma, o proceso de verificacao e validacao do codigo foi realizado
para ambos os sistemas.
muito
Um procedimento de projeto, passo-a-passo, e apresentado nesse trabalho. E
importante mencionar que para o entendimento completo da fsica do escoamento em
compressores e turbinas o projetista deve possuir um solido conhecimento de operacao
dos componentes de uma turbina a gas.
Abstract
With the advent of powerful computer hardware, Computational Fluid Dynamics (CFD)
has been vastly used by researches and scientists to investigate flow behavior and its properties. The cost of CFD simulation is very small compared to the experimental arsenal as
test facilities and wind-tunnels. In the last years many CFD commercial packages were
developed and some of them possess prominence in industry and academia. However,
some specific CFD calculations are particular cases and sometimes need special attention
due to the complexity of the flow. In these cases, meticulous research becomes necessary.
This is the case of turbomachinery flow calculations. The development of CFD codes applied to turbomachinery flow simulations and its implementation issues are not available.
A few institutions have this type of knowledge. Each CFD code has its particularities.
Developing a CFD code is very interest subject in academia.
In this work, a computational code, written in FORTRAN, was developed to calculate
internal flows in turbomachines using CFD techniques. The solver is capable of calculating the three-dimensional flows not only for turbomachines. For instance, internal and
external flows of nozzles and airfoils can be calculated. The approach used allows the
use of unstructured meshes of hexahedral elements. Euler, Navier-Stokes and turbulent
equations can be calculated depending on the user settings. Different numerical schemes
were implemented for time and space integration. Numerical tools to improve the stability
x
and to increase the time-step (local time-step and implicit residual smoothing) were also
implemented and all details are described in this work.
The origin of this solver is to simulate flows in compressors and turbines. Therefore,
both rotating and non-rotating frames of reference are calculated simultaneously. Hence,
the verification and validation processes were run for both inertial and non-inertial systems.
A step-by-step design procedure is presented in this work. It is very important to mention that to have a complete understanding of the flow physics in compressors and turbines
the designer must have a solid knowledge of the operation of gas turbine components.
Contents
List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
xv
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
32
1.1
Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
1.2
Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
1.3
Previous Developments . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
1.3.1
1.3.2
1.3.3
1.4
Author Contribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
1.5
Work Organization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
Mathematical Formulation . . . . . . . . . . . . . . . . . . . .
51
2.1
The fluid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
2.2
The flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
2.2.1
Turbulence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
CONTENTS
2.2.2
3
3.1
3.1.1
3.2
xii
Turbulence Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
Numerical Formulation . . . . . . . . . . . . . . . . . . . . . . .
68
Discretization Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
Finite-Volume Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
Spatial Integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
3.2.1
3.2.2
3.2.3
3.2.4
. . . . . . . . . . . 76
Venkatakrishnans Limiter . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
3.2.6
3.3
Time Integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
3.3.1
3.3.2
3.4
3.4.1
Artificial Dissipation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
3.4.2
3.5
3.6
3.7
3.7.1
Initial Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
3.7.2
Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
3.7.3
CONTENTS
4
4.1
xiii
. . . . . . . . . . . . . . . 113
5.1
5.2
5.3
6
6.1
Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
Flow Simulation in Turbomachines . . . . . . . . . . . . . . . . . . . . . 146
6.1.1
6.1.2
6.2
6.2.1
6.2.2
6.2.3
6.2.4
7
7.1
8.1
Pre-Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
8.2
Solver . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
8.3
Post-Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
Appendix A Artificial Dissipation Issues . . . . . . . . . . . 225
CONTENTS
A.1
xiv
Glossary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
List of Figures
LIST OF FIGURES
xvi
FIGURE 5.7 Mesh generated for NACA0012 airfoil flow calculation . . . . . . . . 130
FIGURE 5.8 Pressure coefficient distribution on the NACA0012 airfoil with zero
angle-of-attack and Mn = 0.8: centered-difference . . . . . . . . . . 131
FIGURE 5.9 Pressure coefficient distribution on the NACA0012 airfoil with zero
angle-of-attack and Mn = 0.8: upwind . . . . . . . . . . . . . . . . . 131
FIGURE 5.10 Mach number contours for centered scheme . . . . . . . . . . . . . . 132
FIGURE 5.11 Mach number contours for upwind scheme . . . . . . . . . . . . . . 132
FIGURE 5.12 Static pressure contours for centered scheme . . . . . . . . . . . . . 133
FIGURE 5.13 Static pressure contours for upwind scheme . . . . . . . . . . . . . . 133
FIGURE 5.14 Mesh used to calculate the flow on a flat-plate . . . . . . . . . . . . 134
FIGURE 5.15 Continuity residue history: first case . . . . . . . . . . . . . . . . . . 136
FIGURE 5.16 Analytical and numerical solutions of flow with M = 0.3 on a
flat-plate with 10 nodes inside of the boundary-layer: first case . . . 136
FIGURE 5.17 Continuity residue history: second case . . . . . . . . . . . . . . . . 137
FIGURE 5.18 Analytical and numerical solutions of an flow with M = 0.3 on a
flat-plate with 15 nodes inside of the boundary-layer: second case . 137
FIGURE 5.19 Continuity residue history: third case . . . . . . . . . . . . . . . . . 138
FIGURE 5.20 Analytical and numerical solutions of an flow with M = 0.3 on a
flat-plate with 15 nodes inside of the boundary-layer: third case . . 138
FIGURE 5.21 Continuity residue history: fourth case . . . . . . . . . . . . . . . . 139
FIGURE 5.22 Analytical and numerical solutions of an flow with M = 0.3 on a
flat-plate with 10 nodes inside of the boundary-layer: fourth case . . 139
FIGURE 5.23 Continuity residue history: fifth case . . . . . . . . . . . . . . . . . . 140
FIGURE 5.24 Analytical and numerical solutions of an flow with M = 0.3 on a
flat-plate with 12 nodes inside of the boundary-layer: fifth case . . . 140
FIGURE 5.25 Analytical and numerical solutions of a flow with M = 0.3 on a
flat-plate with 15 nodes inside of the boundary-layer: sixth case . . 141
LIST OF FIGURES
xvii
FIGURE 5.26 Details of the Mach number contours inside of the boundary-layer
on the flat-plate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
FIGURE 5.27 Details of the velocity vectors profiles inside of the boundary-layer
on the flat-plate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
FIGURE 5.28 Mesh generated for nozzle flow calculation - turbulent flow . . . . . 143
FIGURE 5.29 Comparison of experimental and numerical results for static pressure
ratio 5 for centered and upwind schemes - turbulent flow . . . . . . 144
FIGURE 5.30 Continuity residue histories for two spatial discretization schemes . . 144
FIGURE 5.31 Mach number contours - turbulent flow . . . . . . . . . . . . . . . . 145
FIGURE 5.32 Velocity vectors - turbulent flow . . . . . . . . . . . . . . . . . . . . 145
FIGURE 6.1 Auxiliary curves to drawing a single-stage axial-flow turbine
. . . . 148
FIGURE 6.2 3-D solid drawing of the NGV and rotor of a single-stage axial-flow
turbine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
FIGURE 6.3 Full 3-D view of an single-stage axial-flow turbine . . . . . . . . . . 149
FIGURE 6.4 Single-stage axial-flow turbine H-grid domain . . . . . . . . . . . . . 149
FIGURE 6.5 Closeup of the leading edge of an H-grid . . . . . . . . . . . . . . . 150
FIGURE 6.6 Single-stage axial turbine H-O-H-grid domain . . . . . . . . . . . . . 150
FIGURE 6.7 Closeup of the stator trailing edge and rotor leading edge of a singlestage axial-flow turbine . . . . . . . . . . . . . . . . . . . . . . . . . 151
FIGURE 6.8 Single-stage axial-flow turbine O-grid domain . . . . . . . . . . . . . 151
FIGURE 6.9 Closeup of the O-grid around the stator and rotor blades of a singlestage axial-flow turbine . . . . . . . . . . . . . . . . . . . . . . . . . 152
FIGURE 6.10 Single-stage axial-flow turbine mass-flow outlet convergence history
- inviscid case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
FIGURE 6.11 Static pressure contour of a single-stage axial-flow turbine - inviscid
case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
LIST OF FIGURES
xviii
FIGURE 6.12 Closeup of the flow across the mixing-plane of a single-stage axialflow turbine - inviscid case . . . . . . . . . . . . . . . . . . . . . . . 154
FIGURE 6.13 Closeup of the flow at rotor suction and pressure surfaces - inviscid
case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
FIGURE 6.14 H-grid used for a single-stage axial-flow turbine . . . . . . . . . . . . 155
FIGURE 6.15 Mixing-plane outlet (MPO) and mixing-plane inlet (MPI) ratio turbulent case with H-grid . . . . . . . . . . . . . . . . . . . . . . . 156
FIGURE 6.16 Single-stage axial-flow turbine mass-flow outlet convergence history
of the - turbulent case with H-grid . . . . . . . . . . . . . . . . . . . 156
FIGURE 6.17 Single-stage axial-flow turbine velocity vectors distribution along the
stator blade row - turbulent case with H-grid . . . . . . . . . . . . . 157
FIGURE 6.18 Closeup of the reverse flow at the rotor suction side - turbulent case
with H-grid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
FIGURE 6.19 Scheme of the blocks created to generate the O-grid . . . . . . . . . 158
FIGURE 6.20 Detail of the O-grid on the turbine casing . . . . . . . . . . . . . . . 158
FIGURE 6.21 Detail of the gap between the turbine rotor and casing . . . . . . . . 159
FIGURE 6.22 Mesh elements distribution on the tip clearance region . . . . . . . . 159
FIGURE 6.23 Detail of the O-grid around the rotor blade tip and the refinement
of the clearance region . . . . . . . . . . . . . . . . . . . . . . . . . 160
FIGURE 6.24 Single-stage axial-flow turbine mass-flow outlet convergency history
- turbulent case with O-grid . . . . . . . . . . . . . . . . . . . . . . 160
FIGURE 6.25 Mixing-plane outlet to mixing-plane inlet ratio - turbulent case with
O-grid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
FIGURE 6.26 Single-stage axial-flow turbine pressure ratio monitoring - turbulent
case with O-grid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
FIGURE 6.27 Distribution of the velocity vectors in the turbine rotor - turbulent
case with O-grid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
LIST OF FIGURES
xix
FIGURE 6.28 Effect of Coriolis force in the velocity field close to the turbine rotor
wall - turbulent case with O-grid . . . . . . . . . . . . . . . . . . . . 162
FIGURE 6.29 Global view of the gas expansion along the turbine stage - turbulent
case with O-grid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
FIGURE 6.30 Detail of the static pressure contours along the turbine stage - turbulent case with O-grid . . . . . . . . . . . . . . . . . . . . . . . . . 163
FIGURE 6.31 Detail of the static pressure contours along the turbine stage - turbulent case with O-grid . . . . . . . . . . . . . . . . . . . . . . . . . 164
FIGURE 6.32 Detail of the leakage flow from pressure surface to suction surface of
the turbine rotor - turbulent case with O-grid
. . . . . . . . . . . . 165
LIST OF FIGURES
xx
LIST OF FIGURES
xxi
List of Tables
TABLE 6.1 Distribution of rotor blade inlet angle along the rotor height . . . . 193
TABLE 6.2 Distribution of rotor blade outlet angle along the rotor height . . . . 193
TABLE 6.3 Distribution of stator blade inlet angle along the stator height . . . 193
TABLE 6.4 Distribution of stator blade outlet angle along the stator height . . . 193
TABLE 6.5 Distribution of the blade space-chord ratio along the blade height . 194
TABLE 6.6 Number of blades for each row . . . . . . . . . . . . . . . . . . . . . 194
TABLE 6.7 Rotor loading factor distribution . . . . . . . . . . . . . . . . . . . . 194
TABLE 6.8 Distribution of rotor incidence flow angle along the rotor height
TABLE 6.9 Distribution of stator incidence flow angle along the stator height
. . 194
. 195
TABLE 6.10 Distribution of rotor deviation flow angle along the rotor height . . 195
TABLE 6.11 Distribution of stator deviation flow angle along the stator height . 195
TABLE 6.12 Rotor de Haller number distribution . . . . . . . . . . . . . . . . . . 195
TABLE 6.13 Stator de Haller number distribution . . . . . . . . . . . . . . . . . 196
TABLE 6.14 Axial Mach number for each blade row . . . . . . . . . . . . . . . . 196
TABLE 6.15 Comparison between streamline curvature and CFD results . . . . . 205
Uni-dimensional
2-D
Bi-dimensional
3-D
Three-dimensional
AFCC
AIAA
BOV
Bleed-off valve
CONV
CFD
CFL
CourantFriedrichLewy number
DISS
DP
Design-point
FDS
FluxDifference Splitting
FVM
FVS
FluxVector Splitting
GTAnalysis
IRS
MP
Mixing-Plane
MPO
Mixing-Plane Outlet
Mixing-Plane Inlet
MUSCL
NGV
ODP
Off-design-point
RHS
Right-Hand Side
sm
Surge margin
SA
Spalart-Allmaras
SLC
VIGV
VISC
VSV
xxiv
List of Symbols
Latin Characters
a
Speed of sound
Ai
ARoe
Roe-matrix
Convection term
Blade chord
Cd
Drag coefficient
CP
Pressure coefficient
cp
cv
Diffusion term
D2
D(Qi )
Dest
Destruction term
DISS
LIST OF SYMBOLS
e
E, F, G
Flux vectors
F~
Force
ei
Internal energy
F~
Body forces
Incidence angle
i, j, k
(i0 )10
(iD i2D )
kbl
Blockage coefficient
ki
Correction factor
l0
Characteristic length
Mach number
Mn0
Mni
Mncopt
ml
Rotational speed
xxvi
LIST OF SYMBOLS
P
Static pressure
Pe
Pt
Pr
P rt
P rod
Production term
qj
xxvii
Radius of curvature
~r
Re0
Reynolds number
Re0
Blade spacing
Tt
t/c
Peripherical velocity
Friction velocity
LIST OF SYMBOLS
Vm
Meridional velocity
x, y, z
coordinate system
z0
y+
yp
Distance between the first node on the wall and the wall surface
Relative velocity
xxviii
Greek Characters
Flow angle
Blade angle
Time-step
(d/di)2D
(0 )10
Freestream condition
LIST OF SYMBOLS
4
Limiter
Blade sweep
~n
Normal vector
Gradient operator
Divergent operator
Laplacian operator
Source term
Angular velocity
Density
Blade Solidity
ij
xxix
LIST OF SYMBOLS
xxx
Subscripts
0
Reference value
Inlet
Outlet
Blockage
Centrifugal
CO
Coriolis
Fluctuation term
ghost
Ghost element
Hub location
in
i, j, k
in
internal
face
Meridional
min
Minimumm
neig
Neighbor
LIST OF SYMBOLS
xxxi
out
Outlet
x, y, z
Stall
Radii
Fluctuation term
Wall
Superscripts
1
Inlet
Outlet
Positive eigenvalues
Negative eigenvalues
Nominal
Average
Mass-average mean
Dimensionless
1 Introduction
1.1
Objectives
The goal of this work is to develop a tool to calculate the flowfield within axial-flow
turbomachines. A long way is necessary to reach an accurate and reliable result on
turbomachinery numerical simulations, hence many other objectives should be reached
before understanding the complex process that involves compressor and turbine design
and their performance analysis. With the flowfield calculated along the turbomachinery
streamwise, blade-to-blade and spanwise, it is possible to analyze the physical aspects
of the flow at near wall regions, the boundary-layer behavior and the regions with high
flow losses. With these results, the compressor designer can enhance the compressor blade
geometry and rows matching aiming efficiency improvement. In this work, Computational
Fluid Dynamics (CFD) techniques are applied to calculate pressure, temperature and
velocity distributions of axial-flow compressor and turbine channels. A computational
code written in FORTRAN, was developed to solve the three-dimensional Navier-Stokes
equations in steady-state condition.
CHAPTER 1. INTRODUCTION
1.2
33
Motivation
Gas turbines are used in various applications as: aerospace, marine and land vehicles.
Industrial gas turbines to energy generation are also vastly applied. The use of turbomachines is enormous. Any small gain in turbomachinery efficiency and performance
translates into a major economic worldwide.
The three most important components of gas turbine engine are: compressor, combustion chamber and turbine. The designer should design and test each one of these
components using experimental and numerical tools. Experimental processes generally
are very expensive. Computational simulations are a promising means for alleviating the
cost of the time-consuming and expensive experimental process. If the designer starts
the experimentation upon an accurate preliminary design, the cost on test facilities and
the components manufacturing can be significantly reduced. The development of numerical tools increased dramatically as computers became more powerful in the last decades.
The Gas Turbine Group at ITA has developed several computational codes. A computational program, named Axial Flow Compressor Code (AFCC), developed by Tomita [1]
calculates the preliminary design of axial-flow compressors and its performance, becoming possible to obtain the compressor map. Techniques to improve off-design operation,
such as Variable Inlet Guide Vanes (VIGV), Variable Stator Vanes (VSV) and Bleed-ofValve (BOV) are implemented in AFCC. These results can be used as input data for a
streamline curvature program developed by Barbosa [2]. By using a streamline curvature
program it is possible to calculate all compressor geometry dimensions (annulus, blades,
diameters, length) using streamlines along the blade spanwise. A study of water injection
on the compressor inlet and of the compressor post-stall behavior has been developed. A
computational code developed by Jesus [3] can be used to design and performance anal-
CHAPTER 1. INTRODUCTION
34
ysis of axial turbines. The Gas Turbine Group at ITA also has an engine deck, named
GTAnalysis, developed by Bringhenti [4]. This engine deck is capable of calculating the
engine performance of several gas turbine configurations such as turbo-shaft, turbofan,
turbojet, gas generator with free turbine, twin or triple spool installations for both aeronautical and industrial interests considering steady-state operation. Silva [5] implemented
the transient module in the GTAnalysis code. A robust object-oriented modelling of the
engine deck framework is presented in references [6] and [7].
The goal of this work is to study axial-flow turbomachines flowfield using CFD techniques. Two options are possible: either to use a CFD commercial package or to develop
an in-house computational code. Considering that the historical of the Gas Turbine Group
at ITA is to develop their own computational tools for academic research, the last option
was chosen. With an in-house CFD code other researchers would be able to implement
new techniques to improve the design process, to reduce machine time and to allow optimizations.
1.3
1.3.1
Previous Developments
CFD in Academic Research and Industry
Currently, there are many CFD packages available to calculate the fluid motion for
both internal and external flows, with several configurations and installations. In particular, for industry this is a very good option to obtain a fast problem solution with low cost.
Over the last decades, major progress has been made in areas such as grid generation,
turbulence modelling, boundary conditions, pre- and post-processing computer architec-
CHAPTER 1. INTRODUCTION
35
ture, among others. The boundary conditions encountered in turbomachinery are among
the most complex in CFD [8].
CFD provides a complementary tool for simulation, design optimization and, moreover,
analysis of complex three-dimensional flows experimentaly inaccessible. For academy and
industrial purposes it is a powerful tool to study a specific flow behavior as multi-phase,
ablation of re-entry vehicles, jets, ice formation at leading edge of wings, component
cooling, porous-media, gas and oil extraction and process, thermal comfort, high-lift configurations, chemical reaction, weather forecast, among others. Most improvements of
CFD techniques and codes are due to academic efforts or partnership between academia
and industry.
CFD has been vastly applied to the automotive industry, mainly to racing cars, to
increase the thermal efficiency of engines and to improve the car aerodynamics [9].
In aeronautical and aerospace areas, CFD has been used more and more, for example
to simulate aerodynamic improvements. With the advance of computer hardware and
software, numerical simulations became widely used as well as experimental tests. The
last generation of CFD codes applied in aeronautical engineering are powered by an optimization technique [10] to improve the geometrical design of airfoils, fuselage and nacelles,
decreasing drag and increasing aerodynamic efficiency. A course of PROPESA [11] expressed this development as: An electronic version of a wind tunnel is emerging as a
useful tool for aerodynamistis. Called Computational Fluid Dynamics it uses high-speed
computers to generate mathematically the flow of fluid over a computer-designed model.
The computer can analyse the aerodynamic forces on the aircrafts surface, quickly sort
through a large number of possible design modifications and present the best solution.
Energy conservation is one of the main goals of wind-tunnel testing and CFD research.
CHAPTER 1. INTRODUCTION
36
Commercial airlines in the United States consume about 40 billion liters of fuel every year;
even a one per-cent improvement in fuel efficiency would save millions of liters of fuel.
One example of the use of CFD as a design tool is the design of the A330/A340 Airbus
[11]:
Stringent Design Target: 10% reduction in cruise drag compared with the A310 and
33% reduction in cruise drag compared with the MD11.
CFD Design: 800 different wing geometries analysed. Time: 2 years; Cost: 0.5
million.
Wind Tunnel Design: 800 different wing geometries tested. Time: 150 years; Cost:
65 million.
1.3.2
CHAPTER 1. INTRODUCTION
37
CHAPTER 1. INTRODUCTION
1.3.3
38
Modern multistage turbomachines are already the result of a highly complex design
process based on the designer expertise, supported by numerous computational resources
(hardware and software) and still finally developed and improved by various tests. All
turbomachinery components are affected by aerodynamic requirements, aerodynamic and
structural loads (pressure and temperature), heat transfer mechanisms and material properties. The structural loads should be satisfied by the life targets. Modern techniques to
design turbomachinery need a detailed work division as show in Figure 1.2. The turbomachinery field has multidisciplinary nature [20] and a multidisciplinary engineering team
becomes indispensable. CFD has fostered an unified approach to turbomachinery analysis
and is used as design tool.
Fluid Mechanics:
Experimental
Inviscid Flow
Control Systems
Viscous Flow
Turbulent Flow
Gas Dynamics
Numerical Methods
CFD
Turbomachinery:
Research
Applied and Pure
Mathematics
Development
Aero -Thermodinamic
and Heat Transfer
Design
Manufacturing
Maintence
Manufacturing
Engineering
Acoustic
Material Science
CHAPTER 1. INTRODUCTION
39
Gas turbine design components are based on design by analysis, whereby the blading
is assessed using CFD codes. During the design stage the designer repeatedly adjusts
the blade geometry and channel until finding a suitable geometry that combines acceptable aerodynamic performance with low stress levels, and it is practical and economic
to manufacture. This process of continual refinement of the match geometry can be tedious, expensive and time consuming, since at each stage the geometrical data for the
necessary aerodynamic or stress analysis must be determined. Optimization techniques
such as genetic algorithms, neural network and inverse design can be used to enhance the
preliminary design. Efficiency is probably the most important performance parameter for
most turbomachines, specially for gas turbines whether used for aircraft propulsion or
for land-based power plants, because their net power output is the reference between the
turbine work and the compressor work.
Advances are still possible, not only in the efficiency itself but also in the amount, and
hence cost. Factors influencing efficiency are extremely complex mainly for high pressure
compressor due to the difficulties during the rotor and stator rows matching. The advent
of modern numerical methods applied both whole engines and to individual components.
The flow through turbomachines is one of the most complicated in the field of fluid
dynamics practice. A three-dimensional flowfield calculation inside turbomachinery is
a long and complex task. Firstly, because it is necessary to previously know some important points of the compressor operation to setup correct parameters in the boundary
conditions. To obtain these parameters it is necessary to know the machine designed
and its operational characteristics. Reminding that, nowadays, with CFD techniques it
is impossible to design an axial compressor, however it is possible to improve the design.
But, in many instances, CFD is the only tool available that provides detailed flowfield
CHAPTER 1. INTRODUCTION
40
CHAPTER 1. INTRODUCTION
41
the preliminary design should be very robust [25, 26], accurate and reliable, because
if the pressure ratio calculated during the compressor design is not close to the actual
pressure ratio (obtained in the test facility), the boundary conditions will be unrealistic.
An incorrect pressure ratio can take the compressor to operate in an instability range [27].
The flow behavior when the compressor operate in an instability point is completely
unsteady. Dawes [28] presents a work based on unsteady flows associated with the interblade row interactions in compressor stages. The centrifugal and axial compressors were
studied. Hence, all numerical treatment given between the rows interface must be different compared to the numerical implementation for steady-state condition, including the
numerical treatment at inflow and outflow boundaries due to the high flow perturbations.
Compressor instable operation push the machine to work in surge or choke points.
Each component, such as compressor or turbine, have its particularities. Denton [29],
describe the difficulties to simulate high pressure compressor due to the numerical surge
caused by the complexity in start the CFD calculations for compressors because the
blades can be operating near stall point. Hence, the calculation may fail as a result of
the transient induced by the initial guess rather than because of a genuinely unstable
operating point.
In the work developed by Dawes [30], is possible to observe the difficulties in obtain
boundary condition at compressor outlet due to the complex physical aspects. The flowfield inside a single-stage axial compressor was calculated with and without the stator.
The variation of static pressure along the outlet rotor spanwise is clearly and sometimes
the imposition of a constant static pressure or simple radial equilibrium not represent
correctly the actual values due to the tip-clearance and static pressure drop at casing.
For unsteady flows in turbomachinery applications Chen, Celestina and Adamczyk
CHAPTER 1. INTRODUCTION
42
[31] presents two unsteady wake-blade row interaction models and a rotor- stator unsteady interaction models. These wake-blade row models had produced results that were
qualitative agreement with the results from the rotor-stator interaction model and quantitative agreement was reasonable and could be improved upon by further developing
improvements in the inlet boundary conditions employed in the time-shift model.
Several researchers on turbomachine area participated in the AGARD Propulsion and
Energetics Panel set up Working Group 26 as reported in [32] to help to clarify the issues
involving CFD applied to axial compressors using as wide a range of codes as possible of
two representative single blade row cases: NASA Rotor 37 and an annular turbine cascade
tested by DLR. This report presented a large amount of information about the difficulties
mainly to predict the pressure loss (was up to 40% in error). Some codes shown in [32]
were unable to predict correctly the highly three-dimensional secondary flow.
CFD codes are developed specially to solve specific problems [33, 34]. For example,
internal or external flows, incompressible or compressible, low or high Reynolds numbers,
with or without chemical reactions as dissociations that occur in the hypersonic flows,
porous-media, unsteady or steady-state, among others. There are specific theoretical
and numerical formulations for each case. For complex geometries and flows the mesh
generation and numerical implementation become more difficult. Some authors [35, 36, 37]
studied specific methodologies to calculate regions as shock waves with high accuracy.
Segunpta [38] develop a new compact finite-volume scheme based on a flux-vector splitting
discretization. This scheme has been analyzed by matrix-spectral analysis developed by
author.
Yee [39] present a mathematical formulation for high-order schemes using explicit
and implicit multidimensional compact high-resolution shock-capturing methods for Euler
CHAPTER 1. INTRODUCTION
43
CHAPTER 1. INTRODUCTION
44
1.4
Author Contribution
1. The thesis must contribute with the improvement of the axial design procedures in
use in the Gas Turbine Group at ITA;
2. The thesis might produce data for the establishment of loss models to be incorporated to the streamline curvature code, allowing additional blade profiles;
3. Develop the CFD code in a modular basis to serve as a platform for future studies.
In the Master of Science program, the author worked with compressor design and offdesign performance calculations. Hence, the meanline and streamline curvature methods
were vastly studied and applied. Some results of compressor design and performance
analysis calculated using AFCC program can be found in references [1, 59, 60]. The
use of the AFCC (Axial Flow Compressor Code) and GTAnalysis (Gas Turbine Analysis)
CHAPTER 1. INTRODUCTION
45
programs to predict the gas turbine engine performance using VIGV (Variable Inlet Guide
Vanes) and BOV (Bleed-of-Valve) can be found in [61, 62]. The AFCC program was
used in the Doctoral Program in the Faculdade de Engenharia de ITAJUBA - EFEI to
develop a module to optimize the design of axial compressors using Sequential Quadratic
Programming (SQP). The results are presented in reference [63] and the engine cycle
improvement using this optimization technique in the axial compressor when it is installed
in a gas turbine engine is presented in [64]. The most effective design charts for use in the
preliminary design process are those that allow the engineer to rapidly identify designs
that are unacceptable or provide guidance on the route towards the design optimization.
The purpose here is to develop a tool to calculate three-dimensional flowfield within
axial turbomachines. The developing process of this kind of computational code is very
cumbersome. Another option would be to use a commercial package. The question is:
why not to use a commercial package?
Design tools that combine CFD and optimization techniques, for example, drag minimization, fuselage optimization and inverse problem applied to the blade design, need
some special attention during the numerical implementation process and sometimes only
with the CFD source code it is possible to implement these techniques. In most cases,
turbomachinery designers must understand specific behaviors of the machine to design
improvements in order to decrease loss sources.
Commercial packages are widely used in industry, already validated and verified they
are user friendly due to the Graphic User Interface (GUI). Several problems in the applied
engineer can be solved using these packages. However, in academia, for an autonomous
research laboratory that works with the state-of-art on gas turbine engine design, it is
of great importance to own source codes in order to form human resources capable of
CHAPTER 1. INTRODUCTION
46
CHAPTER 1. INTRODUCTION
1.5
47
Work Organization
CHAPTER 1. INTRODUCTION
48
the solver development and some comments and also recommendations of the author are
presented.
Chapter five describes the code verification and validation. A step-by-step procedure
is described to verify the convective and diffusive momentum equations terms. The Euler
equations were calculated for a supersonic nozzle, using two spatial integration schemes
(centered and upwind) and different pressure ratio values. A NACA0012 airfoil with zero
angle-of-attack was also simulated, considering the transonic regime with Mach number
equal to 0.8. The centered and upwind spatial integration schemes were vastly tested to
analyze the code numerical stability. The implementation of diffusive terms were tested
with laminar flow on a flat-plate for different artificial dissipation models for centered
spatial integration scheme. The upwind scheme was also calculated. The flat-plate numerical results were compared with the analytical solution of Blasius. The nozzle used
in the Euler equations validation was also simulated using turbulent flow. The results of
turbulent flow simulation inside the nozzle were improved on the region after the nozzle
throat when compared to the Euler equations results.
Chapter six describes the results for turbomachinery flow simulations. The mesh
generation process and the mesh types as H-grid, H-O-H-grid and O-grid are presented.
Cases involving turbines and compressors were simulated. A esingle-stage axial-flow turbine, a single compressor rotor with low aspect-ratio and a full multistage axial-flow
compressor design procedure and CFD calculations were carried out.
The study case involving a single-stage axial-flow turbine presents the interaction of
turbine design methodology and CFD techniques to improve the flow through the turbine
rotor. Euler and turbulent flow were calculated using different mesh types (H-grid and
O-grid). The flow on the stator outlet and rotor inlet is evaluated with the use of mixing-
CHAPTER 1. INTRODUCTION
49
plane. The rotor blade geometrical changes are described to solve the problem of reverse
flow on the rotor suction side. After the geometrical modifications an O-grid was generated
with tip-clearance between rotor tip surface and turbine casing to verify the tip leakage
phenomena. Different types of numerical initialization to start the CFD calculations are
discussed.
Based on a 1MW gas turbine engine, a multistage axial-flow compressor was designed
and analyzed in four steps:
CHAPTER 1. INTRODUCTION
50
process is started and the computational domain is presented considering the use of
periodicity condition.
CFD calculation: Based on the streamline curvature results the boundary condition
and numerical initialization process were carried out to start the CFD calculations.
General aspects between design methodology and CFD results are discussed for all
study cases involving turbomachines.
Chapter seven describes important comments and conclusions about the cases simulated, their difficulties and numerical aspects. A brief discussion is done about the
importance of the CFD solver as research and teaching tool.
Chapter eight describes the future implementations to improve the code robustness
from pre-processing data until post-processing results. The techniques to improve the
CFD solver robustness based on numerical aspects including the mesh generation process
are described based on turbomachinery flow simulations.
2 Mathematical Formulation
2.1
The fluid
Aiming at the flow calculation in gas turbines parts like compressors and turbines,
the fluid is usually the atmospheric air or the products of combustion of oil-derived fuels
in air. It is observed that the amount of fuel burned in gas turbines is around 2% of
the air mass flow entering the combustion chamber, so that the fluid properties in a gas
turbine can be considered similar to those of the air. At the usual ranges of pressures and
temperatures encountered in todays gas turbines, air behaves as a newtonian perfect gas
with temperature-dependent specific heats and viscosity.
Bearing in mind that the total specific energy (by volume) is given by
2
1
e = ei + |W | ,
2
(2.1)
ei =
e 1
Wi Wi .
2
(2.2)
52
(2.3)
p
.
RT
(2.4)
An iterative process is need to get the approximate value of . From the calculated T ,
the fluid properties can be calculated [65]:
cv = cv (T ) = R cp(T ).
(2.5)
(2.6)
T2
=b
,
(s + T )
(2.7)
H = ei +
2
2
p 1
1
+ |W | = h + |W | .
2
2
(2.8)
The heat conduction is calculated by Fourier law and is represented by the heat flux
q:
q = T =
,
P r xi
(2.9)
53
where P r is the Prandtl number and is the thermal conductivity coefficient given by:
2.2
(2.10)
The flow
High performance gas turbines require very fast fluid flows inside the compressors and
turbines blade passages. Blade passage are channels that deflect the flow usually a fixed
followed by a rotating (in turbines) or a rotating followed by a fixed (in compressors).
This construction characteristic causes the flow, entering the fixed cascade that follows
a rotating one, to be locally transient, even so it may be considered steady globally, as
when the turbomachine is run at constant speed. Gaps existing at the top of the rotating
passages allow fluid flow from one passage to the other. Experience has shown that the
flow is highly 3 D, viscous and turbulent, so that, in this work, it is modelled using
the full equations os conservation of mass, momentum and energy, written in a rotating
frame of reference. Hence, the relation of absolute and relative velocities is represented
by Figure 2.1.
54
~ =
is the vector of tangential velocity, given by U
~ ~r. Hence, from 2.1 one can obtain
~ +
V~ = W
~ ~r,
with
(2.11)
~ ~r = x
.
(2.12)
Mass
(Wj ) = 0,
+
t xj
(2.13)
p
ij
(Wi ) +
(Wi Wj ) +
+ = 0,
t
xi
xi
xj
(2.14)
e
+
[(e + p)Wj ij Wi + qj ] = 0.
t xj
(2.15)
Momentum
Energy
The viscous stresses are considered linear and proportional to the rate of strain, and are
calculated using the newtonian fluid theory, hence:
ij =
Wi Wj
+
xj
xi
+ i,j
Wk
xk
,
(2.16)
where is the molecular dynamic viscosity and is the second coefficient of viscosity. In
this work the Stokes hypothesis is used, hence = 2/3l . The variable in equation
55
2.14 represents some terms related to the rotation of the fluid flow in rotating channels.
Details of the source term are presented in the Annex A.
The complete form of the momentum equations are, therefore:
(Wz ) +
(Wx Wz zx ) +
(Wz Wy yz ) + (Wz Wz + p zz ) = 0, (2.19)
t
x
y
z
where is the angular velocity.
Analyzing the magnitudes of the variables involved in the above equations it is seen
that they differ by several order of magnitudes. Bearing in mind their numerical solution,
it is possible to avoid the numerical errors related to those differences in numerical values
by normalizing the variables. This is done through the non-dimensionalization equations
,
0
x
x = ,
l0
y
y = ,
l0
z
z = ,
l0
i = Wi ,
W
a0
T
T = ,
T0
e
e = 2 ,
a0
= ,
0
,
0
a0
t = t ,
l0
= ,
0
P
P =
,
P0
and the Reynolds number is calculated by Re0 = 0 a0 l0 /0 for internal flow computations
~ |l0 /0 for external flow computations with the Mach number M =
and Re0 = 0 |W
~ |/a0 , where the subscripts 0 and means referential properties from the stagnation
|W
conditions and freestream conditions, respectively. These new defined parameters are
56
called
= dimensionless density,
x = dimensionless coordinate of a specific point, the same rule is valid for y and z
coordinates,
i = dimensionless velocity,
W
T = dimensionless temperature,
e = dimensionless total energy,
= dimensionless molecular dynamic viscosity,
= dimensionless thermal conductivity,
t = dimensionless time,
= dimensionless second viscosity coefficient,
P = dimensionless pressure,
Re0 = Reynolds number calculated using dimensionless variables.
After the transformation of variables indicated above and collection of terms, a similar
set of equations of conservation is obtained. Details are not supplied here.
2.2.1
Turbulence
Turbulence as understood today states that the flow properties may be represented as
the sum of mean property and a fluctuation of the property:
A = A + A0 ,
(2.20)
57
A = A + A
(2.21)
where the overbar and tilde stands for the mean and the prime and double prime stands
for the fluctuation components of the property A.
There are different ways to extract the average values. In this work, the Reynolds
Averaged Navier-Stokes (RANS) and Favre average methods are used, with the definitions,
respectively,
1
AReynolds = lim
T T
eF avre = 1 lim 1
A
T T
t+T
AdT ,
(2.22)
AdT ,
(2.23)
t
t+T
fi + Wi ,
Wi = W
+ H,
H=H
e = e + e,
T = T + T ,
(2.24)
qi = qi + qi0 ,
= + 0 ,
p = P + p0 ,
58
fj ) = 0
+
(W
t xj
(2.25)
f
f f P
(Wi ) +
Wi Wj +
(ij Wi Wj ) + = 0
t
xi
xi xj
(2.26)
e
Wi Wj
f
f
+
(e + P )Wj Wi (ij Wi Wj ) Wi (ij
) + qj + Wj h = 0
t xj
2
(2.27)
Equations 2.25, 2.26 and 2.27 are similar to equations 2.13, 2.14 and 2.15 except for the
following terms associated with the turbulence:
1
Wi Wi = usually names the kinetic energy of the (turbulent) fluctuations,
2
E F G
Q
= 0,
+
+
+
+
x
y
z
t
(2.28)
where Q is called the vector of conserved variables, E, F and G the flux vectors and
the source term associated to the rotation of the frame of reference.
In equation 2.28 the over hat stands for non-dimensional variables and the overbar for
the averages (RANS or Favre) as indicated by equations 2.22 and 2.23.
The flux vectors are made of non-viscous, viscous and turbulent parts, as indi-
59
E = Ee + Ev + Et ,
(2.29)
F = Fe + Fv + Ft ,
(2.30)
G = Ge + Gv + Gt ,
(2.31)
The convective components are represented by the subscript e. The viscous components
contain only the terms with the viscosity and are represented by the subscript v. The
turbulent components only the terms containing the fluctuations and are represented by
subscript t. The matricial representations of the terms are, therefore
f
=
y
Q
W
(2.32)
f
f
xW
x + P
f
f
xW
x
Ee =
W
f
f
xW
x
x
(e + P )W
(2.33)
Fe =
Ge =
f
y
W
f
f
yW
x
W
f
f
yW
y + P
W
f
f
zW
y
y
(e + P )W
f
z
W
f
f
zW
x
W
f
f
zW
y
W
f
f
z + P
zW
z
(e + P )W
60
(2.34)
(2.35)
x W
x xx
x W
y yx
Et =
W
x W
z zx
x x
(e + P )W
f
y W
x yx
y W
y yy
Ft =
W
z W
y zy
y y
(e + P )W
f
(2.36)
(2.37)
z W
x zx
z W
y yz
Gt =
W
z W
z zz
z z
(e + P )W
f
61
0
xx
1
Ev =
yx
Re0
zx
ei
l
t
f
f
f
x + yx W
y + zx W
z +
xx W
+
P rl P rt x
(2.38)
0
yx
1
Fv =
yy
Re0
zy
ei
l
t
f
f
f
y + yx W
x + yz W
z +
yy W
+
P rl P rt y
zx
1
Gv =
zy
Re0
zz
t
ei
f
f
f
zz Wz + zx Wx + zy Wy +
+
P rl P rt z
(2.39)
(2.40)
(2.41)
2 x + 2Wy
=
2 y 2Wx
62
(2.42)
The variables that have already appeared have the same meaning (but non-dimensionalized).
During the non-dimensionalization and averaging of the variables other definition came
out:
= l + t ,
(2.43)
f
f
f
x +
y xy W
z +
x +
y +
z +
x +
y +
z ,
xf = xx W
xy W
xz W
xx W
xx W
xy W
xz W
xz W
(2.44)
f
f
f
y + yx W
x + yz W
z + yy W
y + yy W
y + yx W
x yx W
x + yz W
z + yz W
z ,
yf = yy W
(2.45)
f
f
f
z + zx W
x + zy W
y + zz W
z + zz W
z + zx W
x zx W
x + zy W
y + zy W
y .
zf = zz W
(2.46)
2.2.2
Turbulence Model
There is not enough affordable computational power today for the flow calculation
using the above equations, since to capture the fluctuations of the flow properties an
63
extremely refined mesh is required. Therefore, to avoid this short coming, it is practice
today to model the turbulent parts of the equations. Details of the physics and of the
available turbulence models, despite not being the scope of this work, can be formal in,
for example, references [66, 67, 68, 69, 70, 8, 71, 72].
Based on the results of published literature related to the calculation of flows in turbomachines [25, 32], the one-equation turbulence model developed by Spalart-Allmaras
[73] was chosen for this work.
2.2.2.1
Reference [73] contains details of the physics behind the model, and of the model. It
is an equation developed for the eddy viscosity , valid for high Reynolds number:
i 2
h
1
c
D
b1
2
+ ft1 4U 2 .
= cb1 [1 ft2 ]S +
(( + )
) + cb2 (
) cw1 fw 2 ft2
Dt
d
(2.47)
The turbulent kinematic viscosity T is given by:
T = fv1 ,
(2.48)
where
fv1 =
3
3 + c3v1
S = S + 2 2 fv2 ,
d
(2.49)
S=
p
2ij ij ,
(2.50)
64
where
1
ij =
2
Wi Wj
+
xj
xi
,
(2.51)
and
fv2
= 1
1 + fv1
r=
fw = g
SK 2 d2
1 + c6w3
g 6 + c6w3
16
wt2 2
2
2
(d + gt + dt )
ft1 = ct1 gt exp ct2
4U 2
(2.53)
4U
gt = min 0.1,
,
wt 4xt
(2.54)
where 4xt is the grid spacing along the wall at the nominated transition point, dt is the
distance between the local point and the transiton point and wt is the vorticity at the
wall at the nominated transition point. The other constants are given by:
2
3
cb2 = 0.622
cb1 (1 + cb1 )
+
cw2 = 0.3
ct3 = 1.2
cb1 = 0.1355
cw1 =
cv1 = 7.1
ct1 = 1
= 0.41,
cw3 = 2,
ct4 = 0.5.
65
i 1
D
1h
((
+ )) + cb2 ()2
+
=cb1 [1 ft2 ]S +
Re0
Dt
" #2
h
cb1 i
1
cw1 fw 2 ft2
+ ft1 U 2 Re0 ,
Re
0
d
(2.55)
where
0 /0
1
S = S +
fv2
2
2
Re0
d
r =
1
.
2 d2 Re0
SK
(2.56)
All other variables are dimensionless by definition. Applying the definition of material
derivative
( )
D( )
~ .( ),
=
+W
Dt
t
(2.57)
and assuming that on the near wall region the velocity is small, the density variation can
be neglected. The non-conservative form of equation 2.47 becomes
i 1
1h
= cb1 [1 ft2 ]S +
((
+ )) + cb2 ()2
+
Re0
t
" #2
h
cb1 i
1
~ ) + ft1 4U 2 Re0 .
cw1 fw 2 ft2
.(W
Re
0
d
(2.58)
~ .() = .(W
~ ) .W
~ was used to obtain Equation 2.58.
The identity W
In the equations above, subscript t stand for transition points, so that the corresponding terms are set to zero outside the transition region.
At walls = 0. For freestream initial condition is usually taken in the interval
[0, /10].
The physical process of fluid motion is governed by the Navier-Stokes equations where
66
unsteady, convection and diffusion terms are considered. The turbulent fluid motion is
related to production, diffusion and dissipation of eddies terms including unsteadiness and
convection. Each flow field problem has different physical aspects. Basically, in the inner
region the small eddies are formed and predominant. Large eddies are formed in the outer
region. The smallest eddies grow and become large eddies and the inverse process occurs
too. This phenomenon is called production of turbulence. Due to the eddies motion in a
turbulent flow, there is a transport of fluid properties that increases the mixing process.
This process is called eddy diffusion. During this mixing process the smallest eddies
dominate the energy dissipation. These eddies are formed due to high velocity gradients,
so the viscous forces become considerable and the smallest eddies are destroyed.
The Spalart-Allmaras turbulence model presented in Equation 2.58 can be written in
the following form:
= C() + D() + P rod() + Dest() + T,
t
(2.59)
(2.60)
D() =
i
1h
((
+ )) + cb2 ()2 ,
(2.61)
(2.62)
67
" #
i 2
c
b1
Dest() = cw1 fw 2 ft2
,
(2.63)
T = ft1 4U 2 .
(2.64)
Details of the development of all terms and the calibration of the model, such as the
choice of constant values, are presented in [73].
3 Numerical Formulation
In this chapter the equations developed in the previous chapters are prepared for the
sake of numerical solution. Stability, consistency and convergence problems are dealt
with. In addition, methods for convergence acceleration, mesh treatments, initial and
boundary conditions for the rotor-stator interaction are discussed. Essential ingredients
for an accurate and efficient solution of the flowfield are:
3.1
69
Discretization Models
Finite volume has been reported as a method which gives good results for flow bound
by complex geometries using unstructured mesh. It is the method adopted in this work.
The reader may refer to [74, 55, 68, 75] if comparisons to other methods are of his interest.
Different space and time discretizations are used in this work, aiming at a platform
for the test, comparison and application of known and to be developed methods. For the
spatial discretization the methods of Jameson (centered), van Leer (Flux Vector-Sppliting)
and Roe (Flux-Difference Sppliting), are studied, while for the temporal discretization the
explicit MacCormack (1969) and the Runge-Kutta (five steps) are chosen. It is possible
to combine spatial and temporal options, resulting in several different methods.
Central schemes cannot recognize and suppress the odd-even decoupling of the solution. Hence, the so-called artificial dissipation has to be added for numerical stabilization.
It is based on a blend of 2nd and 4th differences scaled by the maximum eigenvalue of the
convective flux Jacobian. These combination of an undivided Laplacian and biharmonic
operator is employed for unstructured meshes. The details are described in section 3.4.1.
Upwind schemes are constructed by considering the physical properties of the convective terms (Euler equations). These treatment distinguish the upstream and downstream
influences based on wave propagation directions. Two upwing schemes are implemented
in this work: Flux-Vector Sppliting and Flux-Difference Sppliting.
The Spalart-Allmaras method is used for turbulent calculations. This method was
chosen based on the good results reported in the literature when applied to flows in
turbomachines, the application born in mind in this work. The diffusion and advection
terms are discretized according the suggestions of [73]; the time discretization is implicit
70
3.1.1
Finite-Volume Method
G
Q E F
+
+
+
= 0,
t
x
y
z
(3.1)
E = Ee Ev ,
(3.2)
F = Fe Fv ,
(3.3)
G = Ge Gv ,
(3.4)
with
(3.5)
where Ee , Fe and Ge are Euler (convection) terms and Ev , Fv and Gv are the viscous
terms.
Integrating Equation 3.1 on a finite-volume and applying the Gauss theorem to transform the volume integral into a surface integral:
Z
V
Q
dV =
t
~ P~ )dV =
(.
(P~ .~n)dS,
(3.6)
with
P~ = E~i + F ~j + G~k,
(3.7)
71
S is the control surface and ~n is the outward normal vector. Defining the value of Q by
= 1
Q
V
=
QV
QdV,
(3.8)
QdV,
(3.9)
it follows that
and
Z
V
Q
Q
dV = (QV
)=V
.
t
t
t
(3.10)
1
Q
=
t
V
(P~ .~n)dS.
(3.11)
(3.12)
where k indicates the face of i th volume, nf aces is the number of faces for each volume
and S~k is the outward oriented normal area vector of k face. All fluxes are calculated
based on the face k of an i th volume. In this work, the values of flow properties are
considered to be located at volume centroid.
3.2
Spatial Integration
The spatial integration of vectors E, F and G for each k face given by Equation 3.12
is divided in two parts: the term involving convective flux and the term involving the
72
diffusive or viscous flux. For each i th volume the convective and viscous terms, are
represented by CON V (Qi ) and V ISC(Qi ), respectively.
About the convective terms, to avoid numerical instabilities associated to the centered
method caused by high gradients, artificial dissipation or artificial viscosity is used. Artificial dissipation acts as a damping function to limit high gradients [76] and is added
to the model used in this work. Upwind schemes [77, 78, 43, 79] do not require artificial
dissipation because the flux is calculated following a specific direction, hence naturally
dissipative. This numerical treatment deals with problems in high gradient regions using the vector-splitting methodology [80, 81, 82, 83, 84, 67]. Both, centered and upwind
methods are used in this work and are presented in forthcoming sections.
3.2.1
The centered discretization [76] of flux balance for each i th control volume is based
on information from the current volume and its neighbors. The flux is calculated by
summation of fluxes at all faces of i th control volume. The vectors E, F and G are
divided in two terms, one for convective terms and other for viscous terms,
nf aces
CON V (Qi ) =
~k .
(Ee (Qk )~i + Fe (Qk )~j + Ge (Qk )~k).S
(3.13)
k=1
1
~k ,
CON V (Qk ) = (Qk + Qnei ).S
2
(3.14)
73
where nei stands for the neighbor volume. The vectors E, F and G are written as functions
of the vector of conserved variables Q.
3.2.2
The use of damping functions like artificial dissipation is avoided with the F V S for
the convective terms. This method is an upwind method that is naturally dissipative.
Steger and Warming [85] developed the method for the Euler equations splitting the
eigenvalues of the Jacobian matrix. A disadvantage of this scheme is that the split fluxes
are not continuously differentiable at the sonic region, thus a non-smooth solution at the
sonic region may appear within the domain. To overcome this problem van Leer [83]
proposed some modifications to the Steger and Warming scheme, details of which can be
found in [80]. Basically the method uses a separation of the vectors Ee , Fe and Ge in two
contributions each:
Ee = Ee+ + Ee ,
(3.15)
Fe = Fe+ + Fe ,
(3.16)
Ge = G+
e + Ge .
(3.17)
Contributions + and are associated with positive and negative eigenvalues, respectively.
For M 1 and M 1 the calculations are based on the flux for all directions. The
scheme will be presented only for vector Ee . For vectors Fe and Ge the development of
terms is similar.
74
For M 1:
Ee+ = f,
(3.18)
Ee = {0 0 0 0 0}T ,
(3.19)
Ee+ = {0 0 0 0 0}T ,
(3.20)
Ee = f,
(3.21)
and for M 1:
where,
f=
Wx
(Wx2 + a2 /)
Wx Wy
Wx Wz
(3.22)
For 1 < M < 1 the calculations are based on the forward flux f for all directions
Ee+ = f + ,
(3.23)
Ee = f ,
(3.24)
75
where,
f =
fmass
fmass
[( 1)Wx 2a]/
Wy
fmass
fmass
Wz
fmass
{[( 1)Wx + 2a]2 /[2( 2 1)] + 0.5(Wy2 + Wz2 )}
(3.25)
with
fmass
= a[0.5(M 1)]2 ,
(3.26)
(3.27)
(3.28)
CON V (Qi ) =
X
k=1
~k .
[(Ee )k~i + (Fe )k~j + (Ge )k~k].S
(3.29)
3.2.3
76
The F DS schemes evaluate the convective fluxes at a face of control volume from
left and right states based on idea of Godunov [86]. Several numerical schemes for the
solution of hyperbolic conservation laws are based on exploiting the information obtained
by considering a sequence of Riemann problems. The exact solution of the Riemann
problem, approximate Riemann solvers were developed by Osher et al. [77] and Roe [78].
The methodology of Roe is applied quite often because of its high accuracy for boundary
layers and good resolution of shocks regions.
In this work, Roes approximate Riemann solver was implemented in the framework
of the cell-centered scheme. It is based on the decomposition of the flux difference of
convective terms over a face of the control volume into an addition of wave contributions.
At a k-face of the left (L) and right (R) states of a control volume the difference is
expressed as
~R Q
~ L ),
(P~e )R (P~e )L = (ARoe )k (Q
(3.30)
where ARoe denotes the Roe-matrix. The Roe-matrix is identical to the convective flux
c , where the flow variables are replaced by Roe-averaged variables presented
Jacobian,
bellow
=
Wx =
Wy =
z =
W
=
H
a
=
p
L R ,
WxL L + WxR R
,
L + R
WyL L + WyR R
,
L + R
WzL L + WzR R
,
L + R
HL L + HR R
,
L + R
q
q2 ),
( 1)(H
77
(3.31)
= W
x nx + W
y ny + W
z nz ,
W
x2 + W
y2 + W
z2 .
q2 = W
c T1
With some algebra, inserting the diagonalisation of the Roe-matrix as ARoe = T
decomposing the Roes scheme given by equation 3.30 into waves where
c (C
~R C
~ L ),
(P~e )R (P~e )L = T
(3.32)
(3.33)
where
78
1
~ R ) + P~c (Q
~ L ) |ARoe |K (Q
~R Q
~ L )].
(P~e )k = [P~c (Q
2
(3.34)
The product of |ARoe | and the difference of the L and R states is given by
~R Q
~ L ) = |P~1 | + |P~2,3,4 | + |P~5 |,
|ARoe |(Q
(3.35)
where
p a
W
a
|P~1 | = |W
|
2
a
|
|P~2,3,4 | = |W
p
x
W
Wy
z
W
q2
2
x a
W
nx
y a
W
ny
,
Wz a
nz
H
W
(3.36)
x W nx
W
y W ny
W
,
Wz W nz
x Wx + W
y Wy + W
z Wz W
W
W
(3.37)
79
p + a
W
+a
|P~5 | = |W
|
2
a
x + a
W
nx
y + a
W
ny
.
Wz + a
nz
+a
H
W
(3.38)
|c | =
|c |,
80
if |c | > ;
2
2
|c | + , if |c | ,
2
(3.39)
where is a small value that represents some fraction of the local speed of sound. In
this work the range used is 0.05 0.15. The same entropy correction was also
.
implemented to W
3.2.4
Reconstruction Based on Approximate Monotone UpstreamCentered Schemes for Conservation Laws (M U SCL)
To reach the second-order accuracy, one possibility is to use the M U SCL approach
assuming that the solution to vary over the control volumes in a linear fashion. In order to
calculate the L and R states, a reconstruction of the assumed solution variation becomes
necessary. In this work a piecewise linear reconstruction was implemented. Hence, it is
assumed that the solution is linearly distributed over control volume (Figure 3.1). With
this idea, the L and R states are calculated for the cell-centered scheme with the following
relations
QL = QI + I (QI ~rL )
QR = QJ + J (QJ ~rR )
(3.40)
where the QI and QJ are the gradients at the cell-center I and J respectively, denotes
a limiter function, ~ri and ~rj are the distances from the cell-centroid to the face-midpoints
of the cell.
81
3.2.5
Venkatakrishnans Limiter
82
, if 2 > 0 ;
1,
if 2 = 0 ,
(3.41)
where
1,max = Qmax Qi ,
1,min = Qmin Qi ,
(3.42)
and
1
2 = (Qi ~ri ),
2
Qmax = max(Qi , maxj Qj ),
Qmin = min(Qi , minj Qj ),
(3.43)
where Qmax and Qmin stand for the minimum and maximum values on all neighbors of
control volume i including i itself; parameter 2 is used to control the amount of limiting.
83
Hence, setting 2 equal to zero results in full limiting, but this may stall the convergence.
Otherwise, if 2 is set to a large value, the limiter function will return a value around
unity. In practice, it was found that 2 should be proportional to a local length scale, i.e.,
2 = (Kh)3 ,
(3.44)
where K is a constant set by the user and it is (1) and h is the cube-root of the
volume of control volume. In reference [90] the solution dependence of several k values
are discussed.
3.2.6
The viscous terms are evaluated using cell-centered scheme due to the elliptic nature
of the diffusive fluxes, dynamic viscosity, heat conduction coefficient and stresses terms
resulting:
nf aces
V ISC(Qi ) =
~k .
(Ev (Qk )~i + Fv (Qk )~j + Gv (Qk )~k).S
(3.45)
k=1
The discretization is based on the arithmetical averages for a current ith control volume
and its neighbors nei at k th face. The fluxes for each face can be written as
1
~k ,
V ISC(Qk ) = (Qk + Qnei ).S
2
(3.46)
where nei stands for a determined neighbor volume. The vectors E, F and G are written
as function of the vector of conserved variables Q.
3.3
84
Time Integration
The used time integration scheme is explicit. The MacCormack (1969) and the RungeKutta (five steps) schemes were implemented in this work and are presented bellow.
3.3.1
The MacCormack scheme for time integration is based on a special version of the LaxWendroff method. Instability of centered schemes is avoided if predictor (forward) and
corrector (backward) steps like the used by Veuillot [34] for turbomachinery simulations.
For predictor step,
(n+1)
Qi
(n)
= Qi
ti
(n)
(n)
(n)
[CON V (Qi )f orward V ISC(Qi )f orward DISS(Qi )f orward ].
Vi
(3.47)
(n+1)
Qi
(n)
= Qi
ti
(n+1)
(n+1)
(n+1)
[CON V (Qi
)backward V ISC(Qi
)backward DISS(Qi
)backward ].
Vi
(3.48)
Qi
1 (n+1)
(n+1)
= [Qi
+ Qi
].
2
(3.49)
Constant Courant-Friedrichs-Lewy (CF L) number is used for the calculations of the local
time-step considered, from n to n + 1 time indexes:
ti = CF L
lengthi
.
~ | + ai )
(|W
(3.50)
85
The variable lenghti corresponds to the characteristic length of the volume, calculated
based on the distance between the ith volume centroid and the closest face. The variable
ai is the speed of sound for the i th control volume.
The MacCormack time integration method presents a good convergence for 0 <
CF L < 1.
3.3.2
86
(0)
= Qi ,
(1)
= Q0i 1
(3.52)
(2)
(3.53)
(3)
(4)
(5)
Qi
Qi
Qi
Qi
Qi
Qi
(n+1)
Qi
(n)
ti
(0)
(0)
(0)
[CON V (Qi ) V ISC(Qi ) DISS(Qi )],
Vi
ti
(1)
(1)
(1)
Q0i 2
[CON V (Qi ) V ISC(Qi ) DISS(Qi )],
Vi
ti
(2)
(2)
(2)
Q0i 3
[CON V (Qi ) V ISC(Qi ) DISS(Qi )],
Vi
ti
(3)
(3)
(3)
Q0i 4
[CON V (Qi ) V ISC(Qi ) DISS(Qi )],
Vi
ti
(4)
(4)
(4)
Q0i 5
[CON V (Qi ) V ISC(Qi ) DISS(Qi )],
Vi
(5)
= Qi ,
(3.51)
(3.54)
(3.55)
(3.56)
(3.57)
3.4
87
schemes in which the truncation errors act as a numerical dissipation, but very unstable
in regions with high gradients like in shock waves.
The artificial dissipation or artificial viscosity is responsible for maintaining a good
numerical stability. There are several mathematical models with different characteristics,
linear and non-linear. A flow properties sensor is used to activate or deactivate the
artificial dissipation. Some of such high-order schemes were implemented in this work
and they are presented in the next sub-section.
Jamesons scheme, using centered-difference and Runge-Kutta to perform the spatial
and time integrations, respectively, can be improved in terms of convergence acceleration.
For steady-state simulations it is possible to apply a local time-step based on a linear
stability analysis of the Navier-Stokes equations. This approach is not proper for transient
solutions because the time-marching process must be compatible with the physical aspects
of flow, what requires constant time-step for all control volumes, although the CF L
number may vary. Other alternative is the implementation of implicit residual smoothing
(IRS) to increase the maximum operational CF L number. This technique is presented in
Section 3.4.2. IRS is much used with multigrid methodology [50, 94, 29].
3.4.1
Artificial Dissipation
In this work, four artificial dissipation models were implemented. Tests were carried
out initially with a linear model, in order to check the quality of results. Then, non-linear
models were implemented.
88
Linear Model
The model used in this work is taken from Hoffmann [67]. Let the i th volume and
directions i, j and k be considered. For each direction, the i th control volume has a
neighbor at each face. The neighborhood are identified by i 2, i 1 at the i direction,
and similarly for the others. Therefore, for i, j and k directions the dissipation terms are
DISS(Qi )i = [(Qi )i2 4(Qi )i1 + 6(Qi ) 4(Qi )i+1 + (Qi )i+2 ],
(3.58)
DISS(Qi )j = [(Qi )j2 4(Qi )j1 + 6(Qi ) 4(Qi )j+1 + (Qi )j+2 ],
(3.59)
DISS(Qi )k = [(Qi )k2 4(Qi )k1 + 6(Qi ) 4(Qi )k+1 + (Qi )k+2 ],
(3.60)
(3.61)
3.4.1.2
Non-Linear Models
Non-linear artificial dissipation model used in this work is a blend of second and fourth
differences. The difference between the linear and the non-linear model is that, the last one
is dependent on the local pressure gradient. This requires in region with severe pressure
gradients, treatment and formulation different from other regions of potential flow. Two
procedures have been implemented. The first is based on the work of Jameson [76] for
two-dimensional case, here extended for three-dimensional. Considering the directions in
89
(3.62)
(3.63)
(3.64)
DISS(Qi )i+1 =
where, Vi is the volume of the i th control volume. The coefficients (2) and (4) depend
on the flow behavior, and are calculated using a pressure sensor. For the direction i
(2)
(2)
(3.67)
(3.68)
(3.69)
where
(i )i =
90
The constants k (2) and k (4) assume the values k (2) = 1/4 and k (4) = 1/256 as in [76].
The term (2) is second-order accurate and the term (4) first-order resulting in a thirdorder accurate scheme, as demonstrated by Jameson [76]. In regions with high gradients,
the term (2) reduces the accuracy to first-order and the term (4) , with fourth differences,
tend to induce overshoots during the numerical iterations, being turned-off. For other
directions there are identical procedures.
The second and less dissipative procedure is based on the work of Mavriplis [93] extended to three-dimensions. In this model the artificial dissipation terms are divided by
one undivided Laplacian operator that is first-order accurate and is used only in the regions with strong gradients to avoid numerical oscillations and one undivided biharmonic
operator that is third-order accurate:
(3.70)
D(Qi ) =
(Qm Qi ),
(3.71)
m=1
DISS(Qi )biharmonic =
(2)
X
m=1
[D(Qm ) D(Qi )]
Am Ai
2
,
(3.72)
with
nf aces
Ai =
~k + (Wy )k S
~k + (Wz )k S
~k | + ak Areak ],
[|(Wx )k S
(3.73)
k=1
where ak is the speed of sound, based on the mean value properties of ith control volume
91
and its neighbor at face k; Areak is the area at face k belonging to the i th control
volume and its neighbor.
The Laplacian operator is calculated replacing the quantities D(Qm ) and D(Qi ) in
Equation 3.72 by the quantities Qm and Qi and replacing the constant (2) by (1) :
nneigh
DISS(Qi )Laplacian =
(1)
[Qm Qi ]
m=1
Am Ai
2
.
(3.74)
(1)
(pk pi )|
k=1
nf aces
(3.75)
(pk pi )
k=1
Typical values are (2) = 1/2 and 1/256 (4) 3/256, as indicated by Mavriplis [93].
The values used in this work are: (2) = 1/2 and (4) = 3/256.
Other non-linear formulation can be obtained using the formulation of the first nonlinear model presented is this sub-section, but replacing the term Vi /t in Equation 3.66
by the term (Am Ai )/2 calculated in Equation 3.73. These two last models are less
dissipative than the first model presented and provide conservative dissipation using the
averages given by the terms Am and Ai . Details of artificial dissipation in the boundaries
are presented in Appendix A.
Pulliam studied various artificial dissipation models for Euler equations, for both,
linear and non-linear models using an implicit approximate factorization for transonic
airfoils. The treatment of the artificial dissipation models at boundaries and a fully
implicity application of these models are analysed and can be found in [95].
3.4.2
92
The IRS method [83, 96, 97, 98] is used to increase the maximum operational CF L
number for numerical schemes that use explicit time-integration.
In this work, the IRS model implemented is the one given by Jameson and Mavriplis
[92] and it is extended to three-dimensions. The IRS is applied to the right-hand-side
(RHS) of the governing equations. The mean (RHS) or simply RHS i is calculated by
(3.76)
where 2 is the undivided Laplacian operator. Thus the smoothed residuals are calculated
using:
nf aces
RHSi = (1 + nf aces)RHS i
RHS k ,
(3.77)
k=1
where is 1/8 1/2. Equation 3.77 can be solved numerically using several Jacobi
iterations and can be applied at alternate stages of Runge-Kutta time-integration method.
3.5
Z
Vi
1
i
=
t Re0
93
.[(
+ )]dV
Vi
~ )dV +
.(W
Vi
" #2
1
cb2
2
() + cb1 S cw1 fw
.
Re0
d Re0
(3.78)
i
1
=
t Vi Re0
Z
1
~
~ ).dS+
~
[(
+ )].dS
(W
V
i Si
Si
" #2
1
cb2
()2 + cb1 S cw1 fw
.
Re0
d Re0
(3.79)
C(
)
" #2
1
cb2
D2 ()
(3.80)
Dest(
)
The spatial discretization of the convection and diffusion terms follows Spalart-Allmaras
work [73]. The convection term is discretized using a first-order upwind scheme and the
diffusion term is dicretized using second-order accurate centered-difference with second
diffusion term D2 (), production P rod() and destruction Dest() terms, for the i th
94
("
i + (
)nei
2
+
i + ()nei
2
!#
+ ()nei
2
!)
~k +
.S
nf aces
1 X +
)(k )nei ].S
~k + D2 (i ) + P rod(i ) + Dest(i ),
[(Wk )i + (W
k
Vi k=1
(3.81)
with
i )k + |(W
i )|k
(W
,
2
nei )k |(W
nei )|k
(W
=
.
2
Wk+ =
Wk
(3.82)
(3.83)
n+1
= f + t
f
t
n+1
,
(3.84)
95
[(
+ )].Sk
[(Wk+ )in+1 + (Wk )nei
].Sk +
Vi Re0 k=1
Vi k=1
)
+D2 (in ) + P rod(in ) Dest(in ) .
(3.85)
n+1 =in + ti
Knowing that
in = in+1 in ,
(3.86)
in = ti
!n+1
,
(3.87)
it is possible to write
!n+1
nf aces
nf aces
X
X
1
n+1 ~
~k 1
[(
n + n )].S
[(Wk+ )in+1 + (Wk )nei
].Sk +
Vi Re0 k=1
Vi k=1
)
n
n
n
+D2 (i ) + P rod(i ) Dest(i ) .
(3.88)
(
=
Replacing the variable n+1 following Equation 3.86 and grouping terms, it is possible to
96
write
!n+1
(
=
nf aces
X
1
~k +
[(
n + n )].S
Vi Re0 k=1
nf aces
nf aces
X
1 X
1
+
n
n
n
~k
~k +
[(Wk )i + (Wk )nei ].S
[(Wk+ )in + (Wk )nei
].S
Vi k=1
Vi k=1
)
+D2 (in ) + P rod(in ) Dest(in ) .
(3.89)
The second term in the right-hand-side is evaluated implicitly and all other terms
representing the right-hand-side (involving in ) of the turbulent transport are evaluated
in time n.
Other important remark about how to calculate the convection term implicitly, concerns the mesh. For structured grid, the Thomas algorithm may be directly used, because
it is possible to know the location of the zeros out of the dominant diagonal (tridiagonal,
pentadiagonal or hexadiagonal). For unstructured mesh, the Thomas algorithm cannot be
used directly, because the locations of the non-zeroes are unknown, resulting in a sparse
matrix. In order to take advantage of the large number of zeroes in the matrix, it is
also required special storage algorithms. In this work, the conjugated gradient is used.
References [99] and [100] indicate several methods to solve sparse matrix. Reference [101]
gives robust method for the LU factorization of the original matrices.
3.6
Study of papers available in the literature [53] indicated that the hexahedral element
has better quality, concerning the solution in the areas of high gradients like the ones
in turbomachines. In this work only hexahedral control volumes were considered, since
97
turbomachinery is the application in mind. The meshes for the rotor and for the stator must be usually independent, due to the different numbers of blades in each blade
row. It is worth mentioning that the computational code developed in this work must
to be a platform aiming future development, among which adaptative mesh refinement,
tetrahedral, prismatic, pyramidal elements and hybrid mesh are in mind.
The elements and its neighbors are addressed via connectivity tables. For hexahedral
element with six faces there are six neighbors. Elements in the boundary, the ghost
elements, were created aiming at the boundary conditions imposition. Ghost elements
are similar to real element.
The treatment of the edges, faces, areas, outward normals and volumes follow [83].
3.7
The calculations start from the flow properties established as initial conditions set at
the centroid of each element, respecting the physics of the flow to avoid numerical instabilities and divergence. The closer the initial condition is to the solution the smaller the
number of iterations to reach convergence. For steady-state solution the initial condition
set of values may not represent the actual flow but, for transients, the initial conditions
must be the actual flow conditions at the time when the calculations start.
The elements in the boundary require boundary conditions that are set in the ghost
elements. These elements must provide correct informations about the flow in the real
neighbor elements (see Figure 3.7). All flow properties for both ghost and real elements
98
are cell-centred. Shapes of the ghosts elements are the same as that of the adjacent real
element.
FIGURE 3.2 Scheme implemented to set the boundary condition in a ghost element
Compressors or turbines stages are made of rotating and fixed elements. Therefore the
rotor meshes move with the blades, while the stators remain still. An interface between
two adjacent rows must be created in order to precisely transfer information from the
outcoming flow of one row to the incoming of the next row.
3.7.1
Initial Conditions
Initial conditions are the set up values of the variables used for starting the calculations. They must be qualitatively correct and represent an approximation of the physical
problem, to avoid divergence. They can be chosen from auxiliary meanline or throughflow
calculations. Convergence may be difficult to achieve if adequate initial conditions are not
set up appropriately.
Good initial conditions may also reduce the time machine mainly when full NavierStokes is solved and large number of elements are required.
Computational codes for compressors and turbines, developed at Center of Reference
for Gas Turbine at ITA produce very good initial conditions for the simulations carried
out during this research. For example, the streamline curvature computer code can give
excellent point of departure since it calculates accurate axisymmetric solutions, that may
be used for both design and off-design simulations.
3.7.2
Boundary Conditions
3.7.2.1
Inlet
99
The inlet conditions depend on the flow regime. If the inlet flow is supersonic all
properties must be taken from upstream because there is no communication of the flow
properties from downstream.
For subsonic inlet flow, two possibilities were considered:
1. The first one considers specified stagnation conditions as pressure and temperature
and extrapolates velocities and density from adjacent interior control volumes. The
angles of the meridional and tangential planes should be given by the user following
the turbomachinery design specifications.
2. The second one, for turbomachinery study which starts usually with flow directions
and Mach numbers, the flow velocity components are initially calculated by
1
1
+ [M12 (r)2 ],
( 1) 2
where M1 is the relative flow inlet Mach number; 1 is the relative flow inlet angle based
on meridional plane, 1 is the meridional streamline inclination angle relative to the axial
direction, r is radius of the centroid of the control volume with origin at the axis rotation;
100
x and y are the coordinates of the centroid of the control volume and is the angular
velocity. The inlet flow density is calculated from static pressure and static temperature.
At this point it is interesting to comment that boundary conditions at the inlet interface is commonly used with success. For high performance axial flow compressors, that
involves high velocities levels, it is important to build an extended computational domain
at the compressor inlet to avoid problems with numerical oscillations due to the flow
properties fluctuations. Suitable non-reflecting inlet and outlet boundary conditions were
vastly studied by Giles [102] and [103] to solve this problem. Conversely, non-reflecting
formulation may be used if the inlet is at a short distance from the blade. This affect
directly the mesh size.
Farfield and outlet non-reflecting boundary conditions are presented in sections 3.7.2.3
and 3.7.2.2, respectively. The methodology implemented in the code is based on the work
of Holmes [104] for subsonic inlet, which uses the outgoing Riemann invariant, defined as
~ in ~n 2ain ,
R = W
1
(3.90)
where the subscript in denotes the internal control volume and ~n is the outward unit
normal vector. The Riemann invariant is used to determine either the absolute velocity
or the speed of sound at the inlet boundary, ain calculated by
R( 1)
ain =
( 1) cos2 + 2
(
1 + cos
[( 1) cos2 + 2]a20
( 1)(R )2
)
,
(3.91)
where represents the flow angle relative to the boundary; a0 is the stagnation speed of
101
sound and
~ in ~n
W
,
~ in |
|W
(3.92)
( 1) ~ 2
|Win | ,
2
(3.93)
cos =
and
a20 = a2in +
~ in denoting the relative velocity. Hence, the static temperature, pressure, density
with, W
and velocity are evaluated at the boundary, represented by subscript b, as follows
a2in
,
Tb = Tt
a20
Tb 1
Pb = Pt
,
Tt
Pb
b =
,
RTb
p
~ b | = 2cp (Tt Tb ).
|W
(3.94)
The values os stagnation inlet conditions Pt and Tt are fixed by the user. It should be
~ b | according
observed that the inlet velocity components are obtained by decomposing |W
to the two aforementioned prescribed flow angles.
3.7.2.2
Outlet
The outlet conditions depend also on the flow regime. For supersonic outlet, all properties are extrapolated from adjacent interior control volume. For subsonic exit flow the
static pressure is known and the static temperature is calculated using the density from
adjacent interior control volume. Therefore the internal energy and the total energy can
be calculated. The velocity is extrapolated from adjacent neighbors.
As commented in the section 3.7.2.1, the outlet boundary condition is enforced with
102
non-reflecting condition in the case of subsonic outflow. In this case, four flow variables
( and the three velocity components) have to be extrapolated from the adjacent interior
control volume. For this case the following formulation was inplemented
pb = pout ,
(pb pin )
,
a20
(pb pin )
Wxb = Wxin + nx
,
0 a0
(pb pin )
Wyb = Wyin + ny
,
0 a0
(pb pin )
Wzb = Wzin + nz
,
0 a0
b = in +
(3.95)
where pout is the outlet static pressure specified by the user, pin is the static pressure
calculated with the extrapolated variables.
The static pressure profile known, for example, from a streamline curvature program
or from experimental work, were used in this work.
3.7.2.3
Farfield
For farfield boundary condition the stagnation pressure and stagnation temperature,
the velocity (or Mach number) and flow direction and turbulence level must be given.
Care must be taken to avoid reflection of perturbations back into the computational
domain. Depending on the sign of the eigenvalues of the convective flux, the information
is transported out of or into the domain along the characteristics. In the case of subsonic
inlet, there are four incoming characteristics and one outgoing. For subsonic outlet the
situation is inverse. Hence, the flow can either enter or leave the domain. The Riemann
invariant [92], [105] and [106] was implemented to achieve the correct propagation of the
103
disturbances. Those invariants use the propagation of information among the freestream
properties set up at the ghost elements and its internal neighbors. The calculations are
performed at the face of the control volumes with freestream conditions:
1
qf ace ~n = (Rin + R ),
2
(3.96)
1
af ace = ( 1)(Rin R ),
4
(3.97)
Mf ace =
qf ace ~n
,
af ace
(3.98)
with the subscript standing for freestream and in for the adjacent interior control
volume,
q ~n = (Wx ) nx + (Wy ) ny + (Wz ) nz ,
(3.99)
(3.100)
~n represents the outward normal vector, a and M the speed of sound and Mach number,
respectively. Thus
R = q ~n
2a
,
1
(3.101)
Rin = q ~n
2ain
.
1
(3.102)
For |Mf ace | > 1 there is supersonic propagation of the information at the face, so that
104
if qf ace ~n > 0 the quantities at face are extrapolated from interior of control volume:
f ace = in ,
(Wx )f ace = (Wx )in ,
(Wy )f ace = (Wy )in ,
(Wz )f ace = (Wz )in .
Otherwise, if qf ace ~n < 0, the quantities at the face are extrapolated from freestream
f ace = ,
(Wx )f ace = (Wx ) ,
(Wy )f ace = (Wy ) ,
(Wz )f ace = (Wz ) .
For |Mf ace | < 1 there is subsonic propagation of the information at that face. Therefore
if qf ace ~n > 0 the quantities at that the face are extrapolated from the interior control
volume:
f ace =
in a2f ace
pin
1
1
105
f ace =
a2f ace
p
1
1
and the static pressure and total energy are calculated respectively by
(3.103)
pf ace
1
+ f ace [(Wx )2f ace + (Wy )2f ace + (Wz )2f ace ].
1 2
(3.104)
pf ace =
ef ace =
Qin + Qghost
,
2
(3.105)
(3.106)
Qf ace =
3.7.2.4
Periodic
It is assumed that the channels formed by adjacent blades are identical, so that just
only one blade channel is used, with significant reduction in computer memory and processing time. Being the flow periodic, the periodicity condition must be applied to the
blade inlet and outlet regions. For simulation of rotating frame of reference, this type of
106
= in ,
Wx = (Wx )in cos + (Wy )in sin ,
Wy = (Wx )in sin + (Wy )in cos ,
Wz = (Wz )in ,
e = ein ,
where, = 2/nblades, the subscript in stands for adjacent interior control volumes and
nblades is the number of blades.
Attention must be given to unsteady flows, because the periodicity should provide
terms that include the time-shift conditions across a blade passage and to establish the
dependency of the dissipation and dispersion errors on the time integration and mesh
dependence. For unsteady flow the management of space and time periodicity is calculated
using a time inclined computational grid and the phase lag condition, as described by
[102, 108].
It is evident that if inlet distortion must be taken into account it is necessary to
calculate the flow over the entire compressor inlet, not only over one blade passage.
107
Wall
Walls in a turbomachine are still or rotating. Rotating walls are formed at every piece
that is rotating with the turbomachine shaft: rotor blades, rotor blades base and drums
to which rotor blades are attached. Otherwise the walls are not rotating with the shaft,
therefore still.
Walls can be associated with the inertial frame of reference, as the outer casing and
stator blades. They may be also associated to rotating frames, like the part of the still
walls in a region of rotating meshes. Therefore, appropriate wall conditions must be set.
(3.107)
where, ~n is the unit outward normal vector, ~t is the tangential vector and ~c = ~n ~t.
The normal velocity of the real and ghost control volumes have opposite signals.
Both tangential and cross product are extrapolated from adjacent interior control
volume, as well as density and total energy.
For the viscous case, all quantities are extrapolated from adjacent interior control
volumes. In order to set the flow velocity equal to the wall velocity
108
The remaining variables were set equal to the interior neighbor volume.
2. Fixed Walls (Non-Rotating) in a Rotating Frame of Reference: This category of boundary condition is applied at casing of rotor rows that non-rotates with
the frame of reference. The flow properties are calculated based on the velocity tri~ +W
~ , where the velocity vectors stand for absolute, tangencial
angle equation V~ = U
~ =
and relative velocities, respectively and U
~ ~r and
~ the angular velocity of the
rotating frame of reference. Hence, the flow properties are given by
= in ,
Wx = (Wx )in + 2yw ,
Wy = (Wx )in 2xw ,
Wz = (Wz )in ,
(yw (Wx )in + xw (Wy )in )
+ rin
e = ein + 2in rw
rw
with subscript w meaning wall and the variables xw and yw coordinates of the wall
face centroid.
3. Rotor Walls: For walls with rotating frame, for example rotor blades, the wall
boundary condition presented in Section 3.7.2.5 can be used, with the inclusion of
the term 2in rw (rin rw ) 2 in the energy equation due to the different radii of the
element and its ghost neighbor.
109
Symmetry
The symmetry conditions are the same for wall conditions for inviscid case, set up at
the periodic walls.
3.7.2.7
Rotor/Stator Interface
The rotor-stator interface requires careful attention, because the flow leaving the rotor
rotates, while the stator is still, causing the flow to be unsteady if observed from the
stator. Therefore, the correct transfer of information between rotor and stator is complex. Approximations are made although accuracy should not be lost. Depending on the
physics of the problem, i.e. steady or unsteady, the numerical treatment is different. In
this work, the experiences of Denton, Dawes, Chen and Belardini [29, 30, 109, 110] where
taken into account to calculate the flowfield in the rotor-stator interface. The methodology must represent the ability to capture flow disturbances caused by wakes from the
upstream blade row that may produce significant unsteady lift on the following blade
row and eventually lead to fatigue and material failure. Other disturbances associated
with pressure oscillations may have strong influence if the blade spacing is small, causing
unsteady forces. Chen et al [31] discuss in more details the problem, with numerical simulations. The steady-state simulation is implemented in this work, using the mixing-plane
approach.
3.7.2.8
Mixing-Plane
A surface, called mixing-plane, is defined in the vaneless space between two consecutive
rows, usually lidway between the consecutive rows. At both sides of the surface, two sets
110
of ghost volumes are created to store circumferential averages of the properties of the
flow entering and leaving the mixing-plane. Geometrical properties of those mixing-plane
ghost volumes are calculated. Figures 3.3 and 3.4 show the mixing-plane at a rotor row
outlet and stator row inlet respectively.
The mixing-plane is divided radially in strips. For control volumes contained in each
radial strip the conserved variables are averaged and stored as constant values for those
111
volumes. The average quantities are transferred to the downstream ghost volumes. The
average flow properties in the mixing-plane are re-calculated at each iteration. One of the
first approach to compute the mixing-plane was described by Adamczyk et. al. [108, 111].
Details of such technique can be seen in the work of Gerolymos [112]. The disadvantage
of the mixing-plane is that non-linear effects and convection phenomena may be lost and
additional entropy created.
Some authors [113, 114] compute the mixing-plane with body forces to account for the
potential interaction between closely coupled (staged) rows using deterministic stresses
for the average wake blockage and mixing effects, both axially and radially. For steadystate simulations a very robust methodology is presented in reference [115], what is expected to be implemented in the future in the code developed during this research.
3.7.3
The stop criteria is based on numerical residue or number of iterations that the user
specified. The residue is calculated as the logarithm of RHS of the conserved equations
based on the L2 norm as presented bellow
v
u
2
vol
u 1 N
X
Qi
t
,
error = log10 (RHS) = log10
N vol i=1
t
(3.108)
v
u
vol
u 1 N
X
t
error = log10 (RHS) = log10
(Qni Qin1 )2 ,
N vol i=1
(3.109)
or
where, N vol is the total number of control volumes and n is the time integration iteration
number.
112
Generally only continuity equation is used to calculate the residue. Another option
is calculate the difference Qn Qn1 for each conservation equation to accompany the
decaying of each conserved variable during the iterations. For turbomachinery simulation,
the monitoring of the outlet mass-flow, pressure ratio and efficiency are good option
to observe the convergence. Generally, when the outlet mass-flow, pressure ratio and
efficiency stay unaffected during the numerical iteration, the solution was obtained. For
turbomachinery flow simulations another stop criteria was implemented,
v
u
2
5 N
vol
u 1 X
X
Qik
t
error = log10 (RHS) = log10
N vol k=1 i=1
t
(3.110)
v
u
vol
5 N
u 1 X
X
t
(Qnil Qin1
)2 ,
error = log10 (RHS) = log10
l
N vol l=1 i=1
(3.111)
or
4 Computational Implementation
CFD is based on numerically solving a set of partial differential equations that describes
the conservation of some primary flow variables in time and space. Most flow problems
using Navier-Stokes equations do not have an analytic solution. Thus, CFD methods are
based on numerical solutions of the discretized governing equations through some iterative
process using powerful computers.
At the current state of development, the CFD solver developed in this work, can be
used to simulate external and mainly internal turbomachinery flowfields fairly realistically
and accurate. Many issues should be better understood before starting a CFD simulation:
numerical stability, discretization process, mesh refinement, turbulence modeling, among
many others. Before using a specific CFD code with confidence, it is necessary to verify
and validate the code in order to certify the capability of code to describe the physics of
the flowfield correctly.
4.1
Usually, CFD codes are made of a large number of subroutines, structured in a modular way to make simpler and easier further development, such as the implementation of
new discretization methods, accelerating techniques, boundary conditions and turbulence
114
models. If the target of code developers is to optimize memory storage, CPU time, the
complexity of the program data structure then the vectorization of variables should be
carefully analyzed.
In this work the code was written in FORTRAN 90. The program is procedural and
performs the sequence: read input data, calculate mesh-associated geometrical parameters, solve system of differential equations (inviscid or viscous), write output files.
Although the program has been written for axial turbomachines, internal and external
flows can also be solved. An input file contains the appropriate values for:
1. Fluid properties,
2. Selection of equations to be solved: Euler, Navier-Stokes or Navier-Stokes with
turbulence model,
3. Flow type: internal or external,
4. Inlet condition: subsonic or supersonic,
5. Turbulent Prandtl number (only for turbulent flows),
6. Ratio of turbulent eddy viscosity to molecular viscosity (turbulent /laminar ) or turbulent intensity (only for turbulent flows),
7. Spatial discretization: centred-scheme, flux-vector splitting or flux-difference splitting,
8. Time discretization: Runge-Kutta or MacCormack,
9. Artificial dissipation method and its coeficients (only for centered-scheme),
10. CF L number,
115
After reading these input parameters, the program runs the mesh data treatment
consisting of the following steps:
116
References [83, 87] contain further details on the mesh procedures used in this work.
Boundary conditions are set at the ghost elements. Each ghost element has the same
characteristics of its actual neighbor element. The characteristic length is set as the
minimum distance from the element centroid to the volume faces, and it is used for the
calculation of the CF L number.
Having calculated the mesh-related properties, the initial and boundary conditions
are set. The initial condition depends on the type of flow to be calculated: internal or
external, with or without rotating frame of reference.
For turbomachinery flow simulations the initial condition must comply with the complexity of flow physics. All rotating elements are flagged for the sake of elements flux
calculation based on relative motion. Two frames of reference are used: fixed and rota-
117
tional. Initial condition can be set based on previous estimates from a streamline curvature
code. The more accurate the initial guess the faster the convergence solution is achieved.
Inaccurate initial condition may drive the calculations to an unrealistic solution or to
instable operation such as surge or choke, i.e. unsteadiness. Problems involving these
operation points are extremely unsteady and, they cause numerical divergence. For compressors, the initial condition is more critical because if the initialization means some
instable operation point as stall, surge or choke points the numerical solution will diverge
in the case of steady-state regime simulation.
In this work, the scheme created to identify the number of blades in each blade row
is based on the axial coordinates C1 to C6 shown in Figure 4.1. Those coordinates are
taken from the streamline curvature program. understanding.
For example, to identify the periodicity imposed in the first stator (S1), the coordinates
of C3 and C4 are used. The CFD solver searching for the mesh elements without rotational
speed that is the case of stator blade. For rotors, the same scheme is used, but for elements
with rotational speed, for example, to identify the periodicity imposed in the second rotor
118
For example, to identify the mixing-plane between the first stator (S1) and second
rotor (R2), the coordinates of M P 2a and M P 2b are used.
The spatial and time integration of the equations in the conservative form is then
carried out. This procedure calculates:
119
Concerning the turbulence model, the program gives the user the possibility of delaying
or specifying the frequency at which it is solved, aiming reduction of CPU time.
The iterations might be stopped if either convergence or divergence is achieved or after
a prescribed number of iterations. Output files are written according to the user needs.
A TECPLOT-like file is also generated.
The computational domain for an axial compressor or turbine is divided in segments
for each blade row in order to distinguish the type row that rotates (rotor) or stays still
(stator).
The time integration uses the five-step Runge-Kutta scheme with second-order accuracy. Viscous terms are often evaluated only in the first step for computational efficiency.
120
Convective terms are computed at every step. Centered scheme does not inherently damp
high-wave-number disturbances and hence requires the addition of artificial dissipation
when applied to hyperbolic equations. The fourth-order operators (Biharmonic) are implemented to damp high-wave-number errors and the second-order operators (Laplacian)
to improve shock capturing. The artificial dissipation procedure is introduced at alternate
steps.
Variable time-step based on the linear stability analysis of the discretized Navier-Stokes
equations, is used to accelerate convergence. A IRS method can be used to introduce
implicit-like character to the solver, as well as to increase the maximum operational CF L
number. IRS has the effect of smoothing the high-frequency variations of the RHS of
the conservation equations. This smoothing can be applied either at alternate steps or at
every step.
For space integration, the user can choose either the techniques of Jameson (centereddifference), Van Leer (flux-vector splitting) or Roe (flux-difference scheme). For spatial
discretizations using F DS, as presented in Section 3.2.3, a limiter function is also used at
alternate steps. When an upwind scheme is chosen, the artificial dissipation procedure is
automatically skipped. The dissipation terms in the Roe scheme are calculated using the
Roe-matrix. Special attention must be paid to the matrix of conserved variables of Roe
scheme. Its original form {, Wx , Wy , Wz , e}T . For the implementation pressure and
enthalpy are used instead of and e: {p, Wx , Wy , Wz , h}T .
The F V S method by Van Leer achieved accurate solutions (and is quite robust) and
perform very well in the case of Euler equations. In [116, 117] the authors carried out
this scheme with viscous simulations and the results revealed that splitting errors in the
Navier-Stokes and energy equations smear out the boundary layers and produce inaccurate
121
stagnation and wall temperatures. To avoid this sort of problem, Van Leer [118] suggested
modifications in the momentum flux terms and Hanel [119] presented a modification of the
momentum and energy equations and results for transonic airfoil. These modifications
are not implemented in the CFD solver of current work. In this work, the F V S was
used only to start the inviscid terms calculation to obtain a reasonable velocity, pressure
and temperature fields and after some number of iterations the spatial discretization is
switched by another (centered or FDS).
For turbulent flow simulations it is necessary to supply to the CFD solver the value
of T at boundary. For external flow T must be given at farfield. For internal flow T
must be given at inlet. The Spalart-Allmaras turbulence model calculates the variable
presented in Equation 2.48. The turbulent kinematic viscosity coefficient is used to
calculate the turbulent dynamic viscosity coefficient necessary to calculate the terms of
energy equation. For initial condition the value of can be estimated by /10. Other
option is to estimate the value using the turbulence intensity [120]
r
=
3~
W Il,
2
(4.1)
where I is the turbulent intensity and l is the turbulent length scale. The turbulent
intensity can be estimated by the user or calculated using the following equation
1/8
I = 0.16(Re )DH ,
(4.2)
where the Reynolds number is calculated based on the hydraulic diameter defined by
DH =
Dt2 Dh2
,
Dt + Dh
(4.3)
122
where DT is the tip diameter and DH is the hub diameter. In the case of turboamachinery,
this values are know from the design. The length scale can be estimated by
l = 0.07DH ,
(4.4)
I = 0.499 ,
(4.5)
y+ =
u yp
,
(4.6)
r
u =
w
,
(4.7)
123
and
w =
2
0.037W
1/5
(Re )
As implemented, the code can be used for internal and external flows.
(4.8)
a Inviscid part, using the convergent-divergent NASA nozzle and the NACA0012 in the
transonic regime as guidelines.
b Laminar (viscous) part, introducing time, space and artificial dissipation models in
order to study the flat-plate.
c Turbulent part, using Spalart-Allmaras turbulence model for the study of NASA convergentdivergent nozzle. Results of the turbulent flow in axial turbomachines are presented
in the next chapter.
This chapter deals with the verification of the adequacy of the code developed during
this research, as well as with its validation, aiming at the application to the design and
design improvement of turbomachines.
The verification was carried out for the flow over flat-plate. Several discretization
schemes in time and space aiming at testing different artificial dissipation (for cell-centered
spatial discretization) and upwind schemes.
The validation was carried out through the study of a convergent-divergent nozzle and
a NACA aerodynamic profile aiming at testing different spatial discretization schemes.
125
The reference [121] presents a CFD code validation process and the definition of verification, validation and calibration of computational codes. In order to test the CFD
solver robustness and accuracy of the developed code both external and internal flows
study cases were solved and compared to the literature. The implementation of the code
was validated in a step-by-step manner. First, the Euler equations were implemented
and tested. Afterwards, the Navier-Stokes equations and turbulent flows were simulated.
Considerable amount of time was spent implementing the numerical methods to solve the
governing equations in a 3 D domain considering the full Navier-Stokes equations. It
is important to highlight that the geometry and mesh generation process was very time
consuming due to the geometry complexity.
Specifically, for turbomachinery cases, there are publications available showing the validation process based on experimental results. However, the geometry of the compressor
and turbine are usually not supplied due to restrictions on publishing design information.
Modern blades design have complex camber angles and chord distributions from hub-totip in order to improve machine performance by decreasing internal flow losses. High
performance turbomachines are a key on aircraft and industrial powerplant installations
and business competition. The state-of-the-art on gas turbine components design is rarely
described by researches with details. The mathematical modeling of empirical correlations
is vastly used by designers during the preliminary and refining design process. To obtain
this correlations experimental tests must be carried out at test facilities.
5.1
126
Inviscid Case
For Euler equations, two test cases were simulated, namely a supersonic nozzle and the
NACA0012 airfoil. The nozzle test case is based on the experimental data of a convergingdiverging nozzle at static conditions described by Mason [122]. The work developed by
Carlson [123], presents the same nozzle analysis, but based on numerical simulation. All
nozzle geometrical details can be obtained from both references.
The mesh has 241 61 3 nodes in the x, y and z coordinates, respectively. Hence,
44, 103 nodes representing 28, 800 control volumes. Figure 5.1 shows the nozzle geometry
and mesh.
FIGURE 5.1 Nozzle geometry and mesh (flow is from left to right)
Stagnation pressure and temperature are set at inlet boundary and static pressure is
specified at outlet boundary. The symmetry boundary condition is used on both sides
and the symmetry condition is imposed at the nozzle centerline. Wall boundary condition
is used at the nozzle upper surface.
The first calculation was carried out with pressure ratio equal to 2 between inlet
and outlet. A centered-difference scheme was chosen for the spatial integration with the
artificial dissipation model described in the last paragraph of Section 3.4.1. The artificial
dissipation coefficients used were: (2) = 1/2 and (4) = 3/256. The CF L number was set
as 2. The IRS method was applied at alternate Runge-Kutta steps. Figure 5.2 shows the
127
1.2
p/p0
0.8
0.6
0.4
0.2
Experimental
Numerical Solution
0
1
0.5
0.5
x/x0
FIGURE 5.2 Experimental and numerical results for static pressure ratio 2 - inviscid
flow
The second calculation was carried out with pressure ratio equal to 5 between inlet and
outlet. For this simulation, a centered and a Roe upwind scheme were used for the spatial
integration. The centered-difference scheme was applied with the artificial dissipation
model of Equation 3.70. The flux-difference splitting was calculated with two different
values as described in Section 3.2.3 in order to to analyze the influence of this parameter
on the convergence rate and solution. The Roe discretization scheme was applied using
the piecewise linear reconstruction and the Venkatakrishnans limiter described in Sections
3.2.4 and 3.2.5, respectively. Figure 5.3 shows the experimental and the numerical results
of both spatial discretization methods aforementioned at wall surface. The flux-difference
splitting scheme achieves very similar results for both values (0.007 and 0.15).
Minor differences between the experimental and numerical results can be observed.
In the region after the nozzle throat, these differences are some due to the interaction
128
1.2
p/p0
0.8
0.6
0.4
0.2
Experimental
Centereddifference
FDS, =0.007
FDS, =0.15
0
1
0.5
0.5
x/x0
FIGURE 5.3 Experimental and numerical results for static pressure ratio 5 for centered
and upwind schemes - inviscid flow
between the boundary-layer and the shock waves at near wall regions. They are not
detected because the viscous (or diffusive) terms are neglected in the Euler formulation.
In the section 5.3, the same case was calculated with the full Navier-Stokes equations plus
the turbulence transport equation and the results after the nozzle throat were improved
slightly.
Figure 5.4 shows the convergence histories of the last case described (centered and
Roe upwind schemes). Note that the parameter has a strong influence on the numerical
behavior. This numerical residual is calculated based on Equation 3.109.
The Mach number contours for both centered and upwind schemes are presented in
Figures 5.5 and 5.6. Oblique shock waves can be seen at the divergent nozzle section.
129
log10(RHS)
10
15
20
Centereddifference
FDS, =0.007
FDS, =0.15
25
0
10000
20000
30000
40000
iteration
50000
60000
70000
80000
FIGURE 5.4 Continuity residue histories for different spatial discretization schemes inviscid flow
FIGURE 5.6 Mach number contours for upwind scheme - inviscid flow
130
A transonic NACA0012 airfoil profile was simulated with Mach number 0.8 and zero
angle-of-attack. The mesh generated has 236, 000 control volumes. The computational
domain has 60 times the airfoil chord. An O-Grid mesh type was generated as presented
in Figure 5.7. Other airfoil profiles and experimental results can be found in [124].
Periodic, wall and farfield boundary condition were used. The Riemann invariants
option was applied to increase the numerical robustness and to avoid unphysical numerical perturbations inside of the computational domain. The CF L number used in this
simulation is 1.5. IRS was applied at alternate Runge-Kutta steps.
For this simulation, centered and Roe upwind schemes were chosen for the spatial
integration. The centered-difference scheme was calculated with the artificial dissipation model proposed by Jameson [76] and presented in the Section 3.4.1.2 using artificial
dissipation coefficients: k (2) = 1/4 and k (4) = 1/256.
Figures 5.8 and 5.9 show the pressure coefficient (cp ) distribution along the airfoil
chord for both spatial discretization schemes.
131
Cp
0.5
0.5
Experimental
Centereddifference
1
0
0.2
0.4
0.6
0.8
x/c
FIGURE 5.8 Pressure coefficient distribution on the NACA0012 airfoil with zero angleof-attack and Mn = 0.8: centered-difference
Cp
0.5
0.5
Experimental
FDS, =0.1
1
0
0.2
0.4
0.6
0.8
x/c
FIGURE 5.9 Pressure coefficient distribution on the NACA0012 airfoil with zero angleof-attack and Mn = 0.8: upwind
The shock wave formation can be visualized in Figures 5.10 and 5.11, that show the
Mach number contours for both spatial discretization schemes. Figures 5.12 and 5.13
132
133
5.2
134
Laminar Case
The laminar flow test cases were carried out to validate the CFD solver. The well
known Blasius velocity profile of a flat-plate was calculated numerically by the CF D
solver. The numerical results are compared to the analytical solution presented in [125].
The flat-plate length was calculated based on Re number and freestream Mach number.
The laminar Re number used was 5 105 with M = 0.3.
The solver verification for the laminar case was carried out for different flux computation schemes and artificial dissipation models. The viscous terms were calculated based
on the centered scheme presented in Section 3.2.6.
The number of control volumes inside of the boundary-layer region is very important
to obtain an accurate boundary-layer resolution. Figure 5.14 shows the approach used
to discretize the physical domain. The mesh has 551 61 4 nodes in the x, y and z
coordinates respectively. Hence, 134, 444 nodes representing 99, 000 control volumes.
The part of the mesh in blue shows the computational domain in front of the flatplate leading edge. The green part of the mesh shows the computational domain over
the flat-plate surface. Note that the mesh is finer close to the flat-plate leading edge.
The distance between the first node on wall surface is 2 106 m. This small distance is
necessary to capture the correct boundary-layer thickness. An exponential function was
used to stretch the mesh smoothing along the j direction.
135
Symmetry, wall and farfield boundary conditions were used. The Riemann invariants
option was used to increase the numerical robustness and avoid unphysical numerical perturbations inside of the computational domain. The CF L number used in this simulation
is 1.5 for the centered spatial discretization scheme and 0.7 for the flux-difference scheme.
The IRS was applied only for centered scheme in alternate Runge-Kutta steps. A second case was calculated with the MacCormack time-integration method as described in
Section 3.3.1 without IRS and CF L = 0.3. The artificial dissipation coefficients used are
the same presented in Section 3.4.1.
The coordinates x and y are represented by
r
=y
U
x
(5.1)
f 0 () =
u
U
(5.2)
The first case was solved with 10 nodes inside of the boundary-layer. The centereddifference scheme was calculated with artificial dissipation model proposed by Mavriplis
and Jameson presented in the Equation 3.70. Figure 5.15 shows the continuity numerical residue history (convergence rate). Figure 5.16 shows the analytical and numerical
solutions at 50% of flat-plate distance from leading edge.
The second case was solved with 15 nodes inside of the boundary-layer.
The
centered-difference scheme was used for the spatial integration artificial dissipation model
described in the last paragraph of Section 3.4.1. Figures 5.17 and 5.18 show the continuity
numerical residue and the analytical and numerical solutions at 50% of flat-plate distance
from leading edge, respectively.
136
log10(RHS)
Centereddifference
8
0
5000
10000
15000
20000
25000
iteration
30000
35000
40000
45000
50000
1.2
f()
0.8
0.6
0.4
0.2
Analitical Solution
Centereddifference
0
0
10
FIGURE 5.16 Analytical and numerical solutions of flow with M = 0.3 on a flat-plate
with 10 nodes inside of the boundary-layer: first case
The third case was solved with 15 nodes inside of the boundary-layer. The centereddifference scheme was calculated with artificial dissipation model proposed by Jameson
[76] in Section 3.4.1.2. Figures 5.19 and 5.20 show the continuity numerical residue and
137
log10(RHS)
Centereddifference
8
0
5000
10000
15000
20000
25000
iteration
30000
35000
40000
45000
50000
1.2
f()
0.8
0.6
0.4
0.2
Analitical solution
Centereddifference
0
0
10
FIGURE 5.18 Analytical and numerical solutions of an flow with M = 0.3 on a flatplate with 15 nodes inside of the boundary-layer: second case
the analytical and numerical solutions at 50% of flat-plate distance from leading edge.
The fourth case was solved with 10 nodes inside of the boundary-layer. The high
order flux-difference splitting scheme was calculated with the parameter = 0.1. For this
138
log10(RHS)
Centereddifference
8
0
5000
10000
15000
20000
25000
iteration
30000
35000
40000
45000
50000
1.2
f()
0.8
0.6
0.4
0.2
Analitical Solution
Centereddifference
0
0
10
FIGURE 5.20 Analytical and numerical solutions of an flow with M = 0.3 on a flatplate with 15 nodes inside of the boundary-layer: third case
case, the mesh size is five times lower than the three previous cases. Figures 5.21 and
5.22 show the continuity numerical residue and the analytical and numerical solutions at
25%, 50% and 85% of flat-plate distance from leading edge.
139
log10(RHS)
FDS
8
0
5000
10000
15000
20000
25000
iteration
30000
35000
40000
45000
50000
1.2
f()
0.8
0.6
0.4
0.2
Analitical Solution
FDS: 25% of flatplate
FDS: 50% of flatplate
FDS: 85% of flatplate
0
0
10
FIGURE 5.22 Analytical and numerical solutions of an flow with M = 0.3 on a flatplate with 10 nodes inside of the boundary-layer: fourth case
The fifth case was solved with 12 nodes inside of the boundary-layer. The solver
settings are the same as the fourth case. Figures 5.23 and 5.24 show the continuity
numerical residue and the analytical and numerical solutions at 50% of flat-plate distance
140
log10(RHS)
FDS
8
0
5000
10000
15000
20000
25000
iteration
30000
35000
40000
45000
50000
1.2
f()
0.8
0.6
0.4
0.2
Analitical Solution
FDS
0
0
10
FIGURE 5.24 Analytical and numerical solutions of an flow with M = 0.3 on a flatplate with 12 nodes inside of the boundary-layer: fifth case
The sixth case was solved with 15 nodes inside of the boundary-layer. The solver
141
settings are the same as the fourth case, but the mesh size is the same of the second case
(99,000 control volumes). This case was started with the solution obtained in the second
case. Figure 5.25 show the analytical and numerical solutions at 50% of flat-plate distance
from leading edge.
1.2
f()
0.8
0.6
0.4
0.2
Analitical Solution
FDS
0
0
10
FIGURE 5.25 Analytical and numerical solutions of a flow with M = 0.3 on a flat-plate
with 15 nodes inside of the boundary-layer: sixth case
Figures 5.26 and 5.27 show the details of the Mach number contours and the velocity
vectors of the flow on the flat-plate simulated, respectively.
Comparing the results of the first and fourth cases it is possible to conclude that the
flux-difference splitting scheme is very robust and accurate to capture the boundary-layer
with a smaller number of nodes (inside of the boundary-layer) compared to the centered
scheme. This happens due to the nature of the centered scheme. The centered scheme
is more dissipative than the flux-difference splitting scheme. However, in the CFD solver
developed in this work, the Roe scheme demands about 1.25 times the machine time of
the centered scheme. To make the flux-difference splitting scheme faster per iteration
142
FIGURE 5.26 Details of the Mach number contours inside of the boundary-layer on the
flat-plate
FIGURE 5.27 Details of the velocity vectors profiles inside of the boundary-layer on the
flat-plate
the gradients of the limiter function can be calculated only at alternate Runge-Kutta
time-integration steps.
For centered-difference scheme it is possible to start the solution with higher values
for the artificial dissipation coefficients. However, these values must be decreased as the
calculation progresses in order to avoid too high numerical dissipation.
Each iteration of the MacCormack time-integration is more faster than the five steps
Runge-Kutta scheme. However, the MacCormack time-integration can be very instable,
143
in the case of laminar simulation, for CF L greater than 0.5, without IRS.
It is important to mention that the CFD solver developed in this work aims the solution
of compressible flows. The Mach number of the flat-plate (M = 0.3) is relatively low
for a compressible code (high velocities). Incompressible treatment should be used for
low Mach number simulations. This limitation is not physical. The numerical errors may
dominate the very small variations of the density and compressibility effects.
5.3
A turbulent flow is calculated at the nozzle presented in Section 5.1. The mesh used
has 281913 nodes in the x, y and z coordinates, respectively, 76, 713 nodes representing
50, 400 control volumes. Figure 5.28 shows the mesh generated for nozzle geometry.
FIGURE 5.28 Mesh generated for nozzle flow calculation - turbulent flow
The calculation was carried out with pressure ratio 5 between inlet and outlet. For this
simulation, centered and Roe upwind schemes were chosen for the spatial integration. The
centered-difference scheme was calculated with the artificial dissipation model proposed
by Jameson [76] in Section 3.4.1.2. The flux-difference splitting was calculated with =
0.1 using the piecewise linear reconstruction and Venkatakrishnans limiter described in
Sections 3.2.4 and 3.2.5, respectively. Figure 5.29 shows the experimental and numerical
results for both spatial discretization methods aforementioned.
144
1.2
p/p0
0.8
0.6
0.4
0.2
Experimental
Centereddifference
FDS, =0.1
0
1
0.5
0.5
x/x0
FIGURE 5.29 Comparison of experimental and numerical results for static pressure
ratio 5 for centered and upwind schemes - turbulent flow
Figure 5.30 shows the convergence histories during the numerical iteration.
log10(RHS)
Centereddifference
FDS, =0.1
7
0
10000
20000
30000
40000
iteration
50000
60000
70000
80000
FIGURE 5.30 Continuity residue histories for two spatial discretization schemes
The Mach number contours are presented in Figure 5.31. Oblique shock waves can
145
be seen at the divergent nozzle section. Figure 5.32 shows the velocity vectors in the
boundary layer region.
The turbulent flow solutions of the nozzle presented accurate results compared to the
experimental results, including those in the region after the nozzle throat, for both spatial
discretization scheme.
6 Results
The study described in this chapter refers to the design and improvement design of
axial turbomachines. A 5-stage high performance axial-flow compressor as part of an ongoing design of the gas generator of a 1 MW class turboshaft. Therefore, no experimental
data are available for comparison.
6.1
CFD for turbomachines is inherently complex and laborious due to the complexity of
the geometry and physical aspects. The design process is a complex and laborious task
by itself. To obtain a final 3 D geometry, each projected geometry must be carefully
analyzed through specific tools. CFD should be used as an aid tool to find an optimal
geometry. From the results obtained from the CFD simulation, it is possible to point
out pitfalls and hence improve the design. A long way is necessary until start the CFD
calculations.
Based on a 3 D geometry, a mesh must be generated, creating curves, surfaces,
points, vertices and edges to generate the respective blocks around blades and channels.
For compressors, the process is more difficult, because of dimensions at leading and trailing
edges being very small. The mesh refinement on these regions is critical for turbulent flows.
CHAPTER 6. RESULTS
147
Not only the refinement, but the quality and the aspect ratio of the elements when the
mesh smoothing process is applied must be guaranteed to obtain a good mesh as close as
possible of the orthogonal form.
6.1.1
A single-stage axial-flow turbine designed using the meanline technique was calculated.
The geometry was obtained using an inviscid formulation with the addition of loss models
developed for turbine design. Loss corrections by Ainley and Mathienson [126], Dunham
and Came [127], Kacker and Okapu [128] and Moustapha and Kacker [129] were used.
The parameters used in the design are:
The 3 D views of the turbine are showed in Figures 6.1, 6.2 and 6.3.
The three mesh types most commonly used for turbomachinery domain discretizations
are: H-grid, H-O-H-grid and O-grid. Each one with particular characteristics. H-grids
CHAPTER 6. RESULTS
148
FIGURE 6.2 3-D solid drawing of the NGV and rotor of a single-stage axial-flow turbine
are vastly used due to their relative simplicity compared to both H-O-H and O-grids [29].
Generally, H-O-H and O-grids are superior because the leading and trailing edges are more
accurately represented by the mesh. The problem of the H-grid is that the curvature of
the leading and trailing edges are partially lost. To try to overcome this problem a very
refined grid should be generated in this region. Figure 6.4 shows the H-grid generated in
CHAPTER 6. RESULTS
149
Figure 6.5 shows the mesh close to the stator leading edge. Note that, the blade
leading edge radius is not well represented.
H-O-H-grids use a different blocking scheme. O-grids are generated only around the
CHAPTER 6. RESULTS
150
A closeup of the improved curved part is presented in Figure 6.7. Note that the O-grid
is generated only on the blade surface. Hence, an H-grid is generated between the blades.
The H-O-H-grid type is commonly used for turbine blades. For compressor blades the
CHAPTER 6. RESULTS
151
FIGURE 6.7 Closeup of the stator trailing edge and rotor leading edge of a single-stage
axial-flow turbine
mesh type, depends on the blade thickness and the dimension of the leading edge.
Figure 6.8 shows an O-grid block. For this mesh type, the O-grid is generated around
the blade geometry and the distance of the elements (or nodes) is smoothly distributed
along the channel between blades. An accurate mesh smoothing process can be obtained
using exponential or parabolic functions.
The O-grid blocking has an advantage over the H-O-H-grid. The blade surface is
CHAPTER 6. RESULTS
152
continuous, because the blade does not need to be divided in two parts (suction and
pressure surfaces). For blades with high camber angle, sometimes it is not trivial to
divide the blade in two different surfaces. The periodic condition is applied to the mean
distance between blades along the blade chord (O-grid). Otherwise, only inlet and outlet
regions would be enforced by the periodicity (H or H-O-H-grids). Figure 6.9 shows an
O-grid closeup.
FIGURE 6.9 Closeup of the O-grid around the stator and rotor blades of a single-stage
axial-flow turbine
Assuming that appropriate meshes are available, the CFD simulations can be carried
out. To verify the implementation of the mixing-plane in the region between a non-rotating
frame (stator) and a rotating frame (rotor) of reference, the flow within a single-stage
axial-flow turbine was calculated using an inviscid formulation. An O-grid with 167, 756
nodes was used. The mass-flow outlet were monitored and are presented in Figure 6.10.
Figure 6.10 shows the mass-flow behavior along the numerical iterations. Note that the
mass-flow is negative at the beginning. This happened due to the numerical initialization
process used for this case. As initial conditions the velocities and the rotational speed
were set equal to zero and the stagnation pressure and temperature were set equal to
CHAPTER 6. RESULTS
153
the inlet condition values. The rotational speed was linearly increased during the first
5, 000 iterations. The rotor outlet static pressure was decreased during the first 2, 000
iterations until reaching the static pressure prescribed in the design. The converged
solution presented a mass-flow equal to 17.98 kg/s.
50
40
MassFlow (kg/s)
30
20
10
Outlet MassFlow
10
0
5000
10000
15000
20000
25000
iteration
FIGURE 6.10 Single-stage axial-flow turbine mass-flow outlet convergence history inviscid case
Figure 6.11 shows the static pressure contour at two planes of a single-stage axial-flow
turbine. It is possible to observe the gas expansion along the turbine stage.
FIGURE 6.11 Static pressure contour of a single-stage axial-flow turbine - inviscid case
CHAPTER 6. RESULTS
154
Figure 6.12 shows the flowfield across the mixing-plane from stator outlet to rotor inlet.
The velocities angles and directions at the stator outlet and rotor inlet are in agreement,
note that at the stator the velocity vectors have the same directions of the blade outlet
and at the rotor inlet the relative flow angle is zero, following the design parameter. The
figure also shows that the flow is well aligned with the stator blade as expected.
FIGURE 6.12 Closeup of the flow across the mixing-plane of a single-stage axial-flow
turbine - inviscid case
In the stator, well distributed velocity field can be observed. In the rotor, the velocity
vectors are incorrect due to the reverse flow presented in this simulation, in which is
inconsistent with the Euler formulation. This is a consequence of the artificial viscosity
addition. However, this viscosity is numerical and not physical. This is the reason for
the mass-flow outlet calculated in the CFD solver to be different of the turbine design
mass-flow. A more detailed investigation should be done. Figure 6.13 shows the velocity
vectors at the rotor outlet and reverse flow region is observed (red cricle).
To calculate the Navier-Stokes equations, an H-grid with 313, 446 nodes was generated
and it is presented in Figure 6.14.
The mass-flow ratio between the mixing-plane outlet of the stator and the mixing-plane
CHAPTER 6. RESULTS
155
FIGURE 6.13 Closeup of the flow at rotor suction and pressure surfaces - inviscid case
CHAPTER 6. RESULTS
156
1.4
1.3
1.2
1.1
MPO/MPI
1
0.9
0.8
0.7
0.6
0.5
Mixingplane mass flow ratio
0.4
0
5000
10000
15000
20000
25000
iteration
FIGURE 6.15 Mixing-plane outlet (MPO) and mixing-plane inlet (MPI) ratio - turbulent
case with H-grid
40
MassFlow (kg/s)
20
20
40
Outlet MassFlow
0
5000
10000
15000
20000
25000
iteration
CHAPTER 6. RESULTS
157
FIGURE 6.17 Single-stage axial-flow turbine velocity vectors distribution along the
stator blade row - turbulent case with H-grid
The problem of reverse flow can be solved at the turbine design process. Modifications
on the rotor should be made to avoid this problem. In this case, the thickness-chord-ratio
was increased and the rotor exit blade angle was changed in two degrees (52.50 to
50.50 ). No modification was made on the stator blade.
FIGURE 6.18 Closeup of the reverse flow at the rotor suction side - turbulent case with
H-grid
An O-grid was used to solve the Navier-Stokes equations at the single-stage axial-flow
turbine with the modification in the rotor blade geometry above mentioned. The mesh
used has 538, 721 nodes. Its mesh block is presented in Figure 6.19. The size of the
CHAPTER 6. RESULTS
158
rotor tip clearance is 0.00075 m. In the tip clearance region some blocks must be created
between the rotor tip and the turbine casing, as shown in Figure 6.19. Figure 6.20 shows
a closeup of the mesh generated on the casing and Figure 6.21 show the gap between the
rotor blade and the casing.
The mesh generated on the rotor tip clearance region is presented in the Figures 6.22
and 6.23.
CHAPTER 6. RESULTS
159
FIGURE 6.21 Detail of the gap between the turbine rotor and casing
CHAPTER 6. RESULTS
160
FIGURE 6.23 Detail of the O-grid around the rotor blade tip and the refinement of the
clearance region
Figure 6.24 shows the mass-flow outlet monitoring during the iterations. Note that
the outlet mass-flow is always positive. Figure 6.25 shows the mass-flow ratio between
the mixing-plane outlet of the stator and the mixing-plane inlet of the rotor.
35
30
MassFlow (kg/s)
25
20
15
10
5
Outlet MassFlow
0
0
2000
4000
6000
8000
10000
iteration
12000
14000
16000
18000
FIGURE 6.24 Single-stage axial-flow turbine mass-flow outlet convergency history turbulent case with O-grid
For this case, the pressure ratio was also monitored. The design pressure ratio is 1.62
and CFD calculated pressure ratio is 1.59 as shown in Figure 6.26.
CHAPTER 6. RESULTS
161
1.1
1.08
MPO/MPI
1.06
1.04
1.02
0.98
0
2000
4000
6000
8000
10000
iteration
12000
14000
16000
18000
FIGURE 6.25 Mixing-plane outlet to mixing-plane inlet ratio - turbulent case with
O-grid
1.7
1.6
pressure ratio
1.5
1.4
1.3
1.2
1.1
pressure ratio
1
0
2000
4000
6000
8000
10000
iteration
12000
14000
16000
18000
FIGURE 6.26 Single-stage axial-flow turbine pressure ratio monitoring - turbulent case
with O-grid
The problem of reverse flow is not present anymore, due to geometrical improvements
on the rotor blade. Figure 6.27 shows the velocity vectors in the turbine rotor region.
CHAPTER 6. RESULTS
162
FIGURE 6.27 Distribution of the velocity vectors in the turbine rotor - turbulent case
with O-grid
The Coriolis force can be observed acting in the velocity vectors close to the rotor
wall, as presented in Figure 6.28.
FIGURE 6.28 Effect of Coriolis force in the velocity field close to the turbine rotor wall
- turbulent case with O-grid
The static pressure along the turbine stage decreases until it reaches the static pressure
prescribed at outlet boundary. Figure 6.29 shows a global view of the static pressure
CHAPTER 6. RESULTS
163
contours in two planes along the turbine stage and Figure 6.30 shows the details of the
static pressure contours calculated by CFD solver.
FIGURE 6.29 Global view of the gas expansion along the turbine stage - turbulent case
with O-grid
FIGURE 6.30 Detail of the static pressure contours along the turbine stage - turbulent
case with O-grid
Figure 6.31 shows the velocity vectors across the mixing-plane. The mixing-plane
CHAPTER 6. RESULTS
164
numerical implementation presents accurate results for all cases solved. The transfer of
information from the downstream to the upstream is accurately calculated.
FIGURE 6.31 Detail of the static pressure contours along the turbine stage - turbulent
case with O-grid
Figure 6.32 shows the tip clearance effect and the flow leakage in the region between
the rotor blade pressure side to the suction side. The clearance between the rotor and
casing wall induces leakage flow across the gap. The leakage flow arises due to a pressure
difference between the two surfaces of the blade at the tip. The mass-flow through this
gap, which does not participate in the energy conversion process, depends on the gap
height. The leakage flow tends to roll up into a vortex. Because the leakage jet and the
main flow are at different angles, a flow discontinuity exists. The formation of leakage has
been observed in turbomachinery and cascades. However, in some cases the conditions
necessary for the formation of a strong vortex may not exist because high turbulence levels,
high velocity or separation zone may diffuse the leakage flow before it forms a vortex. The
secondary and leakage flows oppose each other. Sometimes the effect of leakage flow wash
CHAPTER 6. RESULTS
165
out the separated flow, which is beneficial. The leakage flow and its interaction with other
flow features is a complex phenomenon.
FIGURE 6.32 Detail of the leakage flow from pressure surface to suction surface of the
turbine rotor - turbulent case with O-grid
6.1.2
A single rotor designed using meanline technique was calculated. All geometry was
obtained using an inviscid formulation with the addition of loss models developed for axial
compressor design. The Koch and Smith loss correlations were used [23].
The parameters used in the design are:
CHAPTER 6. RESULTS
166
The O-grid mesh type was generated around the rotor row. The mesh size is of 453543
nodes. Figure 6.35 show the 3D mesh and Figures 6.36 and 6.37 show detail of the mesh
generated on the rotor leading edge and the mesh around the rotor hub, respectively.
Figure 6.38 shows the elements distribution along the rotor chord and from hub-totip. Note that in the axis and casing the mesh is refined and smoothly distributed by an
exponential function.
Figures 6.39 and 6.40 present the rotor outlet mass-flow and the pressure ratio along
the iterations.
CHAPTER 6. RESULTS
167
CHAPTER 6. RESULTS
168
MassFlow (kg/s)
2.8
2.6
2.4
2.2
Outlet MassFlow
2
0
5000
10000
15000
iteration
20000
25000
30000
FIGURE 6.39 Rotor outlet mass-flow convergence history - turbulent case with O-grid
1.6
pressure ratio
1.58
1.56
1.54
1.52
RHS residual
1.5
0
5000
10000
15000
iteration
20000
25000
30000
CHAPTER 6. RESULTS
169
formulation used. The design was calculated using an inviscid formulation and the losses
were calculated with empirical formulation. To improve the results the loss model must be
recalibrated. Blockage coefficient and loss distribution from hub-to-tip should be changed.
Despite the discrepancy, the design results agree with the CFD calculation.
Figure 6.41 shows the static pressure increase along the rotor blade and Figure 6.42
shows a detail of the static pressure on two different planes. Note that the computational
domain is quite deformed in the cartesian space. This makes difficult to obtain a clear
cut plane parallel to the rotor axis.
FIGURE 6.41 Increase of static pressure along the rotor blade - turbulent case with
O-grid
FIGURE 6.42 Static pressure in two different planes (near of hub and near of tip) turbulent case with O-grid
CHAPTER 6. RESULTS
170
Figure 6.43 shows the velocity vectors distribution at the rotor leading edge. Note the
effect of the Coriolis force at the rotor suction surface.
FIGURE 6.43 Velocity vector distribution near of the rotor leading edge - turbulent
case with O-grid
6.2
In this section, an axial compressor design is presented. The procedure, from preliminary design to the CFD calculation, is described. The preliminary design of an axial
compressor is the first step to obtain the operational characteristics and the preliminary
dimensions. At this point, basic compressor dimensions are calculated accounting internal
flow losses using empirical correlations. The inlet and outlet blade angles and the flow
properties are calculated at the compressor blade meanline [1]. In this work, the results
of the compressor performance were used to calculate the engine running line using an
engine deck [4] to analyze the gas turbine behavior when equipped with this machine.
The main steps of a typical turbomachinery aerodynamic design are described bellow:
Specification of design parameters (based on gas turbine cycle calculation),
CHAPTER 6. RESULTS
171
If the values of design parameters (diffusion factor, de Haller number, degree of reaction, incidence, deviation, loss mechanisms among others) are not consistent with the
performance required by the customer, no amount of subsequent CFD computational effort or development testing will enable the performance to be achieved. Hence, despite of
the great simplicity of the methods used, the preliminary design tools remain important
in the compressor aero-thermodynamic design.
A compressor is usually designed to achieve optimum performance for the customer,
with minimum development time and cost. The large range of possible turbomachinery
geometry, particulary in multistage machines, makes it possible to investigate all of the
interesting parameter options by using CFD simulations. It is very important to eliminate
inadequate designs and to obtain a near optimum configuration during the preliminary
design process before proceeding with refinements using streamline and CFD.
The streamline curvature code is used to refine the preliminary design based the previous results. The compressor geometry is calculated and analyzed again. Several parameters must be observed to obtain an efficient turbomachinery.
The final geometry is obtained transferring the results of the streamline curvature code
to a CAD software to draw the 3-D compressor view. To simulate the flow a mesh must
be generated and use as input of the CFD solver.
CHAPTER 6. RESULTS
6.2.1
172
The compressor to be designed must deliver a 5:1 pressure ratio in 5 stages and 8.3
kg/s mass-flow rate. Taking into account the constraints generated by the gas generator,
the following parameters were also fixed
Inlet stagnation pressure: 101, 325 Pa,
Inlet stagnation temperature: 288.15 K,
Inlet Mach number: 0.50,
Outlet Mach number: 0.26,
Pressure ratio: 5.0,
Isentropic efficiency: 82%,
Number of stages: 5,
Rotational speed: 25, 650 rpm,
Inlet hub-tip-ratio: 0.40,
Axial channel: Constant Outer Diameter (COD),
Mass-flow: 8, 3 kg/s.
6.2.2
Preliminary Design
6.2.2.1
For the preliminary design of the axial compressor, a computational code named
AFCC, developed by Tomita [1], was used to calculate the main operation character-
CHAPTER 6. RESULTS
173
istics of the machine based on the meanline characteristics. The results of this calculation
is a compressor map with VIGV and BOV. This initial synthesized compressor map was
used in an engine deck to identify possible off-design operation problems.
Loss models were used to account for losses due to the viscous effects and the complexity of the channel flow. The stall incidence was calculated using curves similar to the
ones presented in Figure 6.44.
The incidence angle is defined by i = 1 1 , where 1 is the relative inlet flow angle
and 1 is the inlet blade angle. The deviation angle is defined by = 2 2 , where 2
is the relative outlet flow angle and 2 is the outlet blade angle.
The stall incidence angle is a function of several parameters, like the ones indicated
below:
s/c; ), (6.1)where, 2opt is the optimum relative outlet flow angle; Clvopt is the optimum
CHAPTER 6. RESULTS
174
lift coefficient; is the inlet and outlet blade angle; iopt is the optimum incidence angle;
is the camber; opt is the optimum deviation angle; s/c is the blade space-chord ratio and
is the blade stagger angle.
Minimum loss condition was used for the design. Acceptable incidence and deviation
angles and losses were based on NASA SP-36 [130]. For incidence angle:
(6.2)
where ml stands for minimum loss condition; ki is a correction factor; (i0 )10 is the zerocamber incidence angle; n is the slope factor; (iD i2D ) is a correction factor to account
for two dimensional effects; t/c is the thickness-chord ratio; is the solidity;
The minimum loss incidence angles were corrected to account for off-design operation,
that is, the design point incidence angle was set lower than at design. For modern compressors, the flow velocity exceeds the limits of validity of the relation given by Equation
6.2. For sonic and supersonic flows, cascade experiments have confirmed that there is only
one incidence angle at which minimum losses are achieved and this incidence is named
unique incidence. For these high-speed flows, the minimum loss is attained when the flow
is parallel to the blade suction surface. Reference [2] presents details of this correction.
For deviation angle:
d
di
+ (D 2D ),
(6.3)
2D
where k is the thickness correction factor for zero camber deviation angle; m is a slope
factor for minimum loss deviation; the exponent b is a function of inlet air-angle; (0 )10 is
CHAPTER 6. RESULTS
175
the zero-camber deviation angle at reference minimum loss incidence angle deduced from
low speed cascade data for 10 percent of thickness and (d/di)2D is the slope at reference
incidence.
The model is a function of several parameters as indicated bellow:
(6.4)
where Cd is the drag coefficient; Mn0 is the inlet absolute Mach number at the blade; Mni
is the inlet relative Mach number at the blade; Mncopt is the optimum Mach number at
the blade inlet; 1 and 2 are the inlet and outlet relative flow angles, respectively; i is
the incidence angle and is the deflection. Details can be found in [1].
The sources of losses are, in general, viscous effects in boundary-layers, viscous effects
in mixing processes, shock-waves and heat transfer across temperature differences. The
mechanisms of loss sources as profile loss, end-wall loss and leakage loss, continues
to be widely used although it is now clearly recognized that its mechanisms are seldom
really independent.
The meanline calculation based on stage-stacking method allow an accurate investigation of the incidence and deviation angles for different rotational speeds and variable
geometry configurations. The stall and choke points can also be determined. The tendency at part-speed for the front stages is to move the operational point into stall and the
rear stage to move it towards choke. Hence, in order to eliminate some design difficulties
caused by this, the front stages are matched closer to choke and the rear stages closer to
stall, at the design-point operation.
The area ratio at annulus of a multistage compressor is based on the compressibility
CHAPTER 6. RESULTS
176
of the fluid and the intended stagewise work distribution. At off-design operation, the
stages operate at lower pressure ratios, changing the overall matching of the machine.
The rear stages are unloaded, the front stages are higher loaded and their operating point
shifts towards the surge stability limit. Two features are common to compensate the high
loading: change the stagger angle of the blade and increasing the mass-flow in the front
by taking bleed. The number of variable stator vanes in the compressor is an early project
decision as any late correction is related to high additional cost due to the complexity
of installation. Minimum tolerances have to be introduced or special sealing and bearing
bushes have to be used. All of these tolerances require careful design consideration, and
a sound balance between reduction in tolerance and cost reduction is to be created.
The other possibility for ensuring stable flow in the front stages at part-load speed is
the downstream extraction of bleed. This increases the mass-flow in the upstream annulus,
thus reducing the blade row loading. The energy transforms into the bleed mass-flow is
lost for the performance of the compressor. In Figure 6.54 it is possible to observe the
importance and need of the VIGV and BOV installations for the engine part-load speeds.
These accessories have a direct influence on the gas turbine efficiency.
Figure 6.45 shows the compressor map and its efficiency calculated using meanline
technique. In this figure is possible to observe that for part-load speed as 65%N the
operation range and its efficiency is very low (around 60%).
To improve the compressor efficiency, first the BOV was used. Figure 6.46 shows the
percentage of air bleed used in this simulation. Figure ?? shows the four compressor maps
for each bleed schedule presented in Figure 6.46. Figure 6.48 shows the improvement in
the compressor efficiency at off-design operations.
The installation of VIGV in this compressor was considered in the performance cal-
CHAPTER 6. RESULTS
177
CHAPTER 6. RESULTS
178
CHAPTER 6. RESULTS
179
CHAPTER 6. RESULTS
180
CHAPTER 6. RESULTS
6.2.2.2
181
The adequacy of the designed compressor for the use in a gas generator was studied
using previously designed gas turbine simulator [4]. The result of this study was the compressor performance at part-load and possible adjustments to the compressor preliminary
design.
Calculated the compressor, the scheme of the gas turbine engine is assembly to prepare
the calculations of the engine performance. Aiming at foreseeing off-design operation,
VIGV and BOV were incorporated to the compressor to use in a single-shaft free power
turbine turboshaft. The running line was calculated using an engine simulation deck
named GTAnalysis, developed by the Gas Turbine Group [4]. Figure 6.53 shows the
engine sketch used for this calculation. The engine design point characteristics are:
CHAPTER 6. RESULTS
182
The main engine functional blocks used in this analysis were: ambient, intake, compressor, mass flow splitter, combustion chamber, turbine, mass-flow mixer, power turbine,
and exhaust pipe. At part-load gas generator speed N1 is lowered from 100% down to the
point at which the surge margin vanishes. The power turbine is directly coupled to the
generator, so that its speed is fixed at 60Hz or 3600 rpm. For each chosen off-design point
all thermodynamic parameters were calculated, from which appropriate data should be
selected.
The engine is required to operate off-design due to load variation. Performance deteriorates because at off-design the components operate at regions of lower efficiencies,
caused by bad component matching. The bad-matching results from the passage areas,
calculated at design point conditions, not being adequate to accommodate the flow at
those operating conditions.
CHAPTER 6. RESULTS
183
Compressor surge margin was set to 20%, according to the suggestions of Walsh and
Fletcher [131], at design and was monitored during the calculations at all part-load conditions.
Figure 6.54 shows the compressor map without variable inlet guide vane (IGV) and
bleed (black lines) and the compressor map with variable IGV and bleed (magenta lines),
this being the map corresponding to the best part-load efficiency and surge margin. In this
figure are shown the design-point, running line, surge-line and surge margin. In Figure
6.54, it can be seen that at part-load operation the running line, without VIGV and
compressor bleed, intercepts the surge line at approximately 88% (0.7202 MW). When
variable VIGV and compressor bleed are used this value can be decreased down to 75%
(0.3218 MW). It is therefore desirable to use VIGV and bleed for this engine configuration
and application. Figure 6.55 shows the isentropic efficiency of the two compressors shown
in Figure 6.54. It can be seen that VIGV and compressor bleed keep the isentropic
efficiency at highest values at part-load, nearly at the same levels as at design-point.
CHAPTER 6. RESULTS
184
6.2.3
To improve the meanline design, a previously developed streamline curvature code was
used [2]. The result of this calculations are the whole compressor geometry and all flow
properties in the compressor.
Streamline curvature, is a technique well suited for the design of axial-flow compressors, mainly because it allows quick access to vital flow properties at the blades edges,
from which actions may be taken to improve its performance. Such technique has been
successfully used for analysis, from which more realistic compressor maps can be synthesized early in the design process, even with stators variable geometry. In conjunction
with an engine simulation program, it is possible, well in advance, to evaluate the engine
performance and take actions aiming at better engine tuning. A computer program for
performance analysis was developed at ITA and has been used for many years. This
program is still being in development at ITA, with addition of improved loss models, algorithms for faster convergence and the ability to synthesize a full compressor map. The
use of this compressor code has been granted to other universities, with applications re-
CHAPTER 6. RESULTS
185
cently reported in the literature that testify its great usefulness both for compressor maps
synthesis and engine performance simulation. Its application starts with the compressor
detailed geometry already known. The design of a new compressor can be based also on
streamline curvature techniques.
Starting from the compressor requirements of mass-flow, pressure ratio and efficiency,
a design procedure is set up to provide information of axial-flow channel geometry. Major
required parameters are the inlet hub-tip ratio, the exchange of temperature (or pressure)
increase among the stages, maximum blade tip speed, axial channel type (fixed outer,
fixed inner, fixed mean or variable mean diameter), inlet and outlet Mach numbers, flow
angles at the stators trailing edges, axial velocities (or Mach numbers) at the rotors inlet,
tip clearances, axial chord and spacing ratios, space-chord ratios and maximum thickness
ratios. Variation of these parameters, from hub-to-tip, can also be set. The user can
change these design parameters interactively to force the axial channel to have adequate
shape so that the attainment to the targeted performance is feasible. Having obtained
an acceptable axial channel, a distribution tentative of any number of streamlines is
made, so that the geometrical parameters can be calculated at nodes located at the blade
edges, on the streamlines. Initial location of the streamlines is based on equal flow in the
streamtubes, but it could be any other criteria, since the streamlines are automatically
and iteratively repositioned during the calculations.
The calculations are standard for each row, starting with the integration of the complete radial equilibrium equation at the blade leading edge based on assumed axial velocity
at the hub. The integration of the radial equilibrium equation results in a spanwise axial
velocity distribution, from which the total mass flow at the blade inlet can be calculated.
The exact velocity distribution is obtained by forcing the calculated mass flow to be equal
CHAPTER 6. RESULTS
186
CHAPTER 6. RESULTS
187
~
~
~
W
W W
W
DV~
= Vr
+
+ Vz
,
Dt
r
r
z
(6.5)
CHAPTER 6. RESULTS
188
and
V~ = Vr~er + Vw~ew + Vz~ez ,
(6.6)
(6.7)
~ = r~e .
U
(6.8)
~
~
~
W W
W
W
1~
~ + F~ .
+
+ Vz
+
~
~ ~r + 2~ W
= Vr
P
r
r
z
(6.9)
Therefore,
Details of the streamline curvature method can be found in [2]. Further development
of the previous equation gives the following equation, which is used for the calculation of
the meridional velocity:
Vm
Vm
Vm2
tg( + )
1 S
2 1
=
{sec( + ) +
+
(1 + Mm
) sin()},
m
1 Mm
s
Rc
R m
r
(6.10)
where m stands for meridional, is the streamline slope, is the blade sweep, Rc is the
radius of curvature and S is the distance along blade edges.
The boundary-layer, which is not included in the previous equation, is estimated using
as reference model empirical correlations from [132, 133, 134], that means blockage factor. The literature reports that the application of conventional boundary-layer theory to
estimate the displacement and momentum thickness is generally inadequate since inside
an axial compressor the flow is constantly: changing direction, skewed, energized and deenergized, as it progresses through the rows. The model presented by Jansen [132] uses
CHAPTER 6. RESULTS
189
the meridional velocity near the wall region to account for the effect of the boundary-layer
in restricting the flow through the compressor.
The blockage factor is defined as the decreasing in the compressor annulus height
equivalent to the boundary-layer displacement thickness. Hence, if it is assumed that the
end-wall boundary-layer can be treated as 2 D and turbulent, and that the meridional
velocity is the dominant characteristic, it can be shown that
"R z
(z) = (z0 ) + kbl
V 4 (z)dz
z0 m
Vm3.2
#0.8
.
(6.11)
(z0 ) =
,
H
(6.12)
where H is the shape factor, and initially can be assumed equal to 1.4, and can be evaluated
from
H(z) = 1.5 + 30
d
,
dz
(6.13)
(z) = (z)H(z).
(6.14)
Calculating the equations 6.11, 6.12, 6.13 and 6.14 to the hub and to the casing
(tip) walls, in which Vm (z) is the local meridional velocity, it is possible to obtain H
(z)
and T (z), that are the boundary-layer displacement thickness at hub and at casing,
CHAPTER 6. RESULTS
190
respectively, along the compressor annulus. Hence, the blockage factor is defined as the
ratio of the free-stream area and metal area. Then
kB =
2
(RT T )2 + (RH H
)
,
2
2
RT RH
(6.15)
where RT and RH are the casing (tip) and hub radii. Knowing the kB value, only after
the meridional velocity is known throughout the compressor, the process of calculating
the boundary-layer blockage is iterative. At this point, it is possible to define the hub and
casing (tip) blockage factor due to boundary-layer as
kBH =
2
2
) RH
(RH H
,
2
RT2 RH
(6.16)
kBT =
RT2 (RT T )2
,
2
RT2 RH
(6.17)
then
kB = 1 (kBH + kBT ).
(6.18)
There are different types of losses in a compressor, but a good assessment depends on
the knowledge of the mechanisms through which these losses act. The comprehension of
the interaction of all losses involved inside of the turbomachinery is very complex. Hence,
CHAPTER 6. RESULTS
191
in a simplified approach, the losses are considered independently generated. Most of the
results from test facilities are published and available to the public but key information is
retained as proprietary. Fortunately, some researchers have incorporated their expertise
into correlations of parameters which describe the flow along the compressor annulus.
There are several averages of tests results or their statistical curve fits as presented in
NASA SP-36 [130]. The problem is that there are no expected general formulation to
represent each individual compressor. Hence, the empirical correlation cannot give good
results every time that they are applied because a general model for all compressors is
unrealistic. The adjustments of each case is carried out using the designer expertise.
Profile loss is usually considered to be loss generator due to the blade boundary-layers
well away from the end-walls. It is possible to assume that the flow is 2 D so, the
loss may be based on 2 D cascade tests or in boundary-layer calculations. Friction
and secondary losses also contribute to profile losses. The extra loss arising at blade
trailing-edge is included as profile loss. For high Reynolds numbers, it is necessary to
take account of the additional losses associated with low Reynolds numbers. End-wall
loss or secondary loss is from the secondary flows generated when the compressor annulus
boundary-layers pass through a blade row. However, the loss does not arise directly from
the secondary flow, but it is due to a combination of many factors. Tip leakage loss arises
from the leakage of flow over the rotor tips (and sometimes of stator hub depending on the
design configuration). The detailed loss mechanisms clearly depend on whether the blades
are shrouded or unshrouded. For unshrouded compressor blades, the interaction between
leakage and end wall losses may be strong. Hence, some methods do not distinguish these
losses separately.
The losses are accounted for using loss models, which take account of the following
CHAPTER 6. RESULTS
192
losses:
Profile loss: based on the models of NASA SP-36 [130], Swan [135], Monsarrat
[136], Jansen and Moffat [26], Davis and Millar [137] and one modification of the
last model mentioned is implemented [2],
Shock losses: based on the NACA model of reference [138],
Secondary losses: based on the models developed by Griepentrog and Howell [139,
140],
Friction losses: based on duct loss,
Reynolds number effect: takes account of low Reynolds numbers effect, using the
models of references [130, 141].
The user may arbitrarily combine these models and also used their own models. In
this work, profile loss was evaluated according to reference [136], secondary loss according
to [139] and the Reynolds number effect according to [141].
Tables 6.1, 6.2, 6.3 and 6.4 present blade geometrical characteristics, from hub-totip, calculated by the streamline curvature program, using five streamlines distributed
along the blade spanwise, calculated considering the incidence and deviation angles and
accounting the flow losses using the models aforementioned. With the inlet and outlet
blade angles it is possible to determine the blade stagger and camber angles. To deliver the
flow with low swirl angle at the compressor outlet or combustion chamber inlet, the last
stator (S5) was designed with low outlet blade angle as presented in Table 6.4. Generally,
a high performance compressor has high deflection angles in the first stages to transfer
a high energy from the rotor to the air. Hence, the loading factor is greater than in
CHAPTER 6. RESULTS
193
other stages as presented in Table 6.7. The outlet flow angle can be obtained with the
difference between the deviation angle and blade outlet angle. Table 6.6 presents the
number of blades for each row.
TABLE 6.1 Distribution of rotor blade
Fraction of radial span (%) R1
0
-59.3
25
-68.9
-71.7
50
75
-73.5
100
-75.0
rotor height
R5
-73,8
-75,9
-76,7
-77,3
-77,9
rotor height
R5
-53.6
-60.3
-62.3
-63.7
-64.8
stator height
S5
58.9
55.8
54.7
54.0
53.3
stator height
S5
1.7
1.4
1.3
1.2
1.2
CHAPTER 6. RESULTS
TABLE
span (%)
0
25
50
75
100
194
6.5 Distribution of the blade space-chord ratio along the blade height
R1
S1
R2
S2
R3
S3
R4
S4
R5
S5
0.493 0.553 0.529 0.571 0.598 0.597 0.564 0.563 0.536 0.537
0.732 0.755 0.670 0.699 0.701 0.692 0.634 0.632 0.589 0.596
0.828 0.835 0.725 0.749 0.742 0.728 0.662 0.662 0.612 0.618
0.910 0.903 0.775 0.791 0.779 0.757 0.687 0.681 0.630 0.632
0.987 0.967 0.822 0.830 0.814 0.784 0.711 0.696 0.648 0.643
distribution
R3
R4
0.546 0.454
0.398 0.359
0.353 0.327
0.323 0.305
0.299 0.287
S5
89
R5
0.363
0.300
0.278
0.262
0.249
Tables 6.8, 6.9, 6.10 and 6.11 present the blade incidence and deviation angles from
hub-to-tip calculated by streamline curvature program using 5 streamlines distributed
along the blade spanwise. Table 6.8 shows the incidence angle at the rotor leading and
trailing edges. Note that the rotor five (R5) needs some improvement near the hub region
due to the high incidence angle. High incidence angle causes stall problems. For the
stators, the incidence angle has lower values compared with the rotors (Table 6.9). Note
that the incidence angle at the stator five (S5) is small, but no choke problems was detected
during the compressor design calculation.
TABLE 6.8 Distribution of rotor incidence
Fraction of radial span (%) R1
0
6.88
25
6.66
50
6.64
75
6.63
6.62
100
CHAPTER 6. RESULTS
195
angle
R3
6.60
5.11
4.77
4.61
3.96
angle
S3
8.74
9.21
9.31
9.38
9.44
The de Haller number, defined as the ratio of the relative outlet and inlet flow velocities,
is a parameter that quantifies the deceleration of the flow along the blade-to-blade channel
[1]. The average value of the de Haller number is 0.72 at design-point. Values lower than
0.72 needs some attention to avoid problems with high boundary-layer growth. The
relative or absolute velocity has greater values near tip region. Hence, the de Haller
number, generally is greater on this region. The de Haller number along the rotor and
stator blade spanwise are presented in Tables 6.12 and 6.13
TABLE 6.12 Rotor
Fraction of radial span (%)
0
25
50
75
100
R5
0.637
0.691
0.712
0.727
0.739
CHAPTER 6. RESULTS
TABLE 6.13 Stator
Fraction of radial span (%)
0
25
50
75
100
196
de Haller number distribution
S1
S2
S3
S4
0.737 0.557 0.558 0.570
0.861 0.632 0.610 0.609
0.897 0.659 0.630 0.624
0.919 0.679 0.644 0.636
0.935 0.695 0.656 0.646
S5
0.569
0.605
0.619
0.630
0.639
High Mach number means high loading factor. Table 6.7 shows that the first rotor
blade (R1) has a greater loading factor and the last rotor (R5) has a lower loading factor
due to the energy transfer for each blade row and the design criteria given by the inlet
absolute flow angle for each blade row. High loading factor at the compressor inlet can be
causes serious structural problems for the blade. High Mach number at the compressor
outlet can causes some difficulties to decelerate the flow at the combustion chamber inlet.
Hence, in this work, the inlet and outlet Mach number used as input parameter to design
the compressor are 0.5 and 0.25, respectively. The distribution of the Mach number
calculated through the compressor blade rows is presented in Table 6.14.
TABLE 6.14 Axial Mach number for each blade row
R1
S1
R2
S2
R3
S3
R4
S4
Mach number 0.494 0.384 0.439 0.367 0.382 0.319 0.334 0.282
6.2.4
R5
0.294
S5
0.251
With the results from the streamline curvature program, a solid model of the compressor channel defined by the blades was generated using a CAD software.
The meshes were defined based on such geometry using the version 11 of the ICEMCFD. An interpreter for the ICEM-CFD unstructured hexahedral elements output files
was developed to integrate the calculated mesh to the CFD solver.
Figures 6.56 and 6.57 show a 3 D compressor picture in two different views.
CHAPTER 6. RESULTS
197
CHAPTER 6. RESULTS
198
FIGURE 6.57 3-D view of the 5-stage axial-flow compressor blade profiles
CHAPTER 6. RESULTS
199
FIGURE 6.60 General view of the O-grid generated to the 5-stage axial-flow compressor
CHAPTER 6. RESULTS
200
Figure 6.62 shows the mesh structure of the first-stage of the compressor. In this figure
it is possible to observe the rotor curved surface. All other stages follow the same mesh
fashion.
FIGURE 6.62 Computational domain and mesh structure on the hub of the first-stage
Figure 6.63 shows a detail of the O-grid generated close to the stator trailing edge and
rotor leading edge.
FIGURE 6.63 Detail of the O-grid mesh type around the blades
The elements and their neighbors are addressed via connectivity tables. For an hexahe-
CHAPTER 6. RESULTS
201
dral element there are six neighbors, including the ghost elements. These ghost elements
are used to set the boundary conditions. For this case, the values of the properties were
taken from the streamline curvature program calculations. The properties were specified
at the inlet (stagnation pressure, stagnation temperature and flow angles) and at the outlet (static pressure). The flow was calculated only in one channel, since it is considered
periodic. Avoidance of reflection was assured by the use of non-reflecting inlet and outlet
boundary conditions.
To achieve an acceptable solution, it is essential to initialize the variables accurately.
An accurate start may be based on the streamline program results, what was done in this
research. The results of computational codes for compressors and turbines, developed
at Center for Reference on Gas Turbine at ITA were used to produce the appropriate
initialization. Besides speeding-up convergence, an accurate initialization reduces machine
time, especially for the case of finer meshes, as the one used in the study of the 5-stage
axial compressor. Care has also been taken to guarantee a well-posed problem and hence
numerical stability. The problem is that for high pressure compressor, when the blade
is operating near stall condition, the calculation may become unstable due to numerical
oscillations due to the strong unsteady flow behavior. Hence, the calculated flow separates
or generates a high loss, possibly during the pseudo-transient part of the calculation. The
resulting blockage reduces the mass-flow and increases the flow incidence at the blade
which makes the separation or the loss larger. Thus, the calculation may fail as a result
of the transient behavior by initial condition due to the unsteady operation point.
CHAPTER 6. RESULTS
202
The convergence of the compressor outlet mass-flow and the compressor efficiency are
presented in Figures 6.64 and 6.65, respectively.
10
MassFlow (kg/s)
Outlet MassFlow
5
0
2000
4000
6000
8000
10000
iteration
12000
14000
16000
18000
20000
82
81
efficiency
80
79
78
efficiency
77
0
2000
4000
6000
8000
10000
iteration
12000
14000
16000
18000
20000
FIGURE 6.65 5-stage axial-flow compressor efficiency variation during the numerical
iteration
CHAPTER 6. RESULTS
203
Figure 6.66 shows the static pressure increase along the compressor stages.
A cut of a velocity vectors distribution at the compressor outlet can be seen in the
Figure 6.67.
The mixing-plane velocity distribution at the rotor outlet to the stator inlet, in the
third stage of the axial compressor is presented in Figure 6.68. The velocity vectors at rotor
outlet are relative velocities. This figure shows the plane when the frame of reference was
changed from a rotating frame of reference to non-rotating frame of reference. Periodicity
between suction and pressure sides can be observed too. Figure 6.69 shows the velocity
vectors across the compressor third-stage.
CHAPTER 6. RESULTS
204
FIGURE 6.69 Detail of the mixing-plane velocity distribution at the compressor thirdstage
CHAPTER 6. RESULTS
205
The total pressure distribution is presented in the Figure 6.70. The total pressure in
the rotors region is represented by the relative total pressure.
FIGURE 6.70 Total pressure distribution along the 5-stage axial-flow compressor
General results for the SLC and CFD calculations are presented in Table 6.15. CFD
results are arithmetic means of the properties at each face at the compressor outlet computational plane.
TABLE 6.15 Comparison between streamline curvature and CFD results
Technique
Streamline curvature program CFD
Mass-flow
8.3
7.9
Pressure ratio
5.0
4.9
82.0
79.4
Efficiency (%)
The efficiency was calculated by the ratio of the ideal total temperature increase across
the compressor and the actual total temperature increase. Some differences can be observed between streamline curvature and CFD results. The calculation of the efficiency
and output power by CFD is an issue under study. Maybe the average values at compressor outlet are not a better option to calculate these variables. The calculation of these
variables using CFD is proprietary information on the turbomachinery community. Each
compressor design needs different calibration of the loss models used in the streamline
curvature program. There are no general loss models to apply in the streamline curvature
methodology. Each compressor has its particular calibration and flow behavior. Inter-
CHAPTER 6. RESULTS
206
action between the streamline curvature program and the CFD calculations are being
carried out aiming at the flow analysis at spotted parts where undesirable large gradients
are present to favor flow separation, what it is not the aim of this work.
Interaction between the streamline curvature and the CFD results have been carried
out aiming at the flow analysis at spotted parts where undesirable large gradients are
present to favor flow separation, what is not aim of this work.
208
7.1
209
The CFD platform developed for this work is adequate for future research and teaching
activities. Although the computational capability existing now is not enough for design
using CFD exclusively, the interaction with a streamline curvature program can integrate
the use of CFD during the design part of the project. In this case, the CFD is used to check
whether the streamline curvature design can deliver or not the specified performance.
Therefore, a very fertile field is available for future research on methods of discretization, adequacy of turbulence models for flow calculation, as well as better combination
of technological parameters aiming at a turbomachine to be optimized for a specified
application.
Since the platform was implemented having in mind the modular concept, implementation of many different procedures can be made without much difficulty.
At the time of the writing of this work, a parallel version of the CFD program is
running, as part of a post-doctoral research work carried out at the Center for Reference
on Gas Turbine. This extends the fields of research to the vast area of parallel processing
as well.
8 Future Implementations
The objective of this chapter is to describe some important future works to improve
the present CFD solver robustness. As commented in the last chapters, turbomachinery
flow simulations is very complex and needs some special treatments. The developement of
CFD solver was only possible because the Gas Turbine Group at ITA has turbomachinery
design tools developed as meanline and streamline curvature programs. The process to
obtain the final turbomachinery geometry needs a long time of design and performance
analysis.
The numerical implementation process in a computational code capable to calculate
the full Navier-Stokes equations in a 3 D computational domain and to become it
possible to simulate the rotor and stator blade rows simultaneously required a long time
of programming. Furthermore, many others details before discussed become this work
very extensive. Hence, during the development of this research, the author concluded
that some subjects mentioned in this chapter are strongly recommended to improve the
numerical simulation of the computational code. Not only for the solver, but to improve
the pre- and post-processing.
The CFD platform is not user-friendly, since it has been developed just for the author
use. It can be associated to an user interface that makes its use more friendly.
211
The modular environment used for the program development makes it simple to add
extra modules incorporating other discretization methods, turbulence models, that widen
the use to other applications than turbomachinery.
Several possible continuation work have already been mentioned in the past chapters.
Below follow suggestions that would facilitate the platform development, as MSc or PhD
research programs.
8.1
Pre-Processing
The pre-processing in a turbomachine CFD simulation has many details due to the
complexity of a compressor or a turbine. Blade geometry visualization makes the work
very hard, mainly for high performance multistage axial flow compressors where the
leading and trailing edges radii are too small. The idea of this section is to discuss some
important features in the sense of coupling the streamline curvature program with CFD
solver. Some suggestions are:
212
2. It is of utmost interest to make automatic the process of generating the compressor blade passages in a CAD environment directly from the streamline curvature
program, since this program is the most appropriate for geometry calculation. In
the sequence, the extraction of the geometry data for the grid generation routine.
The authors experience points that the process of generating a CAD solid and the
corresponding grid is too expensive in terms of time consumed until an appropriate
grid is obtained.
3. Although grid generation may not be interpreted as a CFD work, it is the authors
experience that grid generation must be one of the first works to be developed. Having in mind the very specific shape of the blade passages, a grid-generator software
could be developed such that the problem of making available an adequate grid
could be almost transparent for the researcher. In this work the hexahedral grid
was implemented, leaving for future work other grid types to be implemented.
8.2
Solver
The CFD solver developed based on academic interest to become possible the study
of new numerical methods applied to mass, momentum and energy equations, turbomachinery loss mechanisms, new blade geometries, inlet and outlet ducts, inlet flow with
distortions, rotating stall and surge phenomenon, wakes, tip-clearance and its influence
on the machine performance, to create new techniques to improve the tip leakage, passive
endwall treatments for enhancing stability, new design and treatment of casing, unsteady
effects among others. About the numerical improvement, the first-step towards become
the CFD solver faster and possible to run in various processors simultaneously was started
213
with the work involving parallel processing. The first version of the CFD solver code with
parallel processing was verified and publicated [142] in a ASME TurboExpo Conference.
Other suggestions from the author are:
1. The current time integration scheme is explicit. The use of an implicit version would
improve the convergence as far as stability and CPU time are concerned.
2. The implementation of higher-order spatial and time discretization methods will
make the Platform best suited for transient analysis. Methods for convergence
acceleration can be also implemented, aiming better computational performance.
3. In this work grid refinement is not automatic, so grid refinement using all the available criteria can be the focus of further research.
4. The current platform cannot do the calculations of reacting flows. Therefore, the
inclusion of the equations concerning the combustion processes will make it useful
for calculations in combustion chambers as well.
5. There are other flows that can be calculated using the developed software and,
therefore, appropriate turbulence models could be implemented accordingly.
6. Implement a convergence acceleration techniques.
7. Implement other mixing-plane formulations including formulation based on unsteady
flow. For the transient performance study it is required the use of appropriate
techniques, one of which is the interpolation between grid points.
8. Create a powerful framework to simulate the flowfield in all components (compressor,
combustion chamber and turbine) of gas turbine simultaneously.
9. Develop a parallel version of the computational platform.
8.3
214
Post-Processing
Bibliography
1 TOMITA, J. T. Numerical Simulation of Axial Flow Compressors. 2003. 151 p.
Thesis (Master of Science-MSc) - Instituto Tecnologico de Aeronautica - ITA, Sao Jose
dos Campos-SP-Brazil, 2003. 33, 44, 170, 172, 175, 195
2 BARBOSA, J. R. A Streamline Curvature Computer Programme for
Performance Prediction of Axial Flow Compressors. 1987. Thesis (Doctor of
Philosophy-PhD) - Cranfield Institute of Technology, 1987. 33, 174, 184, 188, 192
3 JESUS, G. S. de. Projeto e An
alise de Desempenho de Turbinas Axiais de
V
arios Est
agios com Geometria Vari
avel. 2003. 111 p. Thesis (Master of
Science-MSc) - Instituto Tecnologico de Aeronautica - ITA, Sao Jose dos
Campos-SP-Brazil, 2003. 33
4 BRINGHENTI, C. Variable Geometry Gas Turbine Performance Analysis.
2003. Thesis (Doctor of Science-DSc) - Instituto Tecnologico de Aeronautica - ITA, Sao
Jose dos Campos-SP-Brazil, 2003. 34, 170, 181
5 SILVA, F. J. dos S. Simulac
ao de Desempenho de Turbinas a G
as em Regime
Transit
orio. 2006. 154 p. Thesis (Master of Science-MSc) - Instituto Tecnologico de
Aeronautica - ITA, Sao Jose dos Campos-SP-Brazil, 2006. 34
6 REED, J.; AFJCH, A. Computational Simulation of Gas Turbines - Part I:
Foundations of Component-Based Models: INTERNATIONAL GAS TURBINE &
AEROENGINE CONGRESS & EXHIBITION, 1999. 94-GT-346. 34
7 REED, J.; AFJCH, A. Computational Simulation of Gas Turbines - Part II:
Extensible Domain Framework: INTERNATIONAL GAS TURBINE & AEROENGINE
CONGRESS & EXHIBITION, 1999. 94-GT-347. 34
8 LAKSHMINARAYANA, B. Fluid Dynamics and Heat Transfer of
Turbomachinery. John Wiley & Sons, 1996. 35, 43, 63, 85, 228
9 TREMAYNE, D. The Science of Formula 1 Design. Haynes Publishing, 2004. 35
10 MAKINO, Y.; IWAMIYA, T.; LEI, Z. Fuselage Shape Optimization of a Wing-Body
Configuration with Nacelles. Journal of Aircraft, 2003. v. 40, n. 2, March-April 2003.
35
11 WEATHERILL, N. P.; MORGAN, K. A Course Under the PROPESA 1995/96
Program: Finite-Element Methods in Computational Fluid Dynamics. University of
Wales - Department of Civil Engineering, 1995. 35, 36
BIBLIOGRAPHY
216
12 SCHLUTER, J.; APTE, S.; KATIZIN, G.; HEINZ, P.; WEIDE, v. d.; ALONSO, J.
Unsteady CFD Simulation of an Entire Gas Turbine High-Spool. ASME Turbo Expo
2006. Proceedings of ASME. GT2006-90090. 36
13 GOPINATH, A. K.; WEIDE, V. D.; KALITZIN, G.; ALONSO, J. J.; JAMESON,
J. J.; EKICI, K.; HALL, K. C. Three-Dimensional Unsteady Multi-Stage
Turbomachinery Simulations Using the Harmonic Balance Technique AIAA report, Stanford University, Duke University. 36
14 TOMITA, J.; BRINGHENTI, C.; BARBOSA, J.; JESUS, A. Nacelle Design for
Mixed Turbofan Engines. ASME Turbo Expo 2006. ASME, 2006. GT2006-91212. 36
15 WING, D. Afterbody/Nozzle Pressure Distributions of a Twin-Tail
Twin-Engine Fighter with Axisymmetric Nozzles at Mach Numbers From
0.6 to 1.2, NASA, 1995. NASA-TP-3509. 36
16 COMPTON, W. B. Comparison of Turbulence Models for Nozzle-Afterbody
Flows with Propulsive Jets, 1996. Washington, DC: NASA. 1996. NASA TP-3592.
36
17 WILHELM, R. Inverse Design Method for Design Isolated and Wing-Mounted
Engine Nacelles. Journal of Aircraft, 2002. v. 39, n. 6, Nov.-Dec. 2002. 36
18 FUJINO, M.; KAWAMURA, Y. Wave-Drag Characteristics of an Over-the-Wing
Nacelle Business-Jet Configuration. Journal of Aircraft, 2003. v. 40, n. 6, Nov.-Dec.
2003. 36
19 LAPWORTH, L.; SHAHPAR, S. Design of Gas Turbine Engines Using CFD.
EUROPEAN CONGRESS ON COMPUTATIONAL METHODS IN APPLIED
SCIENCES AND ENGINEERING-ECOMAS 2004, 2004. 37
20 KAU, H. The Multidisciplinary Design Process. In: RTO AVT LECTURE SERIES.
Integrated Multidisciplinary Design of High Pressure Multistage
Compression Systems, 1998. p. 116.(RTO Lecture series, 211) 38
21 XU, J.; GE, M.; ZHAO, X.; WANG, Z. Three-Dimensional Flow Theory of
Turbomachinery, Part 1: Basic Methodology. Propulsion and Power, v. 14, n. 6, p.
899906, nov 1998. 40
22 XU, J.; GE, M.; ZHAO, X.; WANG, Z. Three-Dimensional Flow Theory of
Turbomachinery, Part 1: Design and Analysis System. Propulsion and Power, v. 14,
n. 6, p. 907915, nov 1998. 40
23 KOCH, C.; SMITH, L. Loss Sources and Magnetudes in Axial-Flow Compressors.
Journal of Engineering for Power-Paper N 75-WA/GT-6, p. 114, jul. 1976. 40,
165
24 DENTON, J. Loss Mechanisms in Turbomachines. Journal of Turbomachinery,
v. 115, p. 621656, out. 1993. 40
25 CASEY, M. Computational Methods for Preliminary Design and Geometry
Definition in Turbomachinery. AGARD Lecture Series on Turbomachinery
Design Using CFD. AGARD, 1994. p. 122. 41, 63
BIBLIOGRAPHY
217
BIBLIOGRAPHY
218
40 YAO, J.; SHANHONG, J.; LIU, F.; JENNIONS, I. Computation of Steady Flow
in a Turbomachinery Stage by a Third-Order Accurate High-Resolution
Scheme - Copyright 1997, American Institute of Aeronautics and Astronautics, Inc. 43
41 YANG, J. Third Order Nonoscillatory Schemes for the Euler Equations. AIAA
Journal, v. 29, n. 10, p. 16111618, out. 1991. 43
42 LACOR, C.; SMIRNOV, S.; BAELMANS. A finite Volume Forrmulation of
Compact Central Schemes on Arbitrary Structured Grids. Journal of Computational
Physics, v. 198, p. 535-566, jan 2004. 43, 85
43 CHIMA, R.; LIOU, M. Comparison of the AUSM+ and H-CUSP Schemes for
Turbomachinery Applications. Cleveland, Ohio, 2003. Glenn Research Center,
Cleveland, Ohio, 2003. NASA/TM 2003-212457. 43, 72
44 FRINK, N. Recent Progress Toward a Three-Dimensional Unstructured
Navier-Stokes Flow Solver. AIAA, 1994. 32nd AEROSPACE SCIENCES MEETING.
Washington, DC: AIAA, 1994. AIAA Paper-94-0061, 1994. p. 1-20. 43
45 FRINK, N. Assessment of an Unstructured-Grid Method for Predicting 3-d
Turbulent Viscous Flows. 1996, Reno, Nevada. p. 1-11. AIAA-Paper-96-0292. 43
46 VEUILLOT, J.; CAMBIER, L. Computation Techniques for the Simulation of
Turbomachinery Compressible Flows: Techniques de Calcul pour la Simulation
DEcoulements Compressibles des Turbomachines. 29 Avenue de la Division Leclerc 92320 CHATILLON, 1991. (12th ICNMFD, Oxford (GB), july 9-13). 43
47 STEWART, M. E. Axisymmetric Aerodynamic Numerical Analysis of a
Turbofan Engine ASME Paper. ASME, 1988. ASME 95-GT-338. 43, 85
48 VAVRA, M. Aero-Thermodynmics and Flow in Turbomachines. John Wiley
& Sons, 1960. 43, 228
49 VKI Short Course on Flow in Turbines. Flow in turbines. von Karman Institute
for Fluid Dynamics, 1969. 43
50 DAWES, W. Development of a 3D Navier-Stokes Solver for Application to all Types
of Turbomachinery. ASME Paper. ASME, 1988. ASME 88-GT-70. 43, 87
51 CHEN, J.; GHOSH, A.; SREENIVAS, K.; WHITFIELD, D. Comparison of
Computations Using Navier-Stokes Equations in Rotating and Fixed Coordinates for
Flow Through Turbomachinery. AEROSPACE SCIENCES Meeting & Exhibit. AIAA,
1997. (AIAA-Paper-97-0878). 44
52 ERDOS, J. I.; ALZNER, E. Computation of Unsteady Transonic Flows
Through Rotating and Stationary Cascades. NASA Contractor Report 1997,
2900. 44, 106
53 CHIMA, R. Calculation of Multistage Turbomachinery Using Steady
Characteristic Boundary Conditions. Washington, DC: NASA, 1998. NASA
TM-1998-206613, 1998. 44, 96
BIBLIOGRAPHY
219
BIBLIOGRAPHY
220
BIBLIOGRAPHY
221
BIBLIOGRAPHY
222
100 SAAD, Y. Iterative Methods for Sparse Linear Systems. PWS Publishing
Company, 1996. 96
101 DUFF, I. S.; ERISMAN, A. M.; REID, J. Direct Method for Sparse Matrices.
Oxford, Press, 1986. 96
102 GILES, M. UNSFLO: A Numerical Method for Unsteady Inviscid Flow in
Turbomachinery - internal report, MIT. October 1988. 100, 106
103 GILES, M. B. Nonreflecting Boundary Conditions for Euler Equation Calculation.
AIAA Journal, v. 28, n. 12, p. 2050-2058, dez. 1990. 100
104 HOLMES, D. G. Inviscid 2D Solutions on Unstructured, Adaptive Grids. VKI-LS
1989-06. 100
105 JAMESON, A. Solution of the Euler Equations for Two-Dimensional Transonic
Flow by a Multigrid Method. 1983. Applied Mathemathics and Computation 13:327-355, NY. 102
106 LONG, L.; KHAN, M. M. S.; SHARP, H. T. Massively Parallel Three-Dimensional
Euler/Navier-Stokes Method. AIAA Journal, v. 29, n. 5, p. 657-666, maio 1991. 102
107 HUENE, M.; JUNG, A. R. Comparison of Different Acceleration Techniques and
Methods for Periodic Boundary Treatment in Unsteady Turbine Stage Flow
Simulations. Transactions of the ASME, v. 122, p. 234-246, apr 2000. 106
108 ADAMCZYK, J. J.; R., M.; M., C. A Model for Closing the Inviscid Form of the
Average-Passage Equation System. ASME, v. 108, p. 180-186, out. 1986. 106, 111
109 CHEN, J.; BRILEY, W. A Parallel Flow Solver for Unsteady Multiple Blade Row
Turbomachinery Simulations. In: ASME, 2001, New Orleans,Louisiana,USA. ASME
TURBO EXPO 2001. Proceedings of ASME. p. 1-10. 2001-GT-0348. 109
110 BELARDINI, E. State of the Art in Rotor-Stator Interaction for Turbines.
QNET-CFD, v. 2, n. 3, p. 22-26, dec 2003. 109
111 ADAMCZYK, J. J.; CELESTINA, M.; BEACH, T. A.; BARNETT, M. Simulation
of Viscous Flow Within a Multistage Turbine. ASME, v. 112, 1990. 111
112 GEROLYMOS, G.; HANISH, C. Multistage Three-Dimensional Navier-Stokes
Computation of Off-Design Operation of a Four-Stage Turbine. IMechE, v. 213, p.
243-261, 1999. 111
113 RHIE, C. M.; GLEIXNER, A. J.; SPEAR, D. A.; FISCHBERG, C. J.;
ZACHARIAS, R. M. Development and Application of a Multistage Navier-Stokes
Solver: Part i-Multistage Modelling Using Bodyforces and Deterministic Stresses.
Journal of Turbomachinery, v. 120, p. 205-214, apr 1998. 111
114 RHIE, C. M.; GLEIXNER, A. J.; SPEAR, D. A.; FISCHBERG, C. J.;
ZACHARIAS, R. M. Development and Application of a Multistage Navier-Stokes
Solver: Part ii-Application to a High-Pressure Compressor Design. Journal of
Turbomachinery, v. 120, p. 215-222, apr 1998. 111
BIBLIOGRAPHY
223
BIBLIOGRAPHY
224
A.1
Non-linear artificial dissipation models, generally has terms of second and fourth order.
The stencil of artificial dissipation calculation for each element i is presented in Figure
A.1 only in one direction. For others directions the scheme is straightforward.
In this work, two layers of ghost elements was created to compute the artificial dissipation terms in the boundary elements. At the boundaries, three different schemes was
implemented in GTTeamCFD code and it is presented bellow
Figure A.2 show that two internal elements were used to calculate the values of second
and fourth order terms by the matrix of conserved variables. Figure A.3 show that the
226
FIGURE A.2 First scheme to calculate the second and fourth order terms of artificial
dissipation
FIGURE A.3 Second scheme to calculate the second and fourth order terms of artificial
dissipation
ghost elements uses the same values of the matrix of conserved variables of the first
adjacent internal neighbor. Figure A.4 show that the second and fourth order terms are
calculated using the boundary condition values computed by the matrix of conserved
variables.
In general, the scheme presented in Figure A.4 is more attractive because the terms of
artificial dissipation equation are calculated with the actual condition of the problem. The
connectivity table supply all neighborhood for each element i, but some modifications in
the ghost elements and its adjacent boundary elements are necessary to implement each
scheme above presented.
227
FIGURE A.4 Third scheme to calculate the second and fourth order terms of artificial
dissipation
A.1
~
F~CO = 2~ W
(A.1)
~ ~
F~c = (
~r)
(A.2)
2N
60
(A.3)
229
z =
l0
a0
(A.4)
where N is the rotational speed, l0 is the characteristic length and a0 is the reference
speed of sound. Hence the sum of this forces is calculated by
z2 x + 2Wy z
F~CO + F~c =
z2 y 2Wx z
(A.5)
This shows that a rotation around the z-axis influences only x- and y-component of
the momentum equations. Hence, the definition of the internal energy presented in 2.2
has to be changed by
ei =
e 1
(W 2 + Wy2 + Wz2 z2 y 2 z2 x2 )
2 x
(A.6)
CLASSIFICAC
AO/TIPO
TD
5.
2.
DATA
3.
19 de janeiro de 2009
DOCUMENTO No
4.
No DE PAGINAS
CTA/ITA/TD-001/2009
229
TITULO E SUBTITULO:
Three-Dimensional Flow Calculations of Axial Compressors and Turbines Using CFD Techniques
6.
AUTOR(ES):
OES)/
OES):
INSTITUIC
AO(
ORG
AO(S)
INTERNO(S)/DIVISAO(
Instituto Tecnol
ogico de Aeron
autica - ITA
8.
Din
amica dos fluidos computacional; Turbinas a gas; Turbocompressores; Analise numerica; Fluxo tridimensional; Escoamento turbulento; Turbom
aquinas; Mecanica dos fluidos; Engenharia mecanica
10.
APRESENTAC
AO:
(X) Nacional ( ) Internacional
ITA, S
ao Jose dos Campos. Curso de Doutorado. Programa de Pos-Graduacao em Engenharia Mecanica
Aeron
autica. Area
de Aerodin
amica, Propuls
ao e Energia. Orientador: Joao Roberto Barbosa. Defesa em
07/01/2009. Publicada em 2009.
11.
RESUMO:
With the advent of powerful computer hardware, Computational Fluid Dynamics (CFD) has been vastly used by
researches and scientists to investigate flow behavior and its properties. The cost of CFD simulation is very small
compared to the experimental arsenal as test facilities and wind-tunnels. In the last years many CFD commercial
packages were developed and some of them possess prominence in industry and academia. However, some specific
CFD calculations are particular cases and sometimes need special attention due to the complexity of the flow.
In these cases, meticulous research becomes necessary. This is the case of turbomachinery flow calculations. The
development of CFD codes applied to turbomachinery flow simulations and its implementation issues are not
available. A few institutions have this type of knowledge. Each CFD code has its particularities. Developing a
CFD code is very interest subject in academia.
In this work, a computational code, written in FORTRAN, was developed to calculate internal flows in turbomachines using CFD techniques. The solver is capable of calculating the three-dimensional flows not only for
turbomachines. For instance, internal and external flows of nozzles and airfoils can be calculated. The approach
used allows the use of unstructured meshes of hexahedral elements. Euler, Navier-Stokes and turbulent equations
can be calculated depending on the user settings. Different numerical schemes were implemented for time and
space integration. Numerical tools to improve the stability and to increase the time-step (local time-step and
implicit residual smoothing) were also implemented and all details are described in this work.
The origin of this solver is to simulate flows in compressors and turbines. Therefore, both rotating and nonrotating frames of reference are calculated simultaneously. Hence, the verification and validation processes were
run for both inertial and non-inertial systems.
A step-by-step design procedure is presented in this work. It is very important to mention that to have a complete
understanding of the flow physics in compressors and turbines the designer must have a solid knowledge of the
operation of gas turbine components.
12.
GRAU DE SIGILO:
(X) OSTENSIVO
( ) RESERVADO
( ) CONFIDENCIAL
( ) SECRETO