You are on page 1of 132

DISSERTATION

Modeling Particle Deposition on


Compressor and Turbine Blade Surfaces

submitted to the Vienna University of Technology


in partial fullment of the requirements for the degree of

Doctor of Technical Sciences

Under supervision of
O.Univ.-Prof.Dipl.-Ing.Dr.techn. Hermann Haselbacher
Head of the Institute of Thermal Turbomachines and Powerplants

Presented by

M.S. Hesham El-Batsh


Matr. Nr. 9726880
Pohlgasse 8/1/3, A-1120 Vienna
Vienna, September 2001

DISSERTATION
Modeling Particle Deposition on
Compressor and Turbine Blade Surfaces
ausgefhrt zum Zwecke der Erlangung des akademischen Grades eines

Doktors der technischen Wissenschaften


eingereicht an der Technischen Universitt Wien,
Fakultt fr Maschinenbau

unter der Leitung von


O.Univ.-Prof.Dipl.-Ing.Dr.techn. Hermann Haselbacher
Vorstand des Institut fr Thermische Turbomaschinen und Energieanlagen
von

M.S. Hesham El-Batsh


Matr. Nr. 9726880
Pohlgasse 8/1/3, A-1120 Wien
Wien, September 2001

Acknowledgements
The research work described in this dissertation had been carried out at the Institute of
Thermal Turbomachines and Powerplants, Vienna University of Technology in the period
from November 1997 until September 2001 under supervision of o.Univ.-Prof. Dipl.-Ing.
Dr.techn. Hermann Haselbacher. This work had been funded by the Austrian Academic
Exchange (AD) and the Mission Department of the Egyptian Ministry of High Education
according to the contract between them.
It is a great pleasure to express my deep grateful thanks to o.Univ.-Prof. Dipl.-Ing.
Dr.techn. Hermann Haselbacher for giving me the possibility to carry out my study at this
institute. I would like to express my grateful thanks to him also for his patient guidance, advices, useful ideas and his contribution over the period it has taken me to complete this work.
Special appreciation is given to o.Univ.-Prof. Dipl.-Ing. Dr.techn. Wladimir Linzer for
reviewing this dissertation.
I would like to thank o.Univ.-Prof. Dipl.-Ing. Dr.techn. Hans-Peter Degischer for performing particle testing at the Institute of Materials Science and Testing.
Special thanks are presented to Dipl.-Ing. Dr.techn. Reinhard Willinger for his help in
so many occasions and his suggestions, kind assistance and cooperation.
I extend my thanks to all the sta at the Institute of Thermal Turbomachines and Powerplants where I enjoyed a great hospitality and could benet not only of the knowledge of this
group but also could learn much about Austria during the period of my stay in the Institute.
I would like to express my thanks to Herrn Ing. Gerhard Kanzler for solving many problems
during the period of my stay in Austria.
I am grateful to the Austrian Academic Exchange (AD) and the Mission Department
of the Egyptian Ministry of High Education for supporting me over the entire period of this
research and for supporting workshops and conferences.
Finally, thanks to my wife for her patience and help and thanks to all my family members
for their encouragement to complete this work.

Hesham M. El-Batsh
September 10, 2001

Contents
Abstract
Kurzfassung
Nomenclature
1 Introduction
1.1
1.2
1.3
1.4
1.5
1.6
1.7

Fouling of Gas Turbine Compressors . . . . . . .


Ash Particle Deposition in Gas Turbines . . . . .
Details of Particle Deposition Process . . . . . .
Review for Previous Deposition Models . . . . . .
Shortcomings of the Previous Deposition Models
The Aim of This Work . . . . . . . . . . . . . . .
Outline of the Thesis . . . . . . . . . . . . . . . .

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

I MODELING OF THE FLOW FIELD


2 Models for Fluid Flow and Heat Transfer

2.1 Basic Conservation Equations for Laminar Flows . . . . .


2.1.1 The Mass Conservation Equation . . . . . . . . . .
2.1.2 Momentum Conservation Equations . . . . . . . .
2.1.3 Energy Conservation Equation . . . . . . . . . . .
2.2 Models for Turbulent Flows . . . . . . . . . . . . . . . . .
2.2.1 Reynolds Averaging of the Conservation Equations
2.2.2 Reynolds Averaging of the Momentum Equations .
2.3 Eddy-Viscosity Turbulence Models . . . . . . . . . . . . .
2.4 Selection of Turbulence Models . . . . . . . . . . . . . . .
2.5 The Standard k-" Model . . . . . . . . . . . . . . . . . . .
2.6 The RNG k-" Model . . . . . . . . . . . . . . . . . . . . .
2.6.1 Transport Equations for k and " . . . . . . . . . .
2.6.2 RNG k-" Model for Heat transfer . . . . . . . . . .
2.6.3 RNG k-" Model for Compressible Flows . . . . . .
2.7 Near-Wall Treatments for Turbulent Flows . . . . . . . . .
2.7.1 The Standard Wall Function . . . . . . . . . . . .
2.7.2 The Two Layer-Zonal Model . . . . . . . . . . . . .
2.8 Compressibility Eects . . . . . . . . . . . . . . . . . . . .
2.9 Numerical Aspects . . . . . . . . . . . . . . . . . . . . . .

ii

1
3
5
9

9
10
12
14
16
17
17

18
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

19

19
19
19
20
20
20
21
22
23
24
24
25
26
26
26
28
28
29
29

Contents

iii

II PARTICLE DEPOSITION MODEL

30

3 Particle Transport Model

31

3.1 Overview for Eulerian and Lagrangian Approaches . . . . .


3.2 Particle Motion in Fluids . . . . . . . . . . . . . . . . . . .
3.2.1 Drag Force . . . . . . . . . . . . . . . . . . . . . . .
3.2.2 Rarefaction Eect . . . . . . . . . . . . . . . . . . .
3.2.3 Pressure Gradient Force and Unsteady Forces . . . .
3.2.4 Lift Forces . . . . . . . . . . . . . . . . . . . . . . . .
3.2.4.1 Saman Lift Force . . . . . . . . . . . . . .
3.2.4.2 Magnus Force . . . . . . . . . . . . . . . . .
3.2.5 Gravity Force . . . . . . . . . . . . . . . . . . . . . .
3.2.6 Thermophoretic Force . . . . . . . . . . . . . . . . .
3.2.7 Brownian Force . . . . . . . . . . . . . . . . . . . . .
3.3 Particle Trajectory Calculation . . . . . . . . . . . . . . . .
3.4 Turbulent Dispersion of Particles . . . . . . . . . . . . . . .
3.5 Heat Transfer to or from the Particles . . . . . . . . . . . .
3.6 Coupling between the Dispersed and the Continuous Phases

4 Particle-Wall Interaction

4.1 Particle Sticking . . . . . . . . . .


4.1.1 Dominant Sticking Force . .
4.1.2 Sticking Model . . . . . . .
4.1.3 Particle Capture Velocity .
4.2 Particle Detachment . . . . . . . .
4.2.1 Detachment Model . . . . .
4.2.2 Critical Wall Shear Velocity
4.3 Properties Assumption . . . . . . .

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

5 Deposition Model in Fluent

5.1 Overview for Deposition Calculations . . . . . . . . . . . . . . .


5.2 Building the Deposition Model in Fluent . . . . . . . . . . . . .
5.2.1 Modication of the Drag Coecient . . . . . . . . . . .
5.2.2 Additional Force Components . . . . . . . . . . . . . . .
5.2.3 Particle-Wall Interaction . . . . . . . . . . . . . . . . . .
5.2.4 Menu for the Sticking and the Detachment Parameters .
5.2.5 Changing Particle Inlet Parameters . . . . . . . . . . . .
5.2.6 Output Postprocessing . . . . . . . . . . . . . . . . . . .

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

III TEST CASES AND RESULTS


6 Validity of the Transport Model

6.1 Parameters Aecting Particle Transport . . .


6.2 Particle Transport in a Pipe Flow . . . . . . .
6.2.1 Experimental System . . . . . . . . . .
6.2.2 Numerical Calculations . . . . . . . .
6.2.3 Comparison to Experimental data . .
6.3 Particle Transport to Turbine Blade Surfaces
6.3.1 Experimental System . . . . . . . . . .
6.3.2 Numerical Calculations . . . . . . . .

31
32
32
33
35
35
35
35
35
36
36
36
37
38
38

40

40
41
43
43
45
45
48
49

50

50
52
52
52
52
53
53
53

55
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

56

56
57
57
58
59
61
61
62

Contents

iv

6.3.3 Comparison to Experimental Data . . . . . . . . . . .


6.3.4 Grid Sensitivity Study . . . . . . . . . . . . . . . . . .
6.4 Thermophoretic Particle Transport . . . . . . . . . . . . . . .
6.4.1 Thermophoretic Particle Transport in Laminar Flow .
6.4.1.1 Experimental System . . . . . . . . . . . . .
6.4.1.2 Numerical Calculations . . . . . . . . . . . .
6.4.1.3 Comparison to Experimental Data . . . . . .
6.4.2 Thermophoretic Particle Transport in Turbulent Flow

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

8.1 Selection of the Blade Cascade . . . . . . . . . . . . . . . . .


8.2 Performance of the Clean Blades . . . . . . . . . . . . . . . .
8.2.1 Solving the Flow Field . . . . . . . . . . . . . . . . . .
8.2.1.1 Computational Grid . . . . . . . . . . . . . .
8.2.1.2 Boundary Conditions . . . . . . . . . . . . .
8.2.1.3 Numerical Calculations . . . . . . . . . . . .
8.2.1.4 Comparison with Experimental Data . . . . .
8.2.2 Calculation of the Losses . . . . . . . . . . . . . . . . .
8.2.2.1 Downstream Total Pressure Coecient . . .
8.2.2.2 Grid Sensitivity Study . . . . . . . . . . . . .
8.3 Temperature Eects on Particle Transport and Sticking . . .
8.3.1 Temperature Eect on Particle Transport . . . . . . .
8.3.2 Temperature Eect on Particle Sticking . . . . . . . .
8.3.2.1 Flow Field Numerical Calculations . . . . . .
8.3.2.2 Parameters Aecting the Sticking Coecient
8.3.2.3 Sticking Coecient Calculations . . . . . . .
8.3.3 Resultant Deposition Rate . . . . . . . . . . . . . . . .
8.4 Ash Particle Deposition on Turbine Blades . . . . . . . . . . .
8.4.1 Flow Field Numerical Calculation . . . . . . . . . . . .
8.4.2 Particle Specications and Inlet Boundary Conditions
8.4.3 Numerical Deposition Calculations . . . . . . . . . . .
8.4.4 Deposition Results . . . . . . . . . . . . . . . . . . . .
8.5 Fouled Blade Prole . . . . . . . . . . . . . . . . . . . . . . .
8.6 Deposition on the Fouled Blade . . . . . . . . . . . . . . . . .
8.7 Predicting the Performance of the Fouled Blade . . . . . . . .

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

7 Investigation for the Sticking Model


7.1
7.2
7.3
7.4

Assumptions . . . . . . . .
Experimental System . . . .
Numerical Calculations . . .
Tuning the Sticking Model .

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

8 Ash Particle Deposition on Turbine Blades

9 Summary and Conclusions


Bibliography
List of Figures
List of Tables

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

66
70
72
72
72
73
77
79

81

81
82
83
84

87

87
89
89
89
91
92
92
95
95
95
96
96
96
96
97
98
100
101
101
102
103
104
106
109
110

114
116
122
125

Abstract
Particle deposition is a major source of gas turbine performance loss. This problem is studied
here using a particle deposition model that is based on the three main deposition processes
as: particle transport to the surface, particle sticking at the surface and particle detachment
from the surface.
Particle transport mechanisms in isothermal turbulent ows are the inertia, turbulent
diusion, Brownian diusion, gravity force and electrostatic force.
Based on Lagrangian approach, a particle transport model is used in this study. The
particle trajectory is predicted by solving the particle equation of motion. The eddy lifetime
model is used to represent the eect of turbulence on the particle movement. The turbulent
uctuating velocities are randomly drawn from Gaussian random distribution of the turbulent kinetic energy.
The eect of turbulence models on particle transport is studied. Two turbulence models
are tested: the standard k-" model and the Renormalization group (RNG) k-" model. The
solution near the wall is obtained by two dierent methods: the standard wall function and
the two-layer zonal model. The results are investigated using two experimental data. The
rst concerns particle transport in turbulent pipe ow while the second concerns particle
transport to turbine blade surfaces. It is found that the RNG k-" model with the two-layer
zonal near-wall model is the appropriate turbulence model for particle transport.
In non-isothermal ow, the temperature gradient aects particle transport by the phenomenon called thermophoresis. The eect of the thermophoretic force on particle transport
is studied in laminar pipe ow. The comparison between the numerical calculations and the
experimental data shows good agreement.
Modeling the eect of thermophoretic force on particle transport to turbine blade surfaces
could not be performed in this study due to a deciency in the turbulence models. The k-"
type models, which are used here to solve the ow eld, depend upon the assumption of
isotropic turbulence. At the near-wall region, the assumption of isotropic turbulence is not
appropriate. In this region, the temperature gradient is high and the thermophoretic force
can signicantly aect the particle motion.
A particle-sticking model based on the particle impact velocity on the surface is used.
This model calculates the particle capture velocity, below which no rebound occurs, from the
elastic properties for the particle and the surface. The particle sticks to the surface if its
normal impact velocity is lower than the capture velocity.

Abstract

The sticking coecient is the mass fraction of incident particles that stick to a surface.
The sticking model is used to investigate the eect of blade surface temperature on the sticking coecient of ash particles. The particle thermal response time relates the time required
for a particle to respond to changes in temperature in the carrier uid. The dependence
of the sticking coecient on the particle thermal response time is studied. The results are
tted into universal curves between the thermal response time and the sticking coecient
at various surface temperatures. Three regions are distinguished: perfect sticking region,
cooling aected region and cooling unaected region.
The particle deposition model is extended with a particle detachment model. This model
calculates the critical wall shear velocity required for particle detachment based on the critical moment theory. The particle is detached from the surface when the uid wall friction
velocity is higher than the critical shear velocity for the particle.
The deposition model is used to study the eect of particle deposition on turbine blade
performance. The VKI transonic inlet guide vane is used in this study. The ow eld is
solved for the clean blade and tested using the experimental data. The deposition model is
applied and the deposition distribution on the blade is calculated in three periods of 12 hours
each. After each period, the fouled blade prole is calculated and the ow eld is solved.
The deposition calculations are repeated to account for the unsteady particle deposition. The
ow eld is calculated for the fouled blade after 36 operating hours to investigate the eect of
deposition on the blade performance. The surface velocity and the downstream total pressure
are used in this investigation.

Kurzfassung
Partikelablagerung ist eine Hauptquelle des Gasturbinenleistungsverlustes. Dieses Problem
wird hier mit einem Partikelablagerungmodell studiert, das auf den drei Hauptablagerungprozessen basiert: Partikeltransport zur Oberche, Partikelhaften an der Oberche und
Partikeltrennung von der Oberche.
Partikeltransport mechanismen in der turbulenten isothermischen Strmung sind die
Trgheitkraft, die turbulente Diusion, die Brownsche Diusion, die Schwerkraft und die
elektrostatische Kraft.
Mit dem Ansatz nach Lagrange wird ein Partikeltransportmodell in dieser Studie benutzt. Die Partikelugbahn wird durch Lsen der Partikelbewegungsgleichung vorausberechnet. Das Wirbellebenszeitmodell wird benutzt, um den Eekt der Turbulenz auf die Partikelbewegung darzustellen. Die turbulenten, schwankenden Geschwindigkeiten werden nach
der Gau-Verteilung der turbulenten kinetischen Energie zufllig entnommen.
Der Eekt der Turbulenzmodelle auf den Partikeltransport wird betrachtet. Der Eekt
der Ausung nahe der Wand wird auch getestet. Zwei Turbulenzmodelle werden geprft:
das Standard k-" Modell und das Renormalization-group-k-" Modell (RNG). Die Ausung
nahe der Wand wird durch zwei unterschiedliche Methoden erreicht: die Standardwandfunktion und das Zweischichtzonenmodell. Die Resultate werden mit zwei experimentellen Daten
verglichen. Das erste ist Partikeltransport in einer turbulenten Rohr-Strmung, whrend das
zweite Partikeltransport zur Turbinenschaufel betrit. Es wird gefunden, da das RNG k-"
Modell mit dem Zweischichtzonenmodell das passende Turbulenzmodell fr Partikeltransport
ist.
In der nichtisothermen Strmung beeinut der Temperaturgradient den Partikeltransport durch das Phnomen, das Thermophorese genannt wird. Der Eekt der thermophoretischen Kraft auf den Partikeltransport wird in der laminaren Rohr-Strmung betrachtet. Der
Vergleich zwischen den numerischen Berechnungen und den experimentellen Daten zeigt gute
bereinstimmung.
Der Eekt der thermophoretischen Kraft auf den Partikeltransport zu den Turbinenschaufeloberchen konnte in dieser Studie nicht durchgefhrt werden. Der Grund ist ein
Mangel in den Turbulenzmodellen. Die k-"-Modelle hngen von der Annahme der isotropen
Turbulenz ab. An der Nahwandregion ist die Annahme der isotropen Turbulenz nicht angebracht. In dieser Region ist der Temperaturgradient hoch, und die thermophoretische Kraft
kann die Partikelbewegung erheblich beeinussen.
Festgelegt durch die Partikelaufprallgeschwindigkeit, wird ein Partikelhaft-Modell benutzt. Dieses Modell errechnet die Partikelfanggeschwindigkeit, unterhalb deren kein Rck3

Kurzfassung

prall von den elastischen Eigenschaften fr das Partikel und die Oberche auftritt. Das
Partikel haftet an der Oberche, wenn seine normale Aufprallgeschwindigkeit niedriger ist
als die Fanggeschwindigkeit.
Der Haftkoezient ist der Massenanteil der auftreenden Partikel, die an einer Oberche
haften. Das Haftmodell wird benutzt, um den Eekt der Turbinenschaufeltemperatur auf den
Haftkoezienten der Aschepartikel zu untersuchen. Die Abhngigkeit des Haftkoezienten
auf die Partikeleigenschaften wird betrachtet.
Das Partikelablagerungmodell wird mit einem Partikeltrennungsmodell erweitert. Dieses
Modell errechnet die kritische Wandschergeschwindigkeit, die fr Partikeltrennung erforderlich ist. Das Partikel wird von der Oberche losgelst, wenn die Wandschergeschwindigkeit
der Strmung hher ist als die kritische Schergeschwindigkeit des Partikels.
Das Ablagerungmodell wird benutzt, um den Eekt der Partikelablagerung auf die Turbinenschaufelleistung zu studieren. Das Strmungsfeld wird fr die saubere Beschaufelung
berechnet und mit den experimentellen Daten berprft. Das Ablagerungmodell wird angewendet und die Ablagerungverteilung auf der Beschaufelung wird in drei Perioden von je 12
Stunden errechnet. Nach jeder Periode wird das verschmutzte Beschaufelungprol errechnet
und das Strmungsfeld wird gelst. Die Ablagerungberechnungen werden wiederholt, um
die unstetige Partikelablagerung zu betrachten. Das Strmungsfeld wird fr die beschmutzte
Beschaufelung nach 36 Betriebsstunden errechnet, um den Eekt der Ablagerung auf die
Beschaufelungleistung zu untersuchen. Die Oberchengeschwindigkeit und der stromabwrts gerichtete Gesamtdruckkoezient werden in dieser Untersuchung verwendet.

Nomenclature
Latin Characters
a
a1 ; a2; a3
A
Ap
Acp
A ; A
c
cax
c
cl
cs
CD
C1", C2", C
Cs ; Ct; Cm
C
C
Cp
Cpp
C+
Cu
D
Dp
D
E
E
Ep
Es
f
FD
FL
Fpo
FS
FT
Fx
Gk

[m]
[;]
[J ]
[m2 ]
[m2 ]
[;]
[m]
[m]
[m=s]
[;]
[m=s]
[;]
[;]
[;]
[;]
[;]
[J=kg K ]
[J=kg K ]
[;]
[;]
[m]
[m]

radius of the contact area between particle and surface


constants in the correlation of the drag coecient
Hamaker constant
surface area of a spherical particle
cross section area of a spherical particle
constants in the two-layer zonal model
blade chord length
blade axial chord length
mean molecular speed
constant in the two layer-zonal model
speed of sound
drag coecient
constants in k-" models
constants in the relation of the thermophoretic force
total pressure coecient
mass averaged total pressure loss coecient
gas specic heat at constant pressure
particle specic heat
empirical constant
Cunningham correction factor
inner diameter of a pipe
particle diameter
diusion of 
[;]
empirical constant (chapter 2)
parameter in the sticking model (chapter 4)
[Pa]
Young's modulus of particle material
[Pa]
Young's modulus of surface material
[;]
drag correction factor for the wall eect
[N ]
uid drag force
[N ]
uid lift force
[N ]
particle sticking force
[N ]
Saman lift force
[N ]
thermophoretic force
[N=kg ]
force in the x-direction per unit particle mass
3
[kg=m s ] rate of production of turbulent kinetic energy

Nomenclature
h
hc
H
k
k1 ; k2
kf
kp
kr
ks
Kc
Kn
l
l; l"
L
Ls
mp
M
Mis
Mt
Nu
po
p
Pf
Pr
Q
R
Rey
Rep
s
S
S
Sij
t
T
Tp
TW
u
u0
u
ui
ui
u0i
uc
u+
u+c
up

6
[J=kg ]
[W=m2 K ]
[m]
[m2=s2]
[1=Pa]
[W=m K ]
[W=m K ]
[1=s]
[;]
[Pa]
[;]
[m]
[m]
[m]
[m]
[kg ]
[;]
[;]
[;]
[;]
[Pa]
[Pa]
[;]
[;]
[m3/s]
[J=kg K ]
[;]
[;]
[m]
[m]
[1=s]
[1=s]
[s]
[K ]
[K ]
[K ]
[m=s]
[m=s]
[m=s]
[m=s]
[m=s]
[m=s]
[m=s]
[;]
[;]
[m=s]

uid specic enthalpy


convective heat transfer coecient
blade height
turbulent kinetic energy
parameters in the sticking model
uid thermal conductivity
particle thermal conductivity
local velocity gradient
constant in the sticking force
composite Young's modulus
Knudsen number
turbulence characteristic length
length scales in the two layer-zonal model
characteristic dimension
length of a section
particle mass
Mach number
isentropic Mach number
turbulent Mach number
Nusselt number
total pressure
static pressure
fraction penetration
Prandtl number
volumetric ow rate
gas constant
turbulent Reynolds number
particle Reynolds number
distance along blade surface
blade pitch
modulus of mean rate of strain
strain tensor
time
uid temperature
particle temperature
surface temperature
uid average velocity in x-direction
uid uctuating velocity in x-direction
uid velocity in x-direction
uid velocity in the ith direction
uid average velocity in the ith direction
uid uctuating velocity in the ith direction
velocity at the center of a stuck spherical particle
dimensionless velocity
dimensionless velocity at the center of a stuck spherical particle
particle velocity in x-direction

Nomenclature
u
uc
U
v
v0
v
vcr
V
V+
w
w0
w
WA
x
xi
y
y+
z

7
[m=s]
[m=s]
[m=s]
[m=s]
[m=s]
[m=s]
[m=s]
[m=s]
[;]
[m=s]
[m=s]
[m=s]
[J=m2]
[m]
[m]
[m]
[;]
[m]

wall friction velocity


critical wall shear velocity
ow velocity
uid average velocity in y-direction
uid uctuating velocity in y-direction
uid velocity in y-direction
particle capture velocity
particle transport velocity
dimensionless particle transport velocity
uid average velocity in z-direction
uid uctuating velocity in z-direction
uid velocity in z-direction
work of sticking
distance in x-direction
coordinate in the ith direction
normal distance to the wall
dimensionless wall distance
separation distance

Greek Characters


0
k , "
mol
p

t

ij
"


0


0



t
e


[m]
[;]
[;]
[;]
[;]
[;]
[;]
[s]
[m]
[;]

relative approach between particle and surface (chapter 4)


parameter in the RNG k-" model (chapter 2)
constant in the RNG k-" model
inverse eective Prandtl number for k and ", respectively
inverse Prandtl number
particle volume fraction
constant in RNG k-" model
time scale
distance along blade height
= 1 if i = j
= 0 if i 6= j
[m2=s3] dissipation rate
conserved scalar quantity
mean value of 
uctuating part of 
[;]
ratio of specic heats
[;]
parameter in RNG k-" model
[;]
constant in RNG k-" model
[;]
Karman constant
[m]
mean free path of the gas molecules
[Pa s] uid viscosity
[Pa s] turbulent viscosity
[Pa s] eective viscosity
[m2=s] kinematic viscosity

Nomenclature
p
s
rT

p
k ; "
e
ij
(ij )e
w

+
T


8
[;]
[;]
[K=m]
[kg=m3]
[kg=m3]
[;]
[s]
[Pa]
[Pa]
[Pa]
[s]
[;]
[s]
[;]

Subscripts
1
2
D
e

inlet
exit
drag
eective
i; j; l tensor indices
p
particle
P
point
s
surface
t
turbulent
x
x-direction

Poisson ratio of particle material


Poisson ratio of surface material
temperature gradient
uid density
particle density
constants in the standard k-" model
characteristic life time of uid eddy
stress tensor
stress tensor calculated using eective viscosity
wall shear stress
particle relaxation time (momentum response time)
dimensionless particle relaxation time
particle thermal response time
normally distributed random number

Chapter 1

Introduction
This introduction discusses the problems of compressor fouling and ash particle deposition
in the turbine section of gas turbines when low-grade fuels are used through a review of the
previous theoretical and experimental work in this eld. The previous work showed that
compressor fouling and ash particle deposition on turbine blade surfaces have a great eect
on performance deterioration. It also showed that deposition mechanisms are still not well
understood and further work is required to investigate the problem in detail. Previous particle deposition models are discussed in this chapter and the shortcomings of these models are
summarized.

1.1 Fouling of Gas Turbine Compressors


Gas turbines utilize large quantities of air. The atmospheric air drawn into a gas turbine unit
is not absolutely pure. It always contains solid particles, moisture and sometimes harmful
gaseous substances. Their sources are soil dust, moisture, pollen and waste products generated by trac and industries. Among them are those of the electric power stations themselves
such as combustion products and fog from water coolers. When the gas turbine works in a
chemical processing area, chemical vapours could be drawn also with the air.
If the solid particles are large and hard, their impact may cause erosion damage to the
compressor blade surfaces. Performance deterioration in turbomachinery caused by erosion
of the blade surfaces was studied by Tabako [68, 69] and Tabako et al. [70].
Filtration systems are used to reduce the inlet particle concentration. It is obvious that
an ecient ltration system is necessary. But the pressure drop through the lter will cause
a loss in the power of the gas turbine. Therefore, the main functions of the inlet lters are
to reduce particulate loading and protect the blades from erosion, but compressor fouling
can not be completely eliminated. It was noted by Diakunchak [20] that compressor blade
fouling is mainly caused by dust particles of about 2 m and less in diameter which are not
removed by lters.
Diakunchak discussed also various factors aecting performance deterioration of gas turbine engines. He found that about 70-85% of all gas turbine engine performance loss accumulated during operation is attributed to compressor fouling. Performance deterioration is
also caused by the increase in leakage or the increase in surface roughness of the ow path
due to erosion but the contribution of these factors is small.
9

1.2 Ash Particle Deposition in Gas Turbines

10

The result of fouling is build-up of material which changes the blade shape, changes the
blade inlet angle, increases surface roughness and reduces the ow area. Consequently, fouling
deteriorates the compressors aerodynamically, reduces the air mass ow rate and reduces the
compressor eciency. These, all together, reduce the gas turbine unit output and increase
the specic fuel consumption.
The correlation between performance variables for a General Electric gas turbine unit
was studied by Tarabrin et al. [72]. They found that when the compressor was fouled to the
degree which reduced the pressure ratio by 5.5 %, the gas turbine unit output decreased by
13 % and the specic fuel consumption increased by 6%.
A similar result was mentioned by Diakunchak [20]. He noted that compressor fouling
which causes a drop of about 5% in the mass ow results in 2.5% reduction in the eciency
and 10% in the output.
Tarabrin et al. showed also that the rate of compressor fouling was high during the rst
1000 operating hours. The gas turbine unit output and eciency losses decreased as the operating time increased and they tended to stabilize after 1000-2000 operating hours. Stalder
[67] got a similar result in his work. He found that without cleaning, the power output
degradation stabilized at 90% of the base load.
Saravanamuttoo and Lakshminarasimha [60] investigated the eect of fouling of compressor on its performance. They found that fouling of the rst stages aects all stages by forcing
them to operate at new coecients. Fouling of the last stages does not seriously aect the
upstream stages.
Haq et al. [27] conrmed this result when they investigated the eect of multistage centrifugal compressor fouling on the performance. Their study revealed that fouling of early
stages adversely aected the eciency of later stages. It was also found in their study that
fouling dropped the pressure ratio. The reduction in the pressure ratio was attributed to the
rough surfaces, deviation of the ow angles from metal angles and the reduction in the ow
area.

1.2 Ash Particle Deposition in Gas Turbines


Limited access to and hence high costs of high quality fuels such as natural gas and light
distillates have forced many operators of gas turbine installations to use low-grade fuels.
Most commonly used are crude oils and biomass. These fuels are always contaminated with
varying levels of impurities. Combustion of such fuels produces relatively large amounts of
ash.
Ash particles in the hot gas stream have various eects. Impacting on the turbine surfaces by their inertia they can cause erosion. When the ash particles in the hot gas stick
to the blade surface, they build up layers of deposits depending on the quality of the fuel.
The aerodynamic performance deteriorates due to growing deposit thickness, decreasing the
eciency and the power of the gas turbine. Consequently, the machine has to be shut down
at certain intervals for blade cleaning. Moreover, ash particle deposition on the turbine hot
sections may result in premature damage of turbine blades due to hot corrosion.

1.2 Ash Particle Deposition in Gas Turbines

11

Hot gas clean-up devices are being developed to protect gas turbines from ow path
degradation due to deposition, erosion and corrosion. Cyclone separators have demonstrated
a capability to greatly alleviate the deposition, erosion and corrosion problems. Sethi and
Partanen [62] characterized the products of combustion in a wood-red gas turbine system.
They found that particulate loading is in the 200-300 ppmw range. They found also that
by using cyclone separators, the particulate loading in turbine gas path could be reduced by
a factor of about two. They concluded that erosion of turbine components is not a major
concern in wood-red systems.
Ragland et al. [53] studied the eect of ash particle deposition on the ow rate through
a wood-red gas turbine by using a small four stage gas turbine for 130 hours of test. Their
result showed that ash particle deposition reduced the ow area by 4.2% per 1000 kg of wood
burned.

Kawagishi et al. [33] studied the eect


of particle deposition on turbine performance.
They used a single stage experimental turbine.
In order to simulate the deposits, solids of silicone compounds were articially attached to the
nozzles and the moving blades. Location and
quantity of deposits were chosen from a typical
distribution of deposit in an operating turbine.

Total pressure coefficient C []

Blcs and Sari [9] studied experimentally the eect of deposits on the ow in a turbine
cascade. They attached plastic cement on the blade surface according to a real distribution
of deposits obtained from a gas turbine. They investigated the ow elds for the clean blade
and for the fouled blade. They found that because of the high surface roughness of the deposits, the boundary layer became turbulent and thicker almost immediately downstream of
the blade leading edge on the suction surface. Obviously, the increase in the boundary layer
thickness gives a thicker wake with large recirculation zones and thus high losses.
0.0

-0.1

-0.2

-0.3

Clean nozzle

Fouled nozzle
Figure 1.1 shows the distributions of the to-0.4
tal pressure coecients for the clean nozzle and
0.0
0.2
0.4
0.6
0.8
1.0
Distance
along
nozzle
height
/H
[]

for the fouled nozzle versus the distance along


nozzle height  . The distance  is measured
from the hub to the tip region and is given in Figure 1.1: Distributions of the downdimensionless form using the nozzle height H . stream total pressure coecient, Kawagishi et al. [33]
The total pressure coecient C is dened as:

C = p0o2:5;Up2o1
2

where:

po1
po2

U2

inlet total pressure


exit total pressure
uid density
exit velocity

The gure indicates that the total pressure coecient was strongly aected by particle
deposition. The dierence between the clean nozzle and the fouled nozzle varied between

1.3 Details of Particle Deposition Process

12

0.1 and 0.2 from the dimensionless nozzle height =H = 0:5 to the tip region. This large
decrease in the total pressure was attributed to roughness of the nozzle surface and clogging
of the nozzle throat. Kawagishi et al. measured also the stage eciency at dierent loadings
and found that particle deposition decreased the stage eciency by about 20 % of maximum
eciency over a wide range of loading.
Huang et al. [30] modeled ash particle deposition in coal-red gas turbines. They studied
the eect of deposition on the rst stage performance by considering the ow blockage at the
nozzle throat. Their results showed that about 16 % of the unit power was lost after 250
operating hours when the inlet particle concentration was about 4 ppmw.

1.3 Details of Particle Deposition Process


The processes governing the rate and the distribution of particle deposition are (gure 1.2):
the transport of the particles to the surface, the sticking of the particles at the surface due to
the sticking forces between the particles and the surface and the detachment of the particles
from the surface by the uid ow.
Deposition Process

Particle transport

Particle sticking

Particle detachment

- Inertia force
- Turbulent diffusion
- Brownian diffusion
- Electrostatic force
- Thermophoresis

Van der Waal


force

Fluid force

Figure 1.2: Details of particle deposition process


Particle transport mechanisms are the inertia, turbulent diusion, Brownian diusion,
gravity force, electrostatic force and thermophoresis.
Inertia is the phenomenon caused by the inability of the particles to follow exactly the
curved uid streamlines. For turbine blades, particles will impact directly only near the blade
leading edge and on the blade pressure surface. For a given ow, the inertia eect increases
with increasing particle size.
Turbulent diusion is the process by which the particles in the turbulent boundary layer
migrate to the surface under the inuence of random uctuations in the ow. Although these
eddies dissipate as they approach the surface, the particles continue to travel toward the
surface by the free ight due to their inertia. For very small particles, turbulent diusion is
a very important mechanism for particle transport.

1.3 Details of Particle Deposition Process

13

Brownian diusion is the mechanism of moving the particles by random impact of agitated gas molecules. The eect of molecular impacts increases as the particle size decreases.
The particles may carry electric charges of the same polarity. Therefore, they would produce repulsive forces among themselves causing electrostatic force.
For non-isothermal ow, a new term known as the thermophoretic force is also aecting
the particle motion and the mass rate of deposition. This force is caused by the temperature
gradient in the ow eld. Thermophoretic force is acting on the particles in the direction of
decreasing temperature.
Once particles arrive at the surface, deposit build-up depends on the balance of the sticking forces and the removal forces acting on the particles at the surface. The removal forces
include the rebounding force and the uid forces. Under dry conditions, the van der Waals
force is the major contribution to the particle sticking force. The van der Waals forces arise
from the molecular interaction between the particles and the surface. If the rebounding force
could overcome the sticking force, the particle rebounds. For the stuck particles, when the
removal forces exerted by the uid ow are sucient to prevent the particles from remaining
on the surface, the particles are detached.
The particle momentum response time relates the time required for the particle to respond to the changes in uid velocity and is often called particle relaxation time. Particle
relaxation time depends upon particle size, particle density and uid viscosity.
Basic studies in particle transport investigated the eect of the particle relaxation time
on the rate of particle transport to the surface. Liu and Agarwal [38] measured the transport
of olive droplets of 1.4 to 21 m diameter from turbulent air ow to the internal wall of a
smooth glass tube for nominal pipe Reynolds numbers of 10000 and 50000. They represented
the rate of particle transport in velocity units called particle transport velocity. The particle transport velocity and the particle relaxation time were given in dimensionless forms
using the wall friction velocity and the uid kinematic viscosity. They presented their experimental data as a universal curve between the dimensionless particle transport velocity and
dimensionless particle relaxation time. Figure 1.3 shows the universal curve given by Liu and
Agarwal.
Based on this distribution, many authors distinguished three basic regimes of particle
transport.

 Brownian diusion regime. This regime is characterized by extremely small particles

that are transported to the surface by Brownian diusion.


 Turbulent diusion regime. This regime describes the transport of larger particles under
the inuence of turbulent uctuations. Since turbulent diusion is enhanced by particle
inertia, the transport rate increases as the particle diameter increases.
 Inertia regime. In this regime particle inertia is more important than other transport
mechanisms.

Dimensionless transport velocity [-]

1.4 Review for Previous Deposition Models

14

100

Turbulent
diffusion regime

10-1

10-2

Inertia regime

Brownian
regime

10-3

10-4

10-5

10-1

100

101

102

103

Dimensionless particle relaxation time [-]

Figure 1.3: Dimensionless particle transport velocity versus dimensionless particle relaxation
time, Liu and Agarwal [38]

1.4 Review for Previous Deposition Models


Theoretical studies of particle transport and deposition can be developed either by Eulerian
approach or by Lagrangian approach.
The Eulerian approach is based on the assumption that two continuous elds are present
and the transport equations are solved simultaneously for both phases. In this approach,
particle transport is proportional to the concentration gradient with a constant called diffusion coecient. This assumption is called Fick's law. A uniform particle concentration is
usually assumed at the outer edge of the boundary layer. Therefore, this model is suitable
for very small particles. Increasing the particle size, increases the eect of particle inertia
and increases the error generated from the assumption of Fick's law.
In the Lagrangian approach, a number of particle trajectories are simulated by solving
the particle equation of motion, i.e., the particles are considered individually. Various forces
applied on the particle are considered in the particle equation of motion. Therefore, this
model is valid for all particle sizes. This is a brief description because the complete model is
presented in chapter 3.

Eulerian models. Based on the regimes of particle transport shown in gure 1.3, various

Eulerian particle transport models had been developed. Menguturk and Sverdrup [42] developed a model which calculated the transport of particles by the inuence of turbulent and
Brownian diusions. The model was based on the assumption that the particles were very
small so that the inertia eect could be ignored. They used this model to calculate deposition
rates on the blade suction surface of a linear turbine cascade. They assumed that all particles
smaller than 3 m stick to the surface but larger particles rebound from the surface owing
to their high kinetic energy.
Transport of submicron particles (<1 m) by turbulent and Brownian diusions had been
studied theoretically by other workers assuming that they are the dominant mechanisms and

1.4 Review for Previous Deposition Models

15

ignoring the eect of the inertia force. Yau and Young [80], Wood [79] and Kladas [35]
determined particle transport by giving the volumetric concentration of particles just outside
the boundary layer. Subsequently, they solved the continuity equation for particles under
turbulent ow conditions.
Huang et al. [30] and Ahluwalia et al. [4] studied the transport of ne particles in coalred gas turbines. In their Eulerian deposition models, the Brownian diusion, turbulent
diusion and thermophoresis were taken into account. The models calculated the Brownian
and turbulent diusions of particles and solved the particle transport equations through the
boundary layer.

Lagrangian models. Lagrangian type trajectory models have been used to a much lesser

extent than Eulerian models in the prediction of particle deposition. The reason is primarily
due to their computational expense. The Lagrangian approach provides a more detailed and
realistic model of particle transport due to the fact that the instantaneous particle equation
of motion is solved for each particle moving through the eld of random uid eddies. Moreover, it provides detailed information on particle collision at the surface required for sticking
studies.
Kallio and Reeks [31] presented a Lagrangian approach to model particle transport in
turbulent duct ows. They solved the equation of motion for particles with relaxation time
ranging from 0.3 to 1000. Their results showed very good agreement with the experimental
data of Liu and Agarwal [38] over a wide range of particle sizes.
Young and Yau [81] calculated the transport of fog droplets on steam turbine blades by
the inertia force using a Lagrangian approach. They tracked a number of droplets through a
specied blade channel.
Greeneld and Quarini [24, 25] determined the motion of the particles by solving the
particle equation of motion. They considered the drag force as the principle force acting on
the particle. They included the eect of turbulence on the particles by using the eddy lifetime
model. The turbulence was modeled as a series of random eddies which had a lifetime and
associated random uctuating velocities.
Studies using both Eulerian and Lagrangian models were also performed in the previous
work. Abuzeid et al. [1, 2] modeled the transport of particles to the surface in turbulent
channel ow by using Eulerian and Lagrangian simulations. In the Eulerian approach, they
studied particle diusion by solving the diusion equation. In the Lagrangian approach, the
turbulent uctuating velocity eld was numerically simulated as a Gaussian random process.
The particle trajectories in the channel were calculated by solving the corresponding equation of motion. They studied the transport of various particle sizes. They found that the
Lagrangian simulation was more accurate when compared to the Eulerian one.
Fackrell et al. [21] modeled the transport of particles to turbine blade surfaces using two
approaches. They used a particle trajectory model for large particles which accounts only for
the inertia force as developed by Young and Yau [81]. For very small particles, they modeled the diusive particle transport by treating the particles as a separate continuum phase
transported by the gas phase. They included in this model the eect of Brownian diusion,
turbulent diusion and thermophoresis. They calculated the diusion coecient as given by
Yau and Young [80] and the thermophoretic diusion coecient from Talbot et al. [71].

1.5 Shortcomings of the Previous Deposition Models

16

Fackrell et al. developed also a simple model for diusive particle transport. The model
used the standard analogy between heat and mass transfer to derive a mass transfer coecient from a known heat transfer coecient for the same ow situation. The heat transfer
coecient was derived from experiments or previous heat transfer calculations. To allow for
the thermophoretic eect on the mass transfer coecient, it was multiplied by a factor given
by Gkoglu and Rosner [23]

1.5 Shortcomings of the Previous Deposition Models


The common shortcomings of the previous particle deposition models can be summarized in
the following:
 Various particle transport models had been developed in the previous work based on
turbulent and Brownian diusions by considering that these mechanisms dominate
particle transport and ignoring the eect of particle inertia. Although these models
can predict particle transport to the surface for a certain range of particle sizes, there
are overlapping regimes in which a combination of the inertia force and the turbulent
diusion is in eect as noted by Menguturk and Sverdrup [42]. In addition, these
models assume constant particle concentration outside the boundary layer. Fackrell et
al. [21] noted that when the particles do not undergo signicant direct impact, inertial
eects can still inuence their diusive particle transport by modifying the particle
concentration at the outer edge of the boundary layer.
 Almost all previous deposition models are based on the assumption of perfect sticking
of the particles at the surface and no detachment of the deposit occurs. Therefore,
they are really theories of particle transport. Because this approach does not consider
sticking behaviour due to the physical and the chemical characteristics of the gas, the
entrained particles and the surface, it is inadequate and generally results in an overprediction of the deposition rate.
The sticking coecient is the mass fraction of incident particles that stick to a deposition
target. Ahluwalia et al. [3] determined experimentally the sticking coecients for coal
fuels. They found that the sticking coecients changed between 0.0003 and 0.11 and
were strong functions of gas and surface temperatures. Logan et al. [40] and Anderson
et al. [6] measured the sticking coecients of a series of coals burned at dierent
combustion temperatures. The experiments were carried out at conditions similar to
the conditions expected in the rst stage of a gas turbine. The results indicated that
the sticking coecients changed between 0.001 and 0.1 depending on the gas and the
surface temperatures.
 The probability of particle sticking to a surface is likely to be a function of several
parameters including particle impact velocity and angle of incidence. Dahneke [18]
studied the impact and the sticking of particles. He found that the probability of particle sticking increases as the particle impact velocity decreases. A shortcoming of the
Eulerian particle transport models is that they do not provide the information on particle impaction at the surface. This information is obtained by using the Lagrangian
models.

1.6 The Aim of This Work

17

1.6 The Aim of This Work


The aim of this work is to develop a more general model for particle deposition that can
overcome the mentioned shortcomings and obtain a better insight into the deposition process. The model is based on the Lagrangian approach to accurately predict particle transport
and to provide the complete information on particle impaction at the surface. The particle
equation of motion including dierent applied forces on the particle is solved. The parameters of particle impact on the surface are used to predict the sticking fraction of the hitting
particles. A particle detachment model is used to investigate particle detachment from the
surface by the uid ow. A commercial CFD computer code is extended by routines to allow
the deposition model.
It is aimed to solve the ow eld numerically on the CFD computer code by using the
appropriate models for uid ow and heat transfer. The particle trajectories are solved and
the sticking and the detachment models are applied to the particles at the surface. The results are compared to the available experimental data on particle deposition on turbine blade
surfaces and on some other geometries. The eect of particle deposition on the performance
of turbine blades is investigated by using the deposition model.

1.7 Outline of the Thesis


The thesis is divided into three main parts. Following is a brief description for the contents
of each part:
 Part I, Modeling of the Flow Fluid. This part includes a chapter that introduces
the basic conservation equations for uid ow and heat transfer. Turbulent ow models which are used in this work are also described. Finally, some numerical aspects
concerning the solution of the transport equations are given.
 Part II, Particle Deposition Model. This part introduces the deposition model
as developed in this work through three chapters. In the rst chapter of this part,
the Lagrangian particle transport model is presented. The particle equation of motion
is written considering dierent applied forces on the particle. In the second chapter,
the details of the sticking and the detachment models are given. Various parameters
aecting the sticking and the detachment processes are discussed. In the last chapter
of this part, the main features of the deposition model as built in the commercial CFD
code are described.
 Part III, Test Cases and Results. This part contains three chapters. The rst
chapter concerns the particle transport model and its validation. The results from
various case studies are introduced with comparisons to experimental data. The second
chapter of this part presents the results of an investigation for the sticking model. The
third chapter includes the results from the complete deposition model with comparison
to the available experimental data.
In the nal chapter of this thesis, some conclusions and remarks are drawn form this work.
Some recommendations to reduce ash particle deposition are included. Some directions for
future works are also suggested.

Part I

MODELING OF THE FLOW FIELD

18

Chapter 2

Models for Fluid Flow and Heat


Transfer
This chapter describes the basic physical models for uid ow and heat transfer and their
governing equations. The basic conservation equations for laminar ows are introduced.
For turbulent ows, the time averaging procedure known as Reynolds averaging is used to
obtain the conservation equations. The turbulence models used in this study are discussed
and introduced. These models are the standard k-" model and the ReNormalization Group
(RNG) k-" model. Since the RNG k-" model is used in this study to perform compressible ow
calculations, the modication needed to account for the eect of compressibility is presented.
The near-wall region is accounted by using two methods: the standard wall function and the
two-layer zonal model.

2.1 Basic Conservation Equations for Laminar Flows


The governing equations for uid dynamics are based on the conservation of mass, momentum and energy.

2.1.1 The Mass Conservation Equation

The mass conservation equation for incompressible as well as compressible steady ows is
given by:

@ (u ) = 0
@xi i
where

(2.1)

ui velocity in the ith direction


xi coordinate in the ith direction

2.1.2 Momentum Conservation Equations

Conservation of momentum in the ith direction for a steady ow neglecting body force is
described by:

19

2.2 Models for Turbulent Flows

20

@
@p @ij
@xj (uiuj ) = ; @xi + @xj
where p is the static pressure and ij is the stress tensor given by:

@u
@u
@ul 
i
j
ij =  @x + @x ; 23  @x
ij
j
i
l


and

(2.2)

(2.3)

i; j; l tensor indices = 1, 2, 3

uid viscosity
(
1 if i = j
ij
0 if i =6 j

2.1.3 Energy Conservation Equation

The conservation of energy is given by the following equation [22]:

@ (u h) = @ k @T  + u @p +  @ui
i @x
ij @x
@xi i
@xi f @xi
i
j
where

(2.4)

h uid specic enthalpy


T uid temperature
kf uid thermal conductivity

2.2 Models for Turbulent Flows


Modeling of turbulent ows requires appropriate modeling procedures to describe the eect
of turbulent uctuations of velocity on the basic conservation equations. The time averaging
procedure is appropriate for stationary turbulence, i.e. a turbulent ow that on the average
does not vary with time. The time averaging is commonly known as Reynolds averaging.

2.2.1 Reynolds Averaging of the Conservation Equations

The conservation equations for turbulent ows are obtained from those for laminar ows
using the time averaging procedure. The Reynolds averaging procedure for scalar equations
can be illustrated using a generic transport equation for a conserved scalar quantity 

@ (u ) = D
@xi {z i } |{z}
|
Convection

where D is the diusion of 

Diusion

(2.5)

2.2 Models for Turbulent Flows

21

The value of  in turbulent ow is assumed to be comprised of a mean value  and a


uctuating part 0 .

 =  + 0
where  is the time averaged value of  dened as:
 = 1t

t+t

 dt

(2.6)

(2.7)

and t is a time scale much larger than the largest time scale of turbulent uctuations.
Turbulent uctuations are assumed to be random such that

0 = 0

(2.8)

Substitution of equation 2.6 in the general conservation equation 2.5 and time integration
over a suciently large time interval yields

@ ( u ) = ; @ (u0 0 ) + D

@xi i
@xi i

(2.9)

In all following equations, the overbar will be dropped from the averaged quantities (,
ui ) for the sake of convenience.
The terms in equation 2.9 are similar to those in its laminar ow counterpart, (e.g. equation 2.5) except that each quantity now is represented by its time averaged value and a new
term containing the correlation u0i 0 appears on the right-hand side. Physically, this correlation multiplied by the density represents the transport of  due to turbulent uctuations.

2.2.2 Reynolds Averaging of the Momentum Equations

In the Reynolds averaging of the momentum equations, the velocity at a point is considered
as a sum of the mean ui and the uctuating component u0i :

ui = ui + u0i

(2.10)

Substituting expressions of this form into the basic momentum balance and dropping the
overbar on the mean velocity ui yields the ensemble-averaged momentum equations applied
for predicting turbulent ows [22]:

@ (u u ) = @   @ui + @uj  ; 2  @ul   ; @p + @ (;u0 u0 )


@xj i j @xj
@xj @xi
3 @xl ij @xi @xj i j

(2.11)

2.3 Eddy-Viscosity Turbulence Models

22

Equation 2.11 has the same form as the fundamental momentum balance with velocities
now representing time-averaged values and the eect of turbulence incorporated through the
Reynolds stresses u0iu0j . The term u0i u0j is a symmetric second order tensor since:

u0iu0j = u0j u0i

(2.12)

and hence has six unique terms. The main task of turbulence models is to provide expressions or closure models that allow the evaluation of these correlations in terms of mean ow
quantities.

2.3 Eddy-Viscosity Turbulence Models


In the eddy-viscosity models, the Reynolds stresses are assumed to be proportional to the
mean velocity gradients with the constant of proportionality being the turbulent viscosity t .
This assumption, known as the Boussinesq hypothesis, provides the following expression for
the Reynolds stresses:

u0iu0j

@ui + @uj + 2  @ul 


=  23 kij ; t @x
3 t @xl ij
j @xi


(2.13)

where k is the turbulent kinetic energy given by:

k = 12

u0i2

(2.14)

Equation 2.13 for the Reynolds stresses is analogous to that describing the shear stresses
that arise in laminar ow (equation 2.3) with the turbulent viscosity t playing the same
role as the viscosity . Therefore the form of the Reynolds averaged momentum equations
remain identical to the form of the laminar momentum equations (equation 2.2) except that
 is replaced by an eective viscosity e given by:

e =  + t

(2.15)

Eddy-viscosity models include a number of classes. All of these models approximate the
eect of the turbulence on the mean motion by introducing the eective viscosity as given
in equation 2.15. The dierent classes of eddy-viscosity models are distinguished by the
number of additional dierential equations that are solved to determine turbulent viscosity
t. Dimensional analysis suggests that t is the product of the density, a velocity scale and a
length scale. Two-equation models solve dierential equations to determine these two scales.
The turbulent kinetic energy is used to obtain the velocity scale. A commonly used approach
to determine the length scale is to develop a transport equation for the dissipation rate of
the turbulent kinetic energy " that is dened as:

@u0i @u0i
" =  @x
@x
j

where  is the kinematic viscosity.

(2.16)

2.4 Selection of Turbulence Models

23

2.4 Selection of Turbulence Models


The turbulence model has a great eect on the particle trajectory and therefore on particle
deposition. In turbulent ows, the eddy lifetime model is used to represent the eect of turbulent uctuating velocity on the particle movement. The uctuating velocities are randomly
drawn from a Gaussian random distribution of the turbulent kinetic energy. Consequently,
the turbulence model aects the particle trajectory through the value of the turbulent kinetic
energy. For more details about particle trajectory calculation, refer to chapter 3.
The ow eld is solved in this study using the commercial code Fluent 4.4. The available
turbulence models in Fluent 4.4 are:
 The standard k-" model
 The ReNormalization Group (RNG) k-" model
 The Reynolds Stress Model (RSM)
Both the standard k-" model and the RNG k-" model are available with either wall functions or with the two layer-zonal model for the near-wall region. The Reynolds stress model
is available only with wall functions.
The standard k-" model proposed by Launder and Spalding [36] falls in the category of
the two-equations turbulence models based on an isotropic eddy-viscosity concept. As such,
it is more universal than other low-order turbulence models. Robustness, economy, and reasonable accuracy for a wide range of turbulent ows explain its popularity in industrial ow
and heat transfer simulations.
The RNG k-" model also belongs to the k-" family of models. There is a major dierence
between the RNG k-" and the standard k-" models. The RNG k-" model has an additional
term in its " equation which signicantly improves the accuracy for rapidly strained ows.
The RNG k-" model showed better performance than the standard k-" model in the prediction of turbomachinery heat transfer as found by Orszag et al. [48].
The RSM is the most elaborate turbulence model that Fluent 4.4 provides. The RSM
is available in Fluent 4.4 only with wall functions. Menguturk et al. [43] used a particle
trajectory model to investigate the eect of the blade boundary layer on the trajectories of
particles and thus on the resulting erosion and deposition. Their results showed that the
eect of the boundary layer on the trajectories of particles smaller than 10 m is important.
Since particle deposition is mainly caused by the particles smaller than this size, the required
turbulence model for the study of particle deposition should include the solution for the nearwall region which is accomplished in Fluent 4.4 using the two layer-zonal model.
As a consequence, the turbulence models used in this study are:
 The standard k-" model
 The RNG k-" model

2.5 The Standard k-" Model

24

2.5 The Standard k-" Model


The standard k-" model is an eddy-viscosity model in which the Reynolds stresses are assumed
to be proportional to the mean velocity gradient with the constant of proportionality being
the turbulent viscosity t . The turbulent viscosity t is obtained by assuming that it is
proportional to the product of a turbulent velocity scale and a length scale. In the standard
k-" model, these velocity and length scale are obtained from two parameters: the turbulent
kinetic energy k and its dissipation rate ". Hence t is given by:
2
t = C k"

(2.17)

where C is an empirically derived constant of proportionality.


The values of k and " required in equation 2.17 are obtained by solving their transport
equations [36] as:

@ (u k) = @  t @k  + G ; "
k
@xi i
@xi k @xi

(2.18)

@ (u ") = @  t @"  + C " G ; C  "2


1" k k
2" k
@xi i
@xi " @xi

(2.19)

where Gk is the rate of production of turbulent kinetic energy given by:



@u
@u
i
j
j
Gk = t @x + @x @u
(2.20)
j
i @xi
The coecients C1" , C2", C , k and " are empirical constants which have the values


shown in table 2.1.

C1" C2" C k "


1:44 1:92 0:09 1:0 1:3
Table 2.1: The values of the constants in the standard k-" model

2.6 The RNG k-" Model


The RNG k-" turbulence model follows the two-equation turbulence modeling framework
and has been derived from the original governing equations for uid ow using mathematical techniques called Renormalization group (RNG) methods. Using the RNG k-" model,
the eect of low Reynolds numbers is included
by calculating the eective viscosity using a
p
dierential relation between e and k= " as [22]:
"

e =  1 + C pk"

(2.21)

2.6 The RNG k-" Model

25

2.6.1 Transport Equations for k and "

The RNG theory provides the transport equations for k and " as:

@ (u k) = @   @k  +  S 2 ; "
t
@xi i
@xi k e @xi

(2.22)

@ (u ") = @   @"  + C "  S 2 ; C  "2 ; R


1" k t
2" k
@xi i
@xi " e @xi

(2.23)

where k and " are called in the literature the inverse eective Prandtl numbers for k and
", respectively. They are calculated using the following formula derived analytically by RNG
theory as:



; 1:3929 0:6321 + 2:3929 0:3679 = 


+ 2:3929
0 ; 1:3929
e
0

(2.24)

where 0 =1:0. In the high-Reynolds number limit (=e  1:0), k = "  1:393.

S is the modulus of the mean rate of strain tensor Sij which is dened as:
S = 2Sij Sij
p

and

(2.25)

@ui + @uj
Sij = 12 @x
j @xi


(2.26)

R in the " equation is given by:


3 ; =0) "2
R = C1 +(1 
3
k

(2.27)

where  = Sk=", 0 = 4:38, = 0:012. The model constants C1" and C2" in equation 2.23
are derived analytically by the RNG theory. Table 2.2 shows the values of the constants used
in the RNG k-" model.

C1" C2"

C

1:42 1:68 0:0845


Table 2.2: The values of the constants in the RNG k-" model

2.7 Near-Wall Treatments for Turbulent Flows

26

2.6.2 RNG k-" Model for Heat transfer

In the RNG k-" model, the transport equation for the energy is modeled by [22]:

@ (u h) = @  C  @T  + u @p + ( ) @ui
p e @x
i @x
ij e @x
@xi i
@xi
i
i
j
where

Cp

(2.28)

gas specic heat at constant pressure

(ij )e stress tensor calculated using eective viscosity


The RNG theory gives a formula for in terms of the ratio of the uid viscosity to
eective viscosity, =e , as:

; 1:3929 0:6321 + 2:3929 0:3679 = 



mol ; 1:3929
e
mol + 2:3929
where mol = 1=Pr = kf =Cp and Pr is the Prandtl number.



(2.29)

2.6.3 RNG k-" Model for Compressible Flows

In the RNG k-" model, the eect of compressibility is accounted for by modifying the equation
of the turbulent kinetic energy as [22]:

@ (u k) = @   @k  +  S 2 ; "(1 + 2M 2 )
t
t
@xi i
@xi k e @xi

(2.30)

where Mt is the turbulent Mach number dened as:

p
Mt = c k
s
and

(2.31)

p
cs speed of sound (= RT )
R gas constant
ratio of specic heats

2.7 Near-Wall Treatments for Turbulent Flows


Turbulence models are largely valid for turbulent core ows, i.e., the ows in the regions
somewhat far from walls. When the ow to be calculated involves walls, turbulent ows in
the regions close to the walls are aected by the presence of the walls. First, the mean velocity
eld is aected through the no-slip condition that has to be satised at the wall. Turbulence
is also changed by the presence of the wall. Very close to the wall, turbulence is damped due
to the presence of walls. Toward the outer part of the near-wall region, turbulence is rapidly

2.7 Near-Wall Treatments for Turbulent Flows

27

augmented by the production of turbulent kinetic energy due to Reynolds stresses and the
large gradient of the mean velocity.
Experiments have shown that the near-wall region can be largely subdivided into three
layers. In the innermost layer called the viscous sublayer, the ow is almost laminar-like,
and the viscosity plays a dominant role in momentum and heat transfer. In the outer layer,
called the fully-turbulent layer, turbulence plays a major role. Finally, there is an interim
region between the viscous sublayer and the fully-turbulent layer called buer layer where
the eects of viscosity and turbulence are equally important.
In the near-wall region, the velocity has a universal distribution. Numerous measurements
of this distribution exist. According to these measurements, the viscous sublayer and the fullyturbulent region can be represented as functions between the dimensionless wall distance y +
and the dimensionless velocity u+ as [61]:

u+ = y +
u+ = (1=) ln y + + C +

viscous sublayer:
fully-turbulent region:
where

y + = u y

u+ = uu ;


and


C+
u
w
y

0  y+ < 5
70 < y+

Karman constant (=0.42)


empirical constant (=5.0)p
wall friction velocity (= w =)
wall shear stress
normal distance to the wall

Figure 2.1 shows the distribution of u+ versus y + drawn using the above relations.

Dimensionless velocity u []

30

20
fully turbulent layer
u+=1/ ln y++C+
10

viscous sublayer
u+= y+

0
1

10

100

1000

10000

Dimensionless wall distance y []

Figure 2.1: Universal laws of the wall, Schlichting and Gersten [61]

2.7 Near-Wall Treatments for Turbulent Flows

28

There are two approaches for the treatment of the near-wall region. In one approach
referred to in the literature as the wall function approach, the viscosity-aected, inner region
(viscous sublayer and buer layer) is not solved. Instead, the wall function is used to bridge
the viscosity-aected region between the wall and the fully-turbulent region. Based on this
approach, the standard wall function is used in this study. In another approach which may be
called the near-wall modeling approach, the viscosity-aected region is solved with a mesh all
the way to the wall including the viscous sublayer. In this study, the two layer-zonal model
is used.

2.7.1 The Standard Wall Function

The standard wall function used in this study is based on the proposal of Launder and Spalding [36]. It has been most widely used for industrial ows. The law of the wall for mean
velocity yields [22]:

UP C1=4kP1=2 = 1 ln Ey C1=4kP1=2
P
w =


where

E
UP
kP
yP

(2.32)

empirical constant (=9.81)


mean velocity of the uid at point P
turbulent kinetic energy at point P
distance from point P to the wall

2.7.2 The Two Layer-Zonal Model

In the two-layer zonal model [22], the whole domain is subdivided into a viscosity-aected
region and a fully-turbulent region. The demarcation of the two regions is based on the
turbulent Reynolds number Rey dened as:

p

(2.33)
Rey = ky
In the fully turbulent region (Rey > 200), the k-" or RNG k-" turbulence model is
employed. In the viscosity-aected near-wall region (Rey < 200), a one-equation model
is employed. In this model, the momentum equations and the k-equation are retained as
described in sections 2.5 and 2.6. However, the turbulent viscosity t is calculated from:
p
t = C kl
(2.34)
The "-eld is calculated from:

3= 2

" = kl

"

(2.35)

2.8 Compressibility Eects

29

where l and l" are length scales and are calculated from:

Re
y
l = cl y 1 ; exp ; A


(2.36)


Re
y
l" = cly 1 ; exp ; A


(2.37)

The constants in the length scale formulas are:

cl = C;3=4;

A = 70;

A" = 2cl

(2.38)

2.8 Compressibility Eects


Compressibility eects are encountered in gas ows at high velocity. In compressible ows, the
variation of gas density with pressure has a signicant impact on the ow velocity, pressure
and temperature. Compressible ows can be characterized by the value of Mach number M
dened as the ratio between the ow velocity U and the speed of sound in the gas cs given
by:
p
cs = RT
(2.39)
At Mach numbers much less than one, compressibility eects are negligible and the variation of gas density with pressure can be ignored. The previous models for uid ow and heat
transfer are valid for compressible as well as incompressible ows. It should be emphasised
that in compressible ows, the energy equation is solved with the continuity and momentum
equations. In addition, the gas density is calculated from the gas law instead of using a
constant value. Finally, the transport equation for the turbulent kinetic energy is modied
as mentioned in section 2.6.3

2.9 Numerical Aspects


The previous models for uid ow and heat transfer are solved in this study by using the commercial code Fluent. Fluent is a general purpose computer program for modeling uid ow,
heat transfer and chemical reactions. Fluent solves the conservation equations for mass, momentum, energy and chemical species using a control volume based nite dierence method.
Fluent 4.4.7 used in this study accepts body-tted structured grids.
Interpolation is accomplished in Fluent via dierent schemes. The two schemes used in
this study are the power-law scheme and the Quadratic Upwind Interpolation for Convection
Kinematics (QUICK).
The Semi-Implicit Method for Pressure-Linked Equations (SIMPLE) algorithm is used
for the pressure-velocity coupling.

Part II

PARTICLE DEPOSITION MODEL

30

Chapter 3

Particle Transport Model


Numerical calculations of multiphase ow are performed by using either the Eulerian or the
Lagrangian approach. In the beginning of this chapter, the Eulerian and the Lagrangian approaches are briey discussed. Then, the Lagrangian particle transport model is introduced.
Various forces applied on the particle are explained and the importance of these forces is
discussed. The particle equation of motion is written and the inuence of turbulence on
the particle trajectory is included. The equation for particle heat transfer is given. Finally,
particulate ow in compressors and gas turbines when low-grade fuels are used is classied
according to previous work in this eld.

3.1 Overview for Eulerian and Lagrangian Approaches


Theoretical studies of particle transport can be developed either by Eulerian approach or
by Lagrangian approach. In the Eulerian approach, both phases are considered to exist at
the same time. Particle transport is caused by concentration gradient. Particle velocity is
represented as the average over the computational cell. The consideration of a particle size
distribution requires the solution of a set of basic equations for each size class to be considered. Hence the computational eort increases with the number of size classes. The method
is however preferable for dense two-phase ows with uniform particle size.
In the Lagrangian approach, particle transport is modeled by tracking a large number of
particles through the ow eld by solving the equations of motion taking into account the
relevant forces acting on the particle. Generally, the particles are considered as points, i.e.
the nite dimension of the particles is not considered and the ow around the individual
particles is not solved. Since the number of real particles in a ow system is usually too large
for tracking all the particles, the trajectories of computational particles which represent a
number of real particles with the same properties (i.e. size, velocity and temperature) are
calculated.
The Lagrangian approach was used in this study to model the deposition process because
the previous work showed that it provides a detailed and realistic model for particle transport. In addition, it gives the complete information on particle impact at the surface required
for sticking studies. Furthermore, particle size distribution could be considered easily to represent reality.

31

3.2 Particle Motion in Fluids

32

3.2 Particle Motion in Fluids


3.2.1 Drag Force

The steady state drag is the drag force which acts on the particle in a uniform pressure eld
when there is no acceleration or deceleration of the relative motion between the particle and
the conveying uid. The drag force at various Reynolds numbers is based on the introduction
of the drag coecient CD being dened as:

CD = 1 FD 2
2 (u ; up ) Acp
where

FD
Acp
Dp
u
up

(3.1)

drag force in x-direction


cross-section area of a spherical particle (= (=4)Dp2)
particle diameter
uid velocity in x-direction
particle velocity in x-direction

The Reynolds number for a spherical particle Rep is given by:

Rep = Dp ju ; up j

(3.2)

The dependence of the drag coecient of a spherical particle on Reynolds number is shown
in gure 3.1 based on experimental investigations (Schlichting and Gersten [61]). At very
low Reynolds numbers, the drag coecient varies inversely with Reynolds number. This is
referred to as the Stokes ow and under these conditions CD = 24=Rep. With increasing
Reynolds number, the drag coecient approaches a nearly constant value. At the critical
Reynolds number, there is a sharp decrease in the drag coecient. The critical Reynolds
number represents the transition from laminar to turbulent ow past the particle.

102
101
100

p
Re
4/
=2
CD

Drag coefficient CD []

103

10-1
10-2
10-2

10-1

100

101

102

103

104

105

106

Reynolds Number Rep[]

Figure 3.1: Drag coecient for spherical particles versus Reynolds number, Schlichting and
Gersten [61]

3.2 Particle Motion in Fluids

33

There are several correlations available in the literature that t the drag coecient as a
function of Reynolds number. The general form used in this study is given by:

CD = a1 + a2=Rep + a3 =Re2p

(3.3)
where the a's are sets of constants that apply over various ranges of Rep given by Morsi and
Alexander [47].
The turbulence level of the ambient ow causes a reduction of the critical Reynolds number. With increasing turbulence intensity, the transition from laminar to turbulent boundary
layer is shifted towards smaller Reynolds number. Crowe et al. [17] showed that the critical
Reynolds number reduces by about two to three orders of magnitude by increasing the turbulence level. The particles considered in this work are very small. Therefore, the particle
Reynolds numbers are small. The Reynolds number of a 10 m particle moving in air with a
relative velocity of 10 m=s is about 10. From gure 3.1, this Reynolds number is about four
orders of magnitude smaller than the critical Reynolds number. Consequently, the eect of
the turbulence on the drag coecient was not considered in this study.
The consideration of particle shape in the calculation of particle motion was presented
by Marcus et al. [41]. They described the shape of irregular particles by using a shape
factor. The shape factor was dened as the ratio of the surface area of a sphere which has
the same volume as the particle and the surface area of the particle. They investigated the
drag coecient for various shape factors and found that when the particle Reynolds number
is smaller than about 100, the eect of the shape factor could be ignored and the assumption
of spherical particles could be used. Therefore, the calculations performed in this study were
based on the assumption of spherical particles.

3.2.2 Rarefaction Eect

This eect becomes important when the particles are very small. In such a situation, the gas
ow around the particle can not be regarded as a continuum. Instead, the particle motion is
induced by collisions of gas molecules with the particle surface. This results in a reduction
of the drag coecient. The importance of rarefaction eects may be estimated based on the
particle Knudsen number Kn dened as the ratio of the mean free path of the gas molecules
 to the particle size as:

2
Kn = D

(3.4)

The mean free path of the gas molecules can be calculated from the relation given by
Talbot et al. [71] as:

 = 2c

(3.5)

where c is the mean molecular speed given by:

c = 8RT



and R is the gas constant.

1
2

(3.6)

3.2 Particle Motion in Fluids

34

A classication of dierent ow regimes for very small particles based on the Knudsen
number was given by Crowe et al. [17] and is shown in table 3.1. In the Stokes regime
which is generally valid for very small particles, the reduction of the drag coecient may be
accounted for by a correction function, the so-called Cunningham correction factor Cu given
by (Talbot et al. [71]):

;
0
:
88
Cu = 1 + Kn 1:2 + 0:41 Exp Kn


(3.7)

The drag coecient in this case is given by:

CD = CD;Stokes
Cu

(3.8)

Figure 3.2 shows the variation of the Cunningham correction factor versus Knudsen number. It is obvious that a considerable reduction of the drag coecient occurs for Kn > 0:02.

Cunningham correction factor Cu []

Flow regime
Range of Knudsen number
Continuum ow
Kn < 10;3
Slip ow
10;3 < Kn < 0:25
Transition ow
0:25 < Kn < 10
Free molecular ow
Kn > 10
Table 3.1: Dierent regimes of rareed ows with respect to particle motion, Crowe et al.
[17]

104
103
102
101
100
10-1
10-3

10-2

10-1

100

101

102

103

Knudsen number Kn []

Figure 3.2: Cunningham correction factor versus Knudsen number, Talbot et al. [71]

3.2 Particle Motion in Fluids

35

3.2.3 Pressure Gradient Force and Unsteady Forces

The local pressure gradient in the ow gives an additional force in the direction of the pressure gradient. In addition, the acceleration or deceleration of the relative velocity between
the particle and the uid produces forces which can be divided into two parts: the virtual
mass eect and the Basset force. The virtual mass eect relates to the force required to
accelerate or decelerate the surrounding uid. The Basset term describes the force due to
the lag of boundary layer development with changing relative velocity.
Sommerfeld [64] presented an analysis for the importance of these forces. The results
indicated that these forces could be neglected for large values of the ratio of particle material
density to gas density.

3.2.4 Lift Forces

Lift forces on a particle are caused by the velocity gradient in the uid or by particle rotation.

3.2.4.1 Saman Lift Force


Particles moving in a shear layer experience a traverse lift force due to the non-uniform
relative velocity over the particle and the resulting non-uniform pressure distribution. If a
particle leads the uid motion, then the lift force is negative and the particle moves down the
velocity gradient towards the wall. Conversely, if the particle lags the uid, then the lift force
is positive and it moves up the velocity gradient away from the wall. Saman [59] analysed
this force and found the magnitude of the force FS to be:
r

FS = 1:615Dp2 kr (u ; up)
where kr is the local velocity gradient.

(3.9)

3.2.4.2 Magnus Force


Magnus force is the lift developed due to rotation of the particle. The lift is caused by the
pressure dierence between both sides of the particle resulting from the velocity dierence
due to rotation. Kallio and Reeks [31] noted that in most regions of the ow eld, the Magnus
force is not important and at least an order of magnitude smaller than the Saman force. As
a consequence, it was ignored in this study.

3.2.5 Gravity Force

Gravity force is simply the product of the particle mass and the vector representing the acceleration due to gravity. The gravity force was ignored in this study as the particles considered
were very small.

3.3 Particle Trajectory Calculation

36

3.2.6 Thermophoretic Force

When a particle exists in a ow eld with temperature gradient, another force arises on the
particle which is called thermophoretic force. This force is caused by the unequal momentum
exchange between the particle and the uid. The higher molecular velocities on one side of the
particle due to the higher temperature give rise to more momentum exchange and a resulting
force in the direction of decreasing temperature. An extensive review of thermophoresis by
Talbot et al. [71] indicated that the following equation for the thermophoretic force FT
provides the best t with experimental data over a wide range of Knudsen numbers:

(kf =kp + CtKn)


rT
FT = ;6DpCs (1 + 3C Kn
)(1
+
2
k
=k
+
2
C
Kn
)
T
m
f p
t
where

kp
rT
Cs
Ct
Cm

(3.10)

particle thermal conductivity


temperature gradient
1.17
2.18
1.14

This model is applicable when the Knudsen number Kn is smaller than 2. The particles
considered in this work were larger than the gas mean free path and therefore, this model
was appropriate.

3.2.7 Brownian Force

Brownian force is caused by random impact of the particle with agitated gas molecules. For
submicron size particles, Brownian force could be quite important. In particular, near solid
surfaces where the intensity of turbulence becomes negligibly small, Brownian force could be
an important transport mechanism.
In previous work there was a general thinking that the Brownian force can dominate the
motion of submicron particles. Recently, a clear model of the Brownian force and its eect
on particle trajectory was given by Ounis et al. [49]. They studied the Brownian dispersion
of submicron particles from a point source in a simulated turbulent channel ow eld. The
particle equation of motion including the Brownian force was studied and ensembles of 8192
particle trajectories were evaluated. The results were compared to those obtained from the
exact solution of the corresponding convective diusion equation in the absence of turbulent
uctuations. It was found in this work that the particles with diameters equal to or greater
than 0.03m were strongly aected by turbulence, even those with a distance of one wall
unit from the surface, and the Brownian force could be neglected. For smaller particles, the
Brownian force was an important transport mechanism in this region. Based on these results
and since the particles considered in this study are larger than 0.03 m, the Brownian force
was ignored.

3.3 Particle Trajectory Calculation


Fluent predicts the trajectory of the particle by integrating its equation of motion that is
written (for the x-direction) as:

3.4 Turbulent Dispersion of Particles

37

dup = 18 CD Rep (u ; u ) + F


p
x
dt pDp2 24

(3.11)

The rst term in the right hand side of this equation represents the drag force per unit
particle mass. Fx represents other forces per unit mass which can act on the particle. Fluent
optionally includes the thermophoretic force on particles in the additional force term Fx in
equation 3.11. Fluent uses the form given by Talbot et al. as shown by equation 3.10.
The drag coecient CD is calculated by using the general form given by equation 3.3.
In the present study, the drag coecient was modied by employing a user-dened subroutine to include the rarefaction eect by applying the Cunningham correction factor given by
equation 3.7.
Finally, Fluent allows additional forces to act on the particles by employing user-dened
subroutines. A user-dened subroutine was employed in this study to include the Saman
lift force on the particles as given by equation 3.9.

3.4 Turbulent Dispersion of Particles


The dispersion of particles due to turbulence in the uid phase was predicted by using the
eddy lifetime model [65]. In this model the instantaneous values of the uid velocities u, v
and w appearing in the equations of motion are given by:

u = u + u0 ;

v = v + v 0;

w = w + w0

(3.12)

where u, v and w are the uid average velocities and u0 , v 0 and w0 are the uid uctuating
velocities.
By calculating the trajectories in this manner for a sucient number of representative
particles, the random eects of turbulence on particle dispersion may be accounted for.
In the eddy lifetime model, each eddy is characterized by Gaussian distributed random
velocity uctuations u0 , v 0 and w0, and a time scale e . The values of u0 , v 0 and w0 which
prevail during the lifetime of the uid eddy that the particle is traversing are sampled by
assuming that they obey a Gaussian probability distribution such:

u0 = 

u02;

v0 = 

v 02 ;

w0 = 

w02

(3.13)

where  is a normally distributed random number, and the remainders of the right hand sides
are the local r.m.s. values of the velocity uctuations. Since the kinetic energy of turbulence
is known from turbulent ow calculations, these values of the r.m.s. uctuating components
can be obtained (assuming isotropy) as:
q

u02 =

v 02 =

w0 2 = 2k=3
p

(3.14)

3.5 Heat Transfer to or from the Particles

38

The value of the random number  is applied for the characteristic lifetime of the eddy
given by [65]:

e = 0:3 k"

(3.15)

The particle is assumed to interact with the uid phase eddy over this eddy lifetime.
When the eddy lifetime is reached, a new value of p
the instantaneous
velocity is obtained by
p
p
2
2
2
applying a new value of  . The values u, v , w and u0 , v 0 , w0 are updated whenever
migration into a neighbouring cell occurs.

3.5 Heat Transfer to or from the Particles


The calculations of the heat transfer to or from the particles included in this study considered
only heating or cooling of the particles. Phase changes of the particle are not included.
Assuming the temperature is uniform throughout the particle and radiation is unimportant,
the equation of particle temperature is given by:
p = h A (T ; T )
mpCpp dT
c p
p
dt

where:

mp
Cpp
Tp
Ap
hc

(3.16)

particle mass
particle specic heat
particle temperature
surface area of the particle
convective heat transfer coecient

The convective heat transfer coecient is evaluated using the correlation given by Fluent
[22] and Crowe et al. [17] as:

Nu = hckDp = 2:0 + 0:6Rep2 Pr 31


1

where:

(3.17)

Nu Nusselt number
Pr uid Prandtl number (Cp=kf )

3.6 Coupling between the Dispersed and the Continuous


Phases
In two-phase ows systems, the terms one-way coupling, two-way coupling and four-way coupling are often used to represent the eect of particle phase on the uid ow. In the one-way
coupling, the particle phase has no eect on the uid ow. In the two-way coupling, uid
dynamic interactions between the particles and the uid are considered. In the four-way
coupling, uid dynamic interactions between the particles and the uid in addition to the
collisions between particles are included.

3.6 Coupling between the Dispersed and the Continuous Phases

39

The volume fraction of the dispersed phase p is the volume occupied by the particles in
a unit volume. A classication of dispersed two-phase ows with regard to the importance
of interaction mechanisms was presented by Sommerfeld [64] and is shown in gure 3.3. The
gure indicates that a two-phase system may be regarded as dilute for volume fractions up to
about 4  10;4 . In this regime, the one-way coupling could be considered for p < 4  10;7 .
For higher volume fractions, two-way coupling needs to be accounted for. In the dense regime
(i.e. for p > 4  10;4 ) the four-way coupling should be included.

dilute dispersed
two-phase flow

two-way
coupling

one-way
coupling
10-8

10-7

dense dispersed
two-phase flow

10-6

10-5

four-way
coupling
10-4

10-3

10-2

10-1

100

Volume fraction p []

Figure 3.3: Regimes of dispersed two-phase ows as a function of particle volume fraction,
Sommerfeld [64]
The particle concentration produced from burning low-grade fuels in gas turbines is expected to be in the 200-300 ppmw range as found by Sethi and Partanen [62]. By using
cyclone separators, the inlet particle concentration to the turbine could be reduced by a factor of about two. The inlet particle concentration to the compressor of a gas turbine unit
is much lower than these values. Assuming that the particle density is 2000 times greater
than the gas density, the volume fraction corresponding to the particle concentration of 200
ppmw equals 10;7 . Therefore, particulate ow in either the compressor or the turbine when
low-grade fuels are used is classied as dilute ow and the inuence of the particle phase on
the uid ow can be neglected. Consequently, all the calculations performed in this study
considered only the one-way coupling.

Chapter 4

Particle-Wall Interaction
The particle-wall interaction is the part in the deposition process which determines the fraction of incident particles that remain on the surface. The particle-wall interaction considered
in this study falls into two categories. First, pure mechanical interaction in the absence of
the uid force. It aims to determine the condition at which no rebound occurs and is called
in this study the sticking process. Second, uid dynamic interaction between the uid and
the stuck particles. It studies the stability of the particles at the surface and is called the
detachment process. Dierent sticking forces and mechanisms are explained. The detachment mechanisms are also discussed and the dominant mechanism of particle detachment is
obtained.
A particle sticking model is used from the literature. The model is based on the elastic
properties of the particles and the surface under dry conditions. The sticking process under
wet conditions is not considered in this study. A particle detachment model for the stuck
particles is developed based on the critical moment theory. The gross detachment of previously built-up deposits is not included. The models employed in this study are based on the
assumption of spherical particles and smooth surfaces. Further studies are required to investigate the eect of particle shape and surface roughness on particle sticking and detachment.

4.1 Particle Sticking


Once particles arrive at the surface, deposit build-up depends on the balance of sticking forces
and removal forces acting on the particles at the surface. The sticking coecient is the mass
fraction of incident particles that stick to a surface. Previous deposition models assumed
this coecient. Fackrell et al. [21], Kladas [35], Brown et al. [13] and Ahluwalia et al. [4]
assumed perfect sticking for the particles at the surface. Huang et al. [30] and Menguturk
and Sverdrup [42] performed their deposition calculations assuming that the particles smaller
than 3 m stick to the surface upon impact and the particles larger than this diameter rebound due to their kinetic energy.
The probability of particle sticking to a surface is likely to be a function of several parameters including particle size and composition, target material properties, particle velocity,
angle of incidence and gas and surface temperatures. Ahluwalia et al. [3], Logan et al. [40]
and Anderson et al. [6] measured experimentally the sticking coecients for ash particles.
They found that the sticking coecients changed between 0.1 and 0.0003. Therefore, the assumption of perfect sticking would not lead in many situations to the actual rate of particle
deposition. The most common causes for particle sticking are (gure 4.1):
40

4.1 Particle Sticking

41

 Van der Waals force which arises from molecular interaction between solid surfaces.
 Electrostatic forces caused by charging the particles electrically in the gas stream. The

electric charge developed would play a role in the sticking process.


 Liquid bridges between the particles and the surface under wet conditions. In gas
turbines, the wet conditions are caused by the combustion of low-grade fuels which
contain alkali components. Combustion of these fuels produces alkali vapour in addition
to the ash. If the temperature in the thermal boundary layer is lower than the dew
point of the alkali vapours, alkali vapours condense and modify the sticking process.
Alkali sulphates in the ash work as a glue and increase particle sticking by the so called
liquid bridges. In compressors, wet conditions could be due to the moisture drawn with
air or by oil leakage.

Figure 4.1: Summary of the most common sticking mechanisms

4.1.1 Dominant Sticking Force

Sticking forces/particle weight []

Figure 4.2 shows the signicance of


dierent sticking mechanisms as given
109
by Berbner and Loeer [8]. The stickElectrostatic
108
ing forces are given in dimensionless
van der Waals
7
10
Liquid bridge
forms using particle weight. The g6
10
ure shows the dependence of the di105
mensionless sticking forces on particle
104
diameter. The gure indicates that as
the particle weight increases with the
103
third power of the diameter, the eect
102
of the sticking forces decreases. There101
fore, small particles stick more rmly
100
than large ones. The gure shows also
10-7
10-6
10-5
10-4
10-3
Particle
diameter
D
[m]
that the sticking due to liquid bridges
p
if they exist dominates other sticking
mechanisms. In gas turbines, particle Figure 4.2: Dimensionless sticking forces versus
sticking at the surface by liquid bridges particle diameter, Berbner and Loeer [8]
can be avoided by keeping the surface
temperature and the temperature in
the thermal boundary layer above the dew point of the alkali vapour. In compressors, oil
leakage and moisture should be avoided. Under dry conditions, the van der Waals force is
the major contribution to the particle sticking force.

4.1 Particle Sticking

42

For evaluating the van der Waals force, microscopic and macroscopic approaches were used
in the literature. The microscopic approach is based on the interactions of the individual
molecules. To estimate the magnitude of van der Waals force, the contribution of many
molecules constituting the surfaces are considered. This force can be expressed in terms of
the Hamaker constant. The van der Waals sticking force Fpo is then given by:
p
Fpo = AD
2
12z

where

(4.1)

A Hamaker constant [J ]
z separation distance between the particle and the surface [m]

The macroscopic approach deals directly with the bulk properties of the interacting bodies. When a particle collides with a surface, it actually deforms. The contact area between
the particle and the surface plays a role in particle sticking. The sticking force is calculated
based on the particle size and material properties with constants being derived from experiments. The sticking force Fpo is calculated by Soltani and Ahmadi [63] and Rimai et al. [55]
as:

Fpo = ksWA Dp

(4.2)

where ks is a constant equal to 3=4. The work of sticking WA is a constant which depends
upon the material properties of the particle and of the surface and has the units of J=m2.
This constant is obtained experimentally for some materials.

Coefficient of restitution [-]

Previous investigations for the impact and sticking are available in the
1.00
literature for spherical particles. The
coecient of restitution is dened as
0.95
the ratio of the normal rebound velocity to the normal impact velocity.
0.90
Dahneke [18] studied experimentally
0.85
the eect of particle impact velocity
on the rebound velocity for spherical
0.80
particles. Figure 4.3 shows his result
for 1.27 m diameter particles. The
0.75
gure indicates that when the particle
0
5
10
15
normal impact velocities are relatively
Impact velocity [m/s]
high, the coecient of restitution is
relatively constant. But, as these velocities decrease, the signicance of Figure 4.3: The coecient of restitution versus imthe sticking force increases and the re- pact velocity, Dahneke [18]
bound velocities drop o considerably.
Eventually, a point is reached when no
rebound occurs and the particles are captured. This velocity is called the capture velocity.

4.1 Particle Sticking

43

4.1.2 Sticking Model

The mechanism by which particle sticking


and rebounding are taking place can be illustrated by considering an ideal mass-spring system in which the spring is connected between
the mass and a massless platform (gure 4.4).
Impact is initiated when the platform contacts
a rigid, sticky surface. Following platform contact, the mass will compress the spring due to its
initial kinetic energy. This will be followed by
a rebound with all the kinetic energy restored
when the mass returns to the position it had
at the time of initial contact. At this point,
the platform will not leave the surface because
of the platform's sticking. The spring will be- Figure 4.4: Illustration for the sticking
gin to stretch and develop a tensile force. If mechanism
the strength of the sticking bond is high, the
platform will remain attached and the mass will
continue to oscillate and die out. With a lower strength, the sticking bond will fracture as
the tensile spring force builds up and the mass and spring leave the surface with velocity less
than the impact velocity.
A similar mechanism is attributed to the case of microsphere impact on a at surface. A
sticking bond is established during the impact and is fractured during rebound. The circular
contact area is composed of an inner region of compressive stresses and an outer annulus of
tensile stresses.

4.1.3 Particle Capture Velocity

Brach and Dunn [12] developed a semi-empirical model for the sticking of spherical particles.
Based on experimental data, they calculated the capture velocity below which no rebound
occurs. This model is used in the present study to calculate the particle capture velocity that
is given by:

vcr = 2DE


10=7

(4.3)

"

2
E = 0:51 5 (k13=+2 k2)
4p

2=5

(4.4)

; s )
k1 = (1E

(4.5)

(1 ;  2)
k2 = E p
p

(4.6)

4.1 Particle Sticking


where

vcr
Es
s
Ep
p
Dp
p

44

particle capture velocity, [m=s]


Young's modulus of surface material, [Pa]
Poisson's ratio of surface material, [;]
Young's modulus of particle material, [Pa]
Poisson's ratio of particle material, [;]
particle diameter, [m]
particle density, [kg=m3]

Figure 4.5 shows the capture velocity calculated for dierent particle sizes and materials
with a steel surface. The steel elastic properties were obtained from Soltani and Ahmadi [63]
as Es = 2:15  1014Pa and s = 0:3. p was considered as 0.3 while Ep was changed from
1  105Pa to 1  1010Pa. The gure indicates that as the particle size decreases, the particle
capture velocity increases. Therefore, the particle sticking probability increases. It also shows
that the Young's modulus has a great eect on the capture velocity. As the Young's modulus
increases, the capture velocity decreases.

Capture velocity vcr [m/s]

103
102

rebounding

rebounding

101
100
10-1
10-2
105

sticking
sticking
Dp = 10 m
Dp = 1m
106

107

108

109

1010

Particle Young's modulus Ep [Pa]

Figure 4.5: Particle capture velocity versus particle Young's modulus


To perform deposition calculations in this work, the particle capture velocity at the surface is calculated from the elastic properties for the particle and the surface using equation
4.3. The particle normal impact velocity is compared to the particle capture velocity. If
the particle normal impact velocity is smaller than the capture velocity, the particle sticks,
otherwise it rebounds and continues the trajectory until it leaves the domain or impacts the
surface on another place.

4.2 Particle Detachment

45

4.2 Particle Detachment


When the removal forces from the uid ow are sucient to prevent the particles from remaining on the surface, the particles are detached. Dierent mechanisms of particle detachment
are rolling, sliding and lifting. Soltani and Ahmadi [63] studied dierent mechanisms of particle detachment. They found that spherical particles are released from the surface by rolling
rather than sliding or lifting. Das et al. [19] studied also dierent detachment mechanisms
and got the same result that rolling is the dominant detachment mechanism. Therefore, the
detachment of particles from smooth surfaces is caused by the uid dynamic moment in the
viscous sublayer. Detachment occurs when this moment exceeds the moment exerted on the
particle by the sticking force.

4.2.1 Detachment Model

Figure 4.6 shows the geometric features of a spherical particle attached to a plane surface.
The critical moment theory is applied to describe the mechanism of particle detachment from
a surface. Accordingly, a particle will be detached when external force moment about the
point 'O' overcomes the resisting moment due to the sticking force, i.e.

FD ( D2p ; ) + FL a  Fpo a
where:

(4.7)

relative approach between the particle and the surface


FL uid lift force
a radius of the contact area

Figure 4.6: Geometric features of a spherical particle stuck to a smooth surface


In most cases of elastic particle sticking, is very small in comparison with Dp=2 and
could be neglected. Soltani and Ahmadi [63] found that the eect of lift force on particle
detachment is negligible. Therefore, equation 4.7 becomes:

4.2 Particle Detachment

46

FD D2p  Fpo a

(4.8)

In the viscous sublayer, the dimensionless streamwise velocity is given by:

u+ = y +

(4.9)

where the dimensionless quantities are dened as:

y + = u y

u+ = uu ;


(4.10)

For smooth surfaces, the dimensionless velocity at the center of a stuck spherical particle

u+c is given by:

Dp+
+
u =

(4.11)

uc = u Dp
u
2

(4.12)

uc = D2p u2

(4.13)

where uc is the velocity at the center of the stuck spherical particle.

The drag force acting on a spherical particle is given by:

FD = CD 12 u2c Acp f=Cu

(4.14)

where Acp is the area in the stream direction, Acp = (=4)Dp2, and f (=1.7) is the correction
factor for the wall eect given by Soltani and Ahmadi [63].
For small particles, the Reynolds number is quite small and the Stokes drag may be used.

24 ;
CD = Re
p

Rep = Dpuc

(4.15)

The Cunningham correction factor Cu is given by equation 3.7. Then, the drag force is
given by:
pu
FD = 3fD
Cu c

(4.16)

4.2 Particle Detachment

47

From equation 4.13

f D2u2
FD = 23Cu
p 

(4.17)

The radius a of the contact area can be calculated from the relation given by Soltani and
Ahmadi [63] and Rimai et al. [55] as:

3WADp2
a=
2Kc

!1
3

(4.18)

where Kc is the composite Young's modulus given by:


2 (1 ;  2 )
Kc = 34 (1 ;E s ) + E p
s
p
"

;1

(4.19)

Figure 4.7 shows the variation of the contact radius versus particle diameter for two different materials. The data for the silicon particles were obtained from Soltani and Ahmadi
[63] and are shown in table 4.1. For ash particles, the Young's modulus was measured at the
Institute of Materials Science and Testing, Vienna University of Technology. The work of
sticking for ash particles is not available. Richards et al. [56], Ahluwalia et al. [3] and Spiro
et al. [66] found that ash particles contain 8-50 % silicon. Therefore, it was assumed that the
ash particles have the same work of sticking of silicon particles. The ash material properties
used for this calculation is shown in table 4.2.

Particle contact radius a [m]

10-6

10-7

10-8

10-9

Silicon particles
Ash particles
10-10
10-8

10-7

10-6

10-5

10-4

Particle diameter Dp [m]

Figure 4.7: Variation of the contact radius with the particle diameter

4.2 Particle Detachment

48

Ep [Pa]
17:9  1010
p [;]
0.27
2
WA [J=m ]
0:039
Table 4.1: Properties of silicon particles, Soltani and Ahmadi [63]

Ep [Pa], (measured)
18:4  109
p [;], (assumed)
0.27
2
WA [J=m ], (assumed) 0:039
Table 4.2: Properties of ash particles
The sticking force Fpo is given by equation 4.2 with ks = 3=4 as:

Fpo = 43 WA Dp

(4.20)

Figure 4.8 shows the variation of the sticking force versus particle diameter using the
properties of silicon particles.

Sticking force, Fpo [N]

10-4
10-5
10-6
10-7
10-8
10-9
10-10
10-8

10-7

10-6

10-5

10-4

Particle diameter Dp [m]

Figure 4.8: Variation of the sticking force for silicon particles versus particle diameter

4.2.2 Critical Wall Shear Velocity

Substituting from equations 4.14, 4.18 and 4.20 in equation 4.8 leads to the relation for the
critical wall shear velocity uc :

CuWA WA
c = D
DpKc
p

u2

1
3

(4.21)

4.3 Properties Assumption

49

Critical wall shear velocity uc [m/s]

Therefore, the particle will be removed from the surface if the turbulent ow has a wall
friction velocity which is larger than uc . Figure 4.9 shows the variation of the critical velocity versus particle diameter for silicon particles and for ash particles. The gure shows
that the critical shear velocity needed to detach the particle increases rapidly as the particle
diameter decreases. It also shows that for the same particle size, ash particles need larger
shear velocity than silicon particles. This is caused by the elastic properties of the ash and
silicon particles. From tables 4.1 and 4.2, Young's modulus of ash particles is smaller than
that of silicon. Therefore, the contact area between ash particles and a surface is larger than
the contact area between silicon particles and a surface.

103

102

101

100

Silicon particles
Ash particles
10-1
10-8

10-7

10-6

10-5

10-4

Particle diameter Dp [m]

Figure 4.9: Variation of the critical shear velocity versus particle diameter

4.3 Properties Assumption


The sticking and detachment models presented here are based on the material properties
for the particles and the surface. These properties are: the Young's modulus and Poisson
ratios for particle and surface materials and the work of sticking between the particle and
the surface. The best results are expected from these models if the properties used are
those corresponding to the particles causing fouling of compressors or gas turbines. These
properties are not available in the literature for ash particles. The elastic properties presented
by Soltani and Ahmadi [63] and Brach and Dunn [12] for various types of particles showed that
the Poisson ratio changed from 0.27 to 0.33 while the Young's modulus changed by several
orders of magnitude. According to equations 4.4 and 4.19, the Young's modulus has the
major eect on particle sticking and detachment in this range of Poisson ratio. Therefore,
Young's modulus for ash particles was measured at the Institute of Material Science and
Testing, Vienna University of Technology. Measuring the work of sticking was not possible.
Therefore, the calculations performed in this study are based on the assumption of its value.
The assumed value is obtained from the literature for silicon particles.

Chapter 5

Deposition Model in Fluent


In the previous two chapters, the Lagrangian particle trajectory model and the sticking and
the detachment models were introduced. This chapter explains the approach employed to
build the deposition model. It includes the detailed information on the deposition model and
how it was programmed in the commercial CFD code Fluent 4.4.7.
This particular version of Fluent was used in this study because it models the process of
particle-wall interaction more realistically than other versions. The particle-wall interaction
available in Fluent 5 considers either rebounding or sticking for all the particles hitting a
surface. In Fluent 4.4.7, the condition of the particle at the surface is also determined by
a user-dened subroutine. Fluent 4.4.7 uses an index to distinguish between sticking and
rebounding. Employing the user-dened subroutine USREFL.F, it is possibile to change this
index. In the present study, the sticking and the detachment models were included in this
subroutine. The results of these models determined the index. By this way, the particle-wall
interaction was controlled by using the sticking and the detachment models. This feature is
only available in Fluent 4.4.7 and therefore, the latest versions of Fluent could not be used
here.

5.1 Overview for Deposition Calculations


The deposition calculations performed in this study are based on the ow chart given in
gure 5.1. The rst step is the solution of the ow eld for the given geometry by using
the inlet and the outlet boundary conditions and the appropriate uid ow and heat transfer
models. The particles are distributed and injected at the inlet to the geometry with velocities
and temperatures representative of those of the uid. The particle trajectories are calculated
using the Lagrangian dispersion model explained in chapter 3. The sticking and the detachment models introduced in chapter 4 are applied to the particles hitting the surface in order
to calculate the actual rate of particle deposition. These calculations are repeated as long
as the geometry of the surface is not changed signicantly by deposits. In the case that
the deposits signicantly change the geometry, the fouled surface is generated by using the
deposition distribution. The ow eld is solved for the fouled surface to take into account
the time dependent particle deposition. The calculations are continued until the required
number of operating hours is achieved.
It should be emphasised that this general approach for the model was not used in all
the calculations performed in this thesis but depended on the particular case of the problem
50

5.1 Overview for Deposition Calculations

51

studied. For instance, the sticking and the detachment models were not considered in some
calculations and the assumption of perfect sticking for the particles was used to calculate
the rate of particle transport to the surface. The detailed assumptions for each problem are
discussed as will be shown in the next chapters.

Figure 5.1: Deposition ow chart

5.2 Building the Deposition Model in Fluent

52

5.2 Building the Deposition Model in Fluent


The dispersed phase model available in Fluent 4 was used to predict particle trajectory by
solving the particle equation of motion. The commercial code was extended by user-dened
subroutines to model the deposition process. The user-dened subroutines were employed to
model the particle-wall interaction and to make necessary modications to obtain a general
model for particle deposition. Figure 5.2 shows a ow chart for the deposition model as built
in Fluent 4. The central part of the chart represents the calculations in Fluent. The boxes at
both sides are the user-dened subroutines used in the model. The user-dened subroutines
are described below.

5.2.1 Modication of the Drag Coecient

The particle trajectory calculations performed by Fluent 4 have limitations when the particles
of interest are of submicron size (< 1 m). In Fluent 4, the drag coecient for the particles
is calculated by using the correlation 3.3 with the constants given by Morsi and Alexander
[47]. For submicron particles, the drag coecient should be modied to include the rarefaction eect by using the Cunningham correction factor given by equation 3.7. Therefore, the
user-dened subroutine USERCD.F was used to include the drag correction.

5.2.2 Additional Force Components

The trajectory calculations performed in Fluent 4 do not consider the eect of Saman force.
This force was modeled and implemented in this study by using the user-dened subroutine
USRFOR.F. This subroutine allows new force terms in the particle equation of motion. The
eect of the Saman lift force was included by using equation 3.9.

5.2.3 Particle-Wall Interaction

Fluent 4 applies a unique boundary condition for all the particles hitting a surface. The
user has the choice between reect or stick. This approach was not adequate to model deposition processes because a particle might stick to the surface while another one might
rebound. Therefore, a unique boundary condition for all the particles hitting a surface does
not represent reality. Fluent allows the user to customise the particle-wall interaction via the
user-dened subroutine USREFL.F.
This subroutine was used in this study to include the sticking and the detachment models discussed in chapter 4. The particle capture velocity was calculated from the particle
properties using the sticking model while the particle impact velocity was calculated from
the trajectory model. The particle normal impact velocity was compared to the particle capture velocity in this subroutine. When the particle normal impact velocity was higher than
the capture velocity, the particle was considered to rebound, otherwise, it was considered to
stick. The mass of the stuck particles was accumulated in a variable which is called in Fluent
user-dened function. The user-dened function was visualized during postprocessing.
As discussed in chapter 4, the sticking model includes pure mechanical interaction between
the particle and the surface and can not show the eect of wall shear stress at the surface
on the stability of the particle at the surface. Therefore, a particle detachment model was

5.2 Building the Deposition Model in Fluent

53

also included in this subroutine to check the detachment condition. The wall friction velocity
was calculated from the ow eld and the critical wall shear velocity was calculated from the
particle size and the material properties. The critical wall shear velocity was compared to
the uid friction velocity. When the uid friction velocity was higher than the critical shear
velocity, the particle was considered to detach from the surface and continue the trajectory
until it impacted the surface at another place or left the domain.

5.2.4 Menu for the Sticking and the Detachment Parameters

The sticking and the detachment models used in this study are based on the elastic properties
for the particles and the surface. These properties are not included in the input list of Fluent
4. Therefore, it was necessary to dene a new menu in Fluent to include these material
properties. The user-dened subroutine USRSET2.F was used for this purpose.

5.2.5 Changing Particle Inlet Parameters

In some test cases, it was necessary to distribute the particles at the inlet with a certain
distribution. One of the case studies required a uniform particle distribution at the inlet of
a pipe. To perform this distribution, the user-dened subroutine USERC2.F was used.

5.2.6 Output Postprocessing

Some parameters aecting the rate of particle deposition are not available for postprocessing
in Fluent 4. For instance, the wall shear velocity is not a postprocessing variable in Fluent.
Therefore, its value was calculated and visualized by using the user-dened subroutine USRFN.F. This subroutine was also used to visualize the deposition results.

5.2 Building the Deposition Model in Fluent

Figure 5.2: Deposition model in Fluent

54

Part III

TEST CASES AND RESULTS

55

Chapter 6

Validity of the Transport Model


The Lagrangian particle transport model is tested in this chapter. The validity of the model is
checked by comparing the numerical results to experimental data from three sources. Firstly,
with data obtained by Liu and Agarwal [38] concerning particle transport in turbulent pipe
ow. Secondly, with the data of Parker and Lee [51] concerning particle transport to turbine
blade surfaces. Finally, the thermophoretic particle transport is investigated in laminar pipe
ow with the data of Montassier et al. [45]. All the results presented in this chapter are
based on the assumption of perfect sticking of the particles at the surface because this was
the condition in the three experiments. The sticking and the detachment models are not
considered in the calculations performed here but they are considered in the calculations
presented in chapters 7 and 8.

6.1 Parameters Aecting Particle Transport


The response time of a particle to changes in ow velocity is important in studying particle
transport. It gives indication of the eect of particle inertia on the particle motion. Using
equation 3.1, the particle equation of motion in a gas is given by (in x-direction):
p = 1 (u ; u )2  D2 C
mp du
p 4 p D
dt 2

(6.1)

Using the denition of the Reynolds number of spherical particles given in equation 3.2
and dividing through the particle mass, the equation gives:

dup = 18 CD Rep (u ; u )


p
dt pDp2 24

(6.2)

From gure 3.1 and for the limits of low Reynolds number (Stokes ow), the factor

CDRep=24 approaches unity. The other factor is the momentum response time  given by:
p Dp2
 = 18

56

(6.3)

6.2 Particle Transport in a Pipe Flow

57

The particle equation of motion can be rewritten as:

dup = 1 (u ; u )
p
dt 

(6.4)

This equation indicates that the particle velocity depends upon the momentum response
time for the particle. The momentum response time is sensitive to particle size, particle
density and uid viscosity and is often called particle relaxation time.

6.2 Particle Transport in a Pipe Flow

Liu and Agarwal [38] measured the transport of olive oil droplets of 1.4 to 21 m diameter
from turbulent air ow to the internal wall of a smooth glass tube for nominal pipe Reynolds
numbers of 10000 and 50000.

Figure 6.1: Experimental system of Liu and Agarwal [38]

6.2.1 Experimental System

Figure 6.1 shows a schematic diagram of the system used by Liu and Agarwal. The test
particles were generated using an aerosol generator at its rated output of 1:5  10;3 m3=s

6.2 Particle Transport in a Pipe Flow

58

and transported vertically upward through a nominal 32 mm diameter copper pipe and a
plenum chamber on the top. From the plenum chamber, the particles owed down a nominal 32 mm diameter copper pipe before entering the 1.02 m long, 12.7 mm inside diameter
glass deposition pipe. The glass deposition pipe was connected to the copper pipe through
a 32-12.7 mm copper reducer. Downstream of the glass deposition pipe was a lter followed
by a volumetric ow rate transducer, a regulating valve, and a suction blower. Clean air was
added at the point (B) when the air ow required was higher than 1:5  10;3 m3=s (rated
output of the aerosol generator).
The particles used in these experiments were spherical droplets of olive oil containing less
than 10 % by weight of uranine 1. Therefore, the droplets were sticking once they reached
the surface and the deposition measurements here actually measured the rate of particle
transported to the surface. The deposition pipe was divided into three sections. The central
section was used in the deposition measurement. Deposition measurement in the rst section
near the entrance was not considered because a nite length was needed for the ow to
become fully turbulent. The deposition in the third section near the exit was not considered
either due to the expansion of the ow.

6.2.2 Numerical Calculations

In the Lagrangian particle trajectory model used in this study, the eect of turbulent uctuating velocity on the particle trajectory is simulated by the eddy lifetime as explained in
chapter 3. In this model, the uctuating velocities are randomly drawn from a Gaussian
random distribution of the turbulent kinetic energy. Therefore, the turbulence model aects
the particle trajectory through the value of the turbulent kinetic energy. Consequently, it
was necessary to investigate the eect of turbulence modeling and near-wall treatment on
the rate of particle transport to the surface. Two turbulence models were investigated: the
standard k-" model and the RNG k-" model. The solution near the wall was solved with two
dierent models: the standard wall function and the two-layer zonal near-wall model. The
results were compared to the experimental data.
Two structured two-dimensional grids had been generated. A grid with about 2200 cells
was used to solve the ow eld and to calculate particle transport to the surface using turbulence models with the standard wall function. A dense grid with about 6700 cells was used
to solve the ow eld and to calculate particle transport to the surface by using the two-layer
zonal model.
The velocity inlet boundary conditions were used to satisfy the ow at the inlet cells. Constant velocities of 12 m=s and 61 m=s were considered at the inlet to satisfy the Reynolds
numbers in the experiments of 10000 and 50000, respectively. The inlet turbulence intensity
was assumed at 1 %. The turbulence length scale at the inlet was calculated from the diameter of the pipe using the relation l = 0:07D as given in the Fluent manual where D is the
diameter of the pipe.
The incompressible ow eld was calculated assuming standard air properties. The dimensionless wall distance y + at the wall-adjacent cells was examined for the requirements
of the near-wall treatments. The y + values changed in the range between 30-70 with the
standard wall function and were kept at less than 2 with the two-layer zonal model.
1

Uranine is a di-sodium derivative of uorescein

6.2 Particle Transport in a Pipe Flow

59

Samples of 10000 particles of a particular diameter were injected with velocities equal to
the uid velocity. The density of the particles was obtained from the data of Liu and Agarwal
as 920 kg=m3. The particle trajectories were calculated by solving the particle equation of
motion. The particles were assumed to stick once they had hit the surface. The number of
the transported particles to the wall of each section of the pipe was counted for each of the
turbulence models. Particle transport velocity to the pipe wall V was then calculated for the
central section using the equation given by Liu and Agarwal as:

Q ln 1
V = DL
s Pf
where:

Q
D
Ls
Pf

(6.5)

volumetric ow rate


inner diameter of the deposition pipe
length of the deposition section
the fraction penetration through the section

Pf is dened as the ratio of the number of particles leaving the section to the number of
particles entering the section.
The particle transport velocity was represented in dimensionless form using the uid
friction velocity u as:

V + = uV

(6.6)

The dimensionless particle transport velocity V + was studied against dimensionless particle relaxation time  + that is given by:
2

 + = u 

6.2.3 Comparison to Experimental data

(6.7)

Figure 6.2 shows the variation of the dimensionless particle transport velocity to the pipe
wall versus dimensionless particle relaxation time calculated using dierent turbulence models and near-wall treatments. The RNG k-" model with the two-layer zonal near-wall model
gave good agreement with the experimental data over a wide range of particle relaxation
time. The gure shows also that both models, RNG k-" and standard k-" , with the standard wall function overpredicted the transport velocity to the pipe wall for the particles
with dimensionless relaxation times smaller than 10. The agreement for the particles with
the dimensionless relaxation time greater than 10 was moderate. The agreement between
all turbulence models for large particle relaxation times was due to particle inertia which
made the details of the boundary layer not important for particle trajectories. When the
particles reached the boundary layer, they continued to move toward the surface by their
inertia. Smaller particles were highly aected by the velocity in the boundary layer and the
turbulence inside played also a role in their movement.

6.2 Particle Transport in a Pipe Flow

60

This result agrees with the previous results of Menguturk et al. [43] and Liu et al. [39].
They studied the eect of the boundary layer on particle trajectory and found that the eect
of the boundary layer for particles smaller than 10 m is important.

Dimensionless particle transport velocity V+ []

100

10-1

10-2

10-3

10-4

10-5

10-6
0.1

10

100

1000

Dimensionless particle relaxation time + []


RNG k- model with the two-layer zonal model
Standard k- model with the two-layer zonal model
RNG k- model with the standard wall function
Standard k- model with the standard wall function
Experimental data of Liu and Agarwal

Figure 6.2: Comparison between particle transport velocity to the pipe wall calculated using
dierent turbulence models and the experimental data of Liu and Agarwal [38]

6.3 Particle Transport to Turbine Blade Surfaces

61

6.3 Particle Transport to Turbine Blade Surfaces


Parker and Ryley [52] and Parker and Lee [51] studied experimentally the transport of particles to turbine blade surfaces in an open-circuit low-speed wind tunnel. This data was used
in the present study to test the Lagrangian particle trajectory model and its ability to predict
particle transport to turbine blade surfaces.

6.3.1 Experimental System

The equipment and the techniques used for these measurements is described by Parker and
Ryley [52] while the measured data can be found in Parker and Lee [51]. A schematic diagram of the wind-tunnel used in their study is shown in gure 6.3. It was constructed to
provide an air velocity of about 70 m=s at the exit from a cascade of six blades. The blade
prole is shown in gure 6.4 while the cascade parameters are given in table 6.1.

Figure 6.3: Experimental system of Parker and Lee [51]


One of the central blades was used for particle transport measurements and was removable
for examination, whilst the adjacent blade gave information on the aerodynamic performance
from 31 static pressure tappings in its surface.
The particles were generated in an atomizer unit capable of generating a cloud of drops
from a solution of uorescein. Uranine was used as a uorescent dye. The droplet cloud
entered a holding chamber in which the residence time was sucient to allow the solvent to
evaporate.
Particle transport to the blade surfaces was collected by placing aluminium adhesive foil,
which was securely earthed during testing to remove electrostatic charges. After the test
period, the tape was removed from the blade surface and cut into known and identiable
portions. Each portion was placed in distilled water and the solution was examined to measure the collected mass.

6.3 Particle Transport to Turbine Blade Surfaces

62

Figure 6.4: Blade geometry of Parker and Lee [51]


Chord length, [mm]
Pitch, [mm]
Blade height, [mm]
Aspect ratio [;]

249
127
167
0.67

Table 6.1: Parameters for the cascade of Parker and Lee [51]

6.3.2 Numerical Calculations

The ow eld through the turbine cascade was solved by using two turbulence models to
investigate the eect of turbulence modeling on particle transport to the blade surfaces and
to obtain the appropriate turbulence model. The standard k-" and the RNG k-" models were
tested. The near-wall region was solved by employing the standard wall function and the
two-layer zonal model.
Structured body tted and two-dimensional grids were generated using the preprocessor
preBFC. A grid with about 15000 cells was used to solve the ow eld and to study particle
transport to the blade surface employing turbulence models with the standard wall function.
Alternatively, a dense grid with about 66000 cells was employed to solve the ow eld using
turbulence models with the two-layer zonal model. The grids were generated based on the
grid topology shown in gure 6.5. The inlet plane was chosen at a distance of one axial chord
upstream of the blade leading edge. The outlet plane was at a distance of 0.63 axial chords
downstream of the blade trailing edge. The cyclic boundaries were used to represent the
periodic ow. The blade was represented as an internal region in the computational domain
and the grid points were mapped as shown in gure 6.5. The grids near the blade leading
edge are shown in gures 6.6 and 6.7.
The incompressible air ow eld was solved using the inlet boundary conditions given in
the experiments of Parker and Ryley [52] and as shown in table 6.2. The inlet turbulence
characteristic length l was calculated from the relation given in Fluent manual [22] for duct
ows as: l = 0:07L, where the characteristic dimension L was chosen as the blade pitch.

6.3 Particle Transport to Turbine Blade Surfaces

63

Figure 6.5: The grid topology used to generate the grid for the blade of Parker and Lee

Air inlet velocity, [m=s]


Incidence angle, [o]
Air inlet pressure, [Pa]
Air inlet turbulent intensity, [%]
Turbulence characteristic length, [m]

11.0
0.0

1:013  105
2
0.009

Table 6.2: Inlet boundary conditions of the cascade of Parker and Ryley [52]

6.3 Particle Transport to Turbine Blade Surfaces

64

Figure 6.6: The grid used to solve the ow eld using turbulence models with the standard
wall function

Figure 6.7: The grid used to solve the ow eld using turbulence models with the two-layer
zonal model

6.3 Particle Transport to Turbine Blade Surfaces

65

The surface velocity was deduced from the calculated static pressure at the blade surface.
Figure 6.8 shows the distribution of the surface velocity for the ow eld solved using the
standard k-" model with the standard wall function and compares it with the experimental
data of Parker and Ryley [52]. The gure shows that the acceleration of the air on the blade
pressure surface was smooth while the air accelerated on the blade suction surface to its
maximum value at about 60% of the chord, then the ow decelerated. The gure indicates
that the numerical calculation agreed well with the experimental data.

70

Surface velocity [m/s]

60
50

Suction surface

40
Pressure surface

30
20

Numerical calculation
Experimental data

10
0
0.0

0.2

0.4

0.6

0.8

1.0

1.2

Distance along the blade surface (s/c) [-]

Figure 6.8: Comparison between the numerical calculation and the experimental data for the
surface velocity distribution
Using the statistical software R [74], a sample of 33000 particles was generated with the
same particle mean diameter and standard deviation measured during the experiments. The
particles were uniformally distributed over the blade pitch and injected with velocities equal
to the uid inlet velocity. The particle trajectories were calculated by solving the equation
of motion. Each particle trajectory was calculated 25 times to account for the eect of turbulence. This number was chosen after performing the calculations several times and nding
that with greater numbers, the results did not change signicantly.
The main disadvantage of Lagrangian particle trajectory models is the long computational
time because the particles are calculated individually. To include the eect of turbulent uctuating velocity on the particle trajectory, each particle has to be calculated several times
which increases the computational time further. A way to reduce the computational time is
to nd the zone along the pitch within which the particles have a high probability to reach
the surface. This zone represents the whole blade pitch with a small error. Figure 6.9 shows
the results of several runs when the particles calculated were in the zones of 20, 30, 50 and
100% of the blade pitch. The gure indicates that the calculations performed for only 30%
of the particles injected within 30% of the blade pitch (15% from the stagnation point) can
account for all particles.

Percentage impacted particles [%]

6.3 Particle Transport to Turbine Blade Surfaces

66

3.2

3.1

3.0

2.9

Dp= 0.05 m
Dp= 1 m

2.8
0

20

40

60

80

100

Calculation zone in percent of blade pitch [%]

Figure 6.9: Calculated particles for dierent injected zones


The particles were considered to stick once they had hit the surface and their mass was
accumulated in the corresponding cell. The particle transport to the turbine blade surface
per unit area was then calculated as a percentage of the particles passing through the blade
passage in a zone of unit blade pitch and unit height.

6.3.3 Comparison to Experimental Data

The distributions of the mass fractions of particles transported to the turbine blade surfaces
calculated using dierent turbulence models for a sample with the mean diameter of 0:098 m
are shown in gure 6.10. The gure indicates that the best result was obtained by using the
RNG k-" model with the two-layer zonal model.
The standard k-" model with both near-wall treatments predicted higher transport rates
than the experimental data on the blade pressure surface. This has to do with the fact that
the standard k-" model is a fully turbulent model and simulates the boundary layer as a
turbulent boundary layer. As a consequence, it generated high turbulent kinetic energy and
increased the rate of particle transport to the blade pressure surface. This result agrees with
the previous numerical calculations on heat transfer in turbine blades by Orszag et al. [48].
They found that the standard k-" model overpredicted the heat transfer coecient by a factor
of about 2 on the blade pressure surface.
The adverse pressure at midchord of the blade suction surface shown in gure 6.8 caused
an increase in the thickness of the turbulent boundary layer. The thick turbulent boundary layer contained high turbulent kinetic energy and therefore high uctuating velocities.
Consequently, the rate of particle transport to the surface was high in this region. All the turbulence models used in this study predicted the increase in particle transport in the rear part
of the blade suction surface. The RNG k-" model was the best one to predict the quantity
in this region. The turbulence models which used the standard wall function underpredicted
the rate of particle transport. The reason could be the adverse pressure at midchord of the
blade suction surface which is counter to the assumption of an equilibrium turbulent boundary layer used with the standard wall function.

6.3 Particle Transport to Turbine Blade Surfaces

67

Percentage transported /m2

1.25

Pressure surface
1.00
0.75
0.50
0.25
0.00
1.25

Percentage transported/m2

Suction surface
1.00
0.75
0.50
0.25
0.00
0.0

0.2

0.4

0.6

0.8

1.0

1.2

Distance along blade surface (s/c) []

Experimental data of Parker and Lee


RNG k- model with the two-layer zonal model
Standard k- model with the two-layer zonal model
RNG k- model with the standard wall function
Standard k- model with the standard wall function

Figure 6.10: Comparison between the numerical results using dierent turbulence models
and the experimental data for particle transport to turbine blade surfaces
An analysis was done for the particle impact velocity which is important in the particle
sticking model (as the particle impact velocity decreases, the probability of particle sticking
increases). The distributions of the mean particle impact velocity on the blade suction surface
are shown in gure 6.11. The gure indicates that both turbulence models with the standard
wall function gave high particle impact velocities near the blade leading edge caused by the
turbulent kinetic energy generated in the boundary layer.

6.3 Particle Transport to Turbine Blade Surfaces

68

Mean impact velocity [m/s]

20

15

10

0
0.0

0.2

0.4

0.6

0.8

1.0

1.2

Distance along blade suction surface (s/c) [-]


RNG k- model with the two-layer zonal model
Standard k- model with the two-layer zonal model
RNG k- model with the standard wall function
Standard k- model with the standard wall function

Figure 6.11: Comparison between particle mean impact velocities calculated using dierent
turbulence models
Figure 6.12 shows a comparison between two numerical results and the experimental data
of Parker and Lee for two particle mean diameters. The numerical results were obtained
in this study using the RNG k-" model with the two-layer zonal model. The gure shows
also the numerical results obtained by Menguturk and Sverdrup [42] who calculated particle
transport to turbine blade surfaces using an Eulerian model based on turbulent diusion of
particles.
The gure shows very good agreement between the present results and the experimental
data for both particle mean diameters on the blade suction surface. The model of Menguturk
and Sverdrup predicted particle transport for the particle mean diameter of 0:098 m but underpredicted particle transport to the blade surface for the particle mean diameter of 0:13 m.
The present numerically obtained rates of particle transport to the blade pressure surface
near the blade trailing edge were lower than experimental data for the particle mean diameter
of 0:098 m. Although particle transport to the blade pressure surface is mainly caused by
particle inertia, which increases with increasing particle size, the experimental data showed
high rates of particle transport for the small particle mean diameter. The only explanation
for this result is the eect of secondary ow. As the blade aspect ratio of this geometry is
relatively small (0.67), the passage vortex may aect the particle movements at midspan near
the blade trailing edge. The direction of the passage vortex is from the pressure surface to
the suction surface near endwall and from the suction surface to the pressure surface near
midspan (gure 6.13). If the passage vortex exists near midspan, it carries the small particles
to the blade pressure surface and results in an increase of the rate of particle transport in
this region. The passage vortex is unable to carry relatively large particles (0:13 m). The
two-dimensional numerical calculations considered in this study can not show the eect of the
secondary ow and as a consequence there exists a small discrepancy on the blade pressure
surface for particle with the mean diameter of 0:098 m. This discrepancy may be eliminated
if three-dimensional calculations are performed.

6.3 Particle Transport to Turbine Blade Surfaces

69

Percentage transported /m2

1.5

Particle mean diameter


0.098m

suction surface

1.0

0.5

pressure surface

0.0

Percentage transported /m2

3.0

Particle mean diameter


0.13m

suction surface

2.0

pressure surface
1.0

0.0
-1.2

-0.8

-0.4

0.0

0.4

0.8

1.2

Distance along blade surface (s/c) [-]

Experimental data of Parker and Lee


Present numerical results using the RNG k- model
with the two-layer zonal model
Numerical results of Menguturk and Sverdrup

Figure 6.12: Comparison between the numerical results and the experimental data of particle
transport for two dierent mean diameters

Figure 6.13: Secondary ows in turbine blade passages

6.3 Particle Transport to Turbine Blade Surfaces

70

The discrepancy between the numerical calculations and the experimental data on the
blade leading edge was caused by numerical artifacts due to the junctions in the computational
grid. This problem existed also in the previous numerical results of Orszag et al. [48] when
they calculated the heat transfer coecient on turbine blades. To obtain better agreement in
this region, careful consideration for the grid quality at the junctions is required. The other
possibility is to use O-grids to avoid the junctions on the blade surface.

6.3.4 Grid Sensitivity Study

The best results on particle transport were obtained using the RNG k-" model with the twolayer zonal model. The grid used in the calculations performed so far was relatively dense
for two reasons: First, the grid size was chosen to guarantee the grid independent solution
for both ow eld and particle trajectories. Second, using the near-wall modeling, a dense
grid was required near the wall. The number of grid points was increased in the direction
normal to the wall. To keep the cell aspect ratio within limits, the number of grid points
in the streamwise direction had to be increased as well. In this part of the study, the grid
sensitivity had been investigated to conrm the grid independent solution. In addition, it
aimed to nd the appropriate grid size which can satisfy the independent solution and at the
same time save computational time.
Recently, Hildebrandt and Fottner [29] studied the inuence of grid renement on the ow
eld inside a highly loaded turbine cascade. They modeled the ow eld using low Reynolds
number models all the way to the viscous sublayer as performed in the present study. They
solved the three-dimensional ow eld using a grid which had the plane grid points of 11700
points. They used a block structured grid with an O-grid around the blade surface. They
found that the grid independent solution could not be obtained with coarser grids.
The limitations in the preBFC and Fluent 4 make the O-grid type and the block structured grid unavailable. Therefore, it was decided to increase the number of the grid points
by 50 % of that used by Hildebrandt and Fottner and use a grid with a total number of cells
of about 17000 to investigate the grid sensitivity. Figure 6.14 shows the grid used for this
study.
The air ow eld was solved using the inlet boundary conditions given above in table
6.2. The particles were injected at the inlet and the trajectories were calculated. The rate of
particle transport to the blade surface was calculated as explained before. Figure 6.15 shows
a comparison between the rate of particle transport using the grid of 17000 cells and the
previous grid with 66000 cells. Particles transported to the blade leading edge were omitted
in this gure because of the problem of the junctions explained in the previous section. The
gure indicates that the solution obtained with the coarser grid agreed with that obtained
with the ne grid. The gure indicates also that a grid size of about 17000 cells could predict the rate of particle transport with acceptable error. Consequently, all the calculations
performed in the next chapters considered a grid size of about 17000 cells per turbine blade
passage.

6.3 Particle Transport to Turbine Blade Surfaces

71

Figure 6.14: A grid of about 17000 cells used in the sensitivity study

Percentage transported /m

1.5
1.2

Experimental Data
Calculation with a grid of 17000 cells
Calculation with a grid of 66000 cells

0.9
0.6
0.3
0.0
-1.2

-0.8

-0.4

0.0

0.4

0.8

1.2

Distance along blade surface (s/c) [-]

Figure 6.15: Comparison between the numerical calculations using two grids and the experimental data

6.4 Thermophoretic Particle Transport

72

6.4 Thermophoretic Particle Transport


Thermophoresis is the term describing the phenomenon that small particles suspended in a
gas move in the direction of decreasing temperature. The thermophoretic force is generated
according to the fact that gas molecules colliding with particles deliver greater momentum if
they come form hot gas zones than if they come from cold zones. The result of this unequal
transfer of momentum is a force applied to the particle in the direction opposite to the temperature gradient.

6.4.1 Thermophoretic Particle Transport in Laminar Flow

The experimental data of Montassier et al. [45, 46] was used in this study to investigate the
eect of the thermophoretic force on particle transport in laminar pipe ow.

6.4.1.1 Experimental System


Montassier et al. measured particle transport to the pipe wall in a temperature gradient along
its length for dierent sizes of particles (from 0:05 m to 8 m in diameter). The experimental
system they used is shown in gure 6.16. The particles and the puried compressed air entered
a tube surrounded by a heating element. The wall of the tube was brought to a temperature
of 373 K . When the gas left the tube it was at a uniform temperature of 373 K measured
continuously by a temperature probe and had a parabolic velocity prole. The gas and the
particles then arrived at a cooler. The cooler functioned as a heat exchanger. The wall of
the test tube (20 mm in diameter, 0.6 m long) was kept at a constant temperature (293 K
or 283 K ) by circulation of water. The temperature dierence created in this way between
the tube wall and the hot gas resulted in thermophoretic transport of some particles to the
wall. The remaining particles were collected by a high eciency lter at the cooler outlet.

Figure 6.16: Experimental system of Montassier et al. [45]

6.4 Thermophoretic Particle Transport

73

These experiments were performed with a laminar ow regime for a ow rate of

2  10;4kg=s and a pipe Reynolds number of about 700. After each test, the tube containing
the transported particles was removed. The mass distribution of the particles transported to
the wall was determined by cutting the test tube into 12 sections each of 50 mm length. The
collected mass on each section was measured.

6.4.1.2 Numerical Calculations

Initially, two-dimensional calculations were performed as they are simple and the computations are not time consuming. A two-dimension grid of about 12300 cells was used to model
the non-isothermal ow eld in the test tube shown in gure 6.16. A parabolic velocity
boundary condition was used to represent the inlet velocity. The temperature distribution
at the inlet was uniform. The boundary conditions are given in table 6.3. The air ow eld
was calculated by solving the governing equations for laminar ow consisting of continuity,
momentum and energy equations.
A sample of particles with diameter of 0:1 m was uniformally distributed at the pipe
inlet and injected with velocities and temperatures equal to those of the uid. The particle
properties were obtained from the data of Montassier et al. [45] as given in table 6.4. The
particle trajectories were calculated allowing the heat transfer between the particles and the
uid to take place. To calculate the total mass of particles transported to the wall, the particles were assumed to stick once they hit the surface.
Inlet mean velocity [m=s] 0:54
Inlet temperature [K ]
373
Wall temperature [K ]
293
Table 6.3: Inlet boundary conditions for the two-dimensional ow eld calculations prior
thermophoretic particle transport
Inlet particle temperature [K ]
373
3
Particle density [kg=m ]
1500
Particle thermal conductivity [W=mK ] 0.43
Table 6.4: Particle inlet boundary conditions for the two-dimensional thermophoretic particle
transport
The two-dimensional thermophoretic particle transport showed very bad agreement with
the experimental data. This suggested that the bad agreement was caused by two eects:
 The experimental set-up consisted of two main parts: the heating tube where the
air and the particles were heated; and the cooling (test) tube where the particles were
transported to the surface by thermophoresis and measured. The assumption of uniform
particle concentration at the inlet to the test tube was not correct because heating the
air and the particles in the gas heater aects with thermophoretic force on the particles.
The thermophoretic force moves the particles away from the surface. Therefore, the
concentration near the tube center is expected to be higher than the concentration near
the tube surface.

6.4 Thermophoretic Particle Transport

74

 The experimental set-up shown in gure 6.16 contained a bend between the gas heater

and the test tube. The bend could aect particle trajectories by the secondary ow
and change the particle distribution at the inlet to the test tube.
As a consequence, three-dimensional calculations were necessary to include these two
eects which could not be simulated using two-dimensional calculations. Montassier et al.
[45, 46] provided only information on the test tube but they did not provide any information
on the dimensions of the heater or the bend upstream of the test tube. The secondary ow
generated depends upon the dimension of the bend and its eect on the particle trajectories
depends on the distance between the bend and the test tube. Therefore, it was necessary
to assume these dimensions. It was assumed that the diameter of the heating tube and the
diameter of the bend were equal to the diameter of the test tube. The radius of the bend
was assumed as 0:3 m. It was assumed also that the bend was positioned at a distance of
0:2 m upstream of the test tube (gure 6.17).

Figure 6.17: Assumed dimensions of the bend


A three-dimensional grid with about 147000 cells was used to solve the ow eld in the
heater and cooler (test) tubes. Figure 6.18 shows the grid at the entrance to the test tube.
The grid represented only half of the pipe and the symmetrical boundary conditions were
applied to represent the symmetry of the ow.
The three-dimensional non-isothermal ow was solved assuming parabolic velocity inlet
boundary conditions with ambient air inlet temperature to the heater. The wall of the heater
tube was kept at 373 K . The boundary condition for the cooler had a constant temperature of 293 K or 283 K corresponding to the experiments. Two assumptions for the change
from heater-wall temperature to cooler-wall temperature were investigated. First, a step
change from heater-wall temperature to cooler-wall temperature was used (case A). Second,
the temperature boundary conditions allowed the wall temperature to vary linearly from the
heater to the cooler over a distance of 44 mm measured from the inlet of the test tube (case
B). Although these assumptions did not represent the temperature distribution between the
heater and the cooler, they permitted to perform the calculations without solving the heat
conduction in the tube.

6.4 Thermophoretic Particle Transport

75

Figure 6.18: The grid at the entrance to the test tube


1.5

Velocity [m/s]

Velocity distribution at section Y-Y


Velocity distribution at section X-X
1.0

0.5

0.0
0.0

0.2

0.4

0.6

0.8

1.0

Dimensionless distance along pipe diameter [-]

Figure 6.19: Calculated velocity at the inlet to the test tube


The assumed distance between the bend and the test tube was not long enough to attain
a fully developed pipe ow. Therefore, it was necessary to investigate the velocity prole at
the inlet to the test tube. Figure 6.19 shows the calculated velocity proles at the inlet to
the test tube.
Figure 6.20 shows the secondary ow at the inlet to the test tube. The gure indicates
that it had a signicant eect.
Samples of about 44000 particles were injected at the inlet to the heater with velocities
and temperatures equal to the uid inlet velocity and temperature respectively. Each sample
consisted of particles of the same size as those measured in the experiments. Particle distribution at the inlet was assumed to be uniform. The particle trajectories were calculated
allowing the heat transfer between the particles and the uid to take place. The mass of
transported particles to the wall of the test tube was calculated for each particle size.
Figure 6.21 shows the trajectories of some particles inside the test tube viewed in the axial
direction to the tube. The gure indicates that the secondary ow aected the trajectories
for the assumed bend dimensions.

6.4 Thermophoretic Particle Transport

Figure 6.20: The secondary ow eld at the inlet to the test tube

Figure 6.21: Particle trajectories in the test tube

76

6.4 Thermophoretic Particle Transport

77

6.4.1.3 Comparison to Experimental Data


Figure 6.22 shows a comparison between the three-dimensional numerical calculations and
the experimental data for the particles transported to the wall of the test tube sections for
the particle size of 0:1 m. Two numerical calculations are given: case A and case B. The
gure indicates that the assumption of linear distribution (case B) gave better agreement
with the experimental data than the step change assumption (case A). The gure shows that
particle transport to the surface with the step change assumption was high in the rst section which was caused by the sudden decrease in temperature resulting in a high temperature
gradient and, therefore, in a large thermophoretic force. As a consequence, more particles
were transported to the surface by thermophoresis.

Particles transported [%]

2.5

Numerical calculation, case A


Numerical calculation, case B
Experimental data

2.0
1.5

Dp = 0.1 m

1.0
0.5
0.0
0

15

30

45

60

75

Distance from the inlet [mm]

Figure 6.22: Comparison between the numerical calculations and the experimental data for
the thermophoretic transport of 0:1 m particles
The cumulative eciency is the ratio between the mass of the particles transported to
the wall to the mass of the injected particles. Figure 6.23 shows the cumulative eciency
along the pipe. The agreement between the numerical results and the experimental data is
good. The discrepancy might be reduced if the complete dimensions for the experimental
set-up were given and by performing the heat conduction calculations in the pipe.
Figure 6.24 shows the comparison between the numerical calculations using the assumption given in case B and the experimental data of Montassier et al. for the thermophoretic
particle transport to the wall of the test tube for dierent particle sizes ranging from 0:05 m
to 8 m. The agreement is very good for the particle sizes of 0:1 m, 4 m and 8 m but
moderate for the particle sizes 0:05 m, 0:2 m, and 1 m. The results indicate that the
eect of thermophoretic force increases as the particle size decreases. Considering all circumstances of the experiments, the result indicate that the thermophoretic force as modeled by
Talbot et al. [71] can predict thermophoretic particle transport in laminar pipe ows.

6.4 Thermophoretic Particle Transport

78

Cumulative efficiency [%]

Dp = 0.1 m

Numerical calculation
Experimental data
0
0

15

30

45

60

75

Distance from the inlet [mm]

Figure 6.23: Comparison between the numerical calculations using the assumption given in
case B and the experimental data for the cumulative thermophoretic transport of 0:1 m
particles

Cumulative efficiency [%]

20

Experimental data
Numerical calculations
15

10

wall temperature
= 293 K
wall temperature
= 283 K

0
0.1

10

Particle diameter Dp [m]

Figure 6.24: Comparison between the numerical calculations using the assumption given in
case B and the experimental data for the thermophoretic transport of dierent size particles

6.4 Thermophoretic Particle Transport

79

6.4.2 Thermophoretic Particle Transport in Turbulent Flow

Although thermophoretic particle transport in turbulent ow had been studied experimentally and theoretically in the previous work, there is no clear trend about the eect of thermophoresis on particle transport. In one hand, Owan [50] and Ryley and Davies [57] measured
particle transport to turbine blade surfaces and found that the rate of particle transport was
reduced by 30-90% by heating the blades. In the other hand, Fackrell et al. [21] developed a
model for particle transport to turbine blade surface. Their results showed that thermophoresis had little eect in the turbulent parts of the ow because turbulent diusion dominates.
Huang et al. [30] evaluated the transport rate of ne particles in coal-red gas turbines. In
their particle transport model, Brownian diusion, turbulent diusion, thermophoresis and
inertial impaction were taken into account. The numerical results of Huang et al. which
included the eect of thermophoresis showed no signicant change from those without the
eect of thermophoresis.
Recent numerical calculations of He and Ahmadi [28] for thermophoretic particle transport of 0:1 m particles at Reynolds number of 7000 showed that in turbulent ows, when
the particles are not suciently close to the wall, the eect of thermophoresis is overwhelmed
by the turbulent dispersion.
Greeneld and Quarini [25] used the relation for the thermophoretic force given by Talbot
et al. [71] in their calculations to test the eect of thermophoresis on particle transport in a
pipe ow. The Reynolds number in the pipe was 15000 and the particles considered had a
diameter of 0:8 m. Including the thermophoretic force, the numerical calculations failed to
predict any change in rate of particle transport from that without thermophoresis.
Romay et al. [58] showed in their thermophoretic results in a pipe ow at Re = 9600
that the turbulent eddy impaction mechanism was acting in addition to the thermophoresis.
Their experimental data at this Reynolds number showed that the thermophoretic transport
was signicantly higher than the theoretical predictions using the relation of Talbot et al.
[71]. The measured data was about 2 times greater than the theoretical particle transport
for particle diameter ranging from 0.1 to 0:7 m. They concluded that the theory for the
thermophoretic particle transport given by Talbot et al. underpredicts the rate of particle
transport and the dierence between the theory and the data widens with increasing particle
Reynolds number. They concluded also that a more detailed theoretical model needs to be
developed to demonstrate how turbulent eddies aect the rate of particle transport to the
wall by thermophoresis.
The calculations performed in the present study on the eect of thermophoresis on particle
transport to turbine blade surfaces used the relation of Talbot et al.. They failed to predict
the decrease in the rate of particle transport when the blade is heated as found experimentally
by Ryley and Davies [57] and Owen [50]. This could be attributed to the following:

 The previous work showed that the relation given by Talbot et al. needs modication
to account for the eect of turbulence in the calculation of thermophoretic particle
transport. To the author's knowledge, the validity of their relation was not investigated
at high Reynolds numbers. Further work is required in this eld.
 The k-" type models, which were used here to solve the ow eld, depend upon the
assumption of isotropic turbulence. In the particle trajectory model, the eect of turbulence was included using the eddy lifetime model which calculated the uctuating

6.4 Thermophoretic Particle Transport

80

velocity from the turbulent kinetic energy assuming isotropic turbulence. As a consequence, the reconstructed uctuating part of the ow eld was isotropic. This feature
is a deciency of the k-" type turbulence models which becomes signicant in regions
where the turbulence structure is anisotropic. At the near wall region, the assumption
of isotropic turbulence is not appropriate. In this region the temperature gradient is
high and the thermophoretic force can signicantly aect the particle motion. As a
consequence, to accurately predict particle dispersion by thermophoresis, anisotropy
based turbulence models should be used.

Chapter 7

Investigation for the Sticking Model


In the preceding chapter, the particle transport model was tested with experimental data
assuming perfect sticking of the particles at the surface. In the calculations performed in
this chapter, the assumption of perfect sticking was not considered. The sticking model was
included to calculate the sticking coecient dened as the mass fraction of incident ash that
sticks to a deposition target. The sticking model was tuned using experimental data to permit particle sticking calculations at dierent surface temperatures.

7.1 Assumptions
Once ash particles arrive at the surface, deposit build-up depends on the balance of the sticking forces and the removal forces acting on the particles at the surface. The most common
causes for particle sticking are van der Waals force, electrostatic force and liquid bridges
between the particles and the surface under wet conditions. Combustion of fuels which contain alkali components produces vapours of alkali-metal sulphates. If the temperature in the
thermal boundary layer is lower than their dew points, alkali vapours condense and enhance
the sticking process. Alkali sulphates in the ash work as a glue and increase ash particle
sticking. Under dry conditions, the van der Waals force is the major contribution to the
particle sticking force.
Simulation of the sticking process including chemical vapour condensation and deposition with phase changes is complicated and needs complete information on the fuel and the
ash analysis. The sticking mechanism studied in this work considers the case when the ash
particles are frozen or semi-molten. The mechanism of alkali vapour condensation when the
temperature is lower than the dew point is not considered.
The sticking model used here calculates the capture velocity of the particle at the surface.
The capture velocity depends upon the size of the particle and the material properties of the
particle and the surface. One of the problems encountered when calculating the capture
velocity is the detail information on the material properties of the particle and the surface.
These properties ideally should be based on the micro-material and not on the bulk material.
Unfortunately, these properties are not available. Furthermore, to study the eect of design
and operating conditions such as surface temperature on particle sticking, the dependence of
the material properties on temperature is required.
These parameters namely the Young's modulus of particle material and the Young's
modulus of surface material were obtained in this study by tting experimental data. The
81

7.2 Experimental System

82

experimental data provided the conditions expected in gas turbines. Previous experimental
work of Richards et al. [56], Ahluwalia et al. [3], Wenglarz and Fox [75, 76] and Cohn
[16] showed that the sticking coecient is a strong function of surface temperature. The
experimental data of Richards et al. was appropriate to perform the sticking calculations.
Therefore, it was used in this study. The process of choosing the particle and the surface
properties to obtain the same sticking coecient measured in the experiments is called in
this chapter tuning the sticking model.

7.2 Experimental System


Richards et al. measured the sticking coecient as a function of operating conditions for a
series of coals burned in a combustor especially developed for this purpose. At the exit of
the combustor, the combustion products were accelerated through a 3.2 mm diameter nozzle,
creating a jet with a mean velocity of 300 m=s (gure 7.1). This velocity has the same order
of magnitude as the velocity expected in the rst stage of a gas turbine. The high-velocity
jet was directed perpendicularly to a deposition target located 6 mm below the nozzle exit.
The target was a removable 12.7 mm diameter platinum disk. By cooling the target, its
temperature was changed in the range between 1056 K and 1251 K . The sticking fraction
was determined by replacing the target with an open-faced lter holder. A vacuum pump
was used to draw the gas through a lter. The sticking fraction was calculated from the ratio
of the ash stuck to the target versus the ash which was collected on the lter during the same
time interval.

Figure 7.1: Experimental set-up of Richards et al. [56]

7.3 Numerical Calculations

83

7.3 Numerical Calculations


A structured two-dimensional grid with about 3700 cells was used to solve the ow eld and
particle trajectories between the nozzle and the target. The inlet boundary conditions satised the inlet velocity and temperature measured during the experiments. The inlet boundary
conditions are shown in table 7.1.
Inlet velocity, [m=s]
300
Inlet temperature, [K ]
1573
Target temperature, [K ] 1056-1251
Table 7.1: Inlet boundary conditions for the experiments of Richards et al. [56]

The Mach number at the exit from the nozzle is given by M = U= RT , where U is
the ow velocity, is the ratio of specic heats, T is the uid temperature and R is the gas
constant. At this temperature, thepratio of specic heats equals to 1.3. Therefore, the Mach
number at the exit equals to 300= 1:3  287  1573 = 0:39. As a consequence, the ow was
considered to be incompressible and the error created by this assumption is not high. The
ow eld was solved by using the RNG k-" model with the two-layer zonal model for the
near-wall region. It was assumed that the gas had air properties. The energy equation was
also solved to obtain the temperature eld in the domain between the nozzle and the target.
The dimensionless wall distance in the wall adjacent cells was kept at less than 5.

Fraction of total mass [%]

A sample of particles satisfying the


particle size distribution measured in the
30
experiments was generated and injected.
Figure 7.2 shows the size distribution
20
measured by Richards et al. The particles were injected from a point source at
the center of the nozzle with velocities and
10
temperatures equal to the uid velocity
and temperature, respectively. The particle trajectories were calculated allowing
0
0
5
10
15
20
heat transfer between the particles and
Particle size D [m]
the surrounding uid to take place. The
heat transfer calculation assumed that the
particle was at a uniform temperature Figure 7.2: Particle size distribution measured
throughout. This was an appropriate as- by Richards et al. [56]
sumption because the particles considered
in this work were very small.
p

For the particles arrived at the target surface, the sticking model was applied to calculate
the particle capture velocity. The main diculty in applying the sticking model was the right
choice of particle properties such as Young's modulus and Poisson's ratio. In addition, to
investigate the eect of surface temperature on particle sticking, the dependence of particle
properties on the temperature was required, which was not easy to quantify. Consequently,
the sticking model was tuned by choosing the dependence of Young's modulus on temperature to t the experimental data.

7.4 Tuning the Sticking Model

84

The tendency of ash particles to stick to a surface depends on the ash fusion temperatures. The ash fusion temperatures are measured by pressing the ash into small cone-shape.
The ash is placed in a furnace in which the temperature is slowly rising. The shape of the
cone is observed and four temperatures are reported. The initial deformation temperature is
the temperature at which the rst rounding of the apex of the cone occurs. The softening
temperature is the temperature at which the cone has fused down into a semi-spherical lump
in which the height is equal to the width at the base. The hemispherical temperature is
the temperature at which the height is one-half of the width of the base. Finally, the uid
temperature is determined when the mass has spread out in a nearly at layer [10].
The chosen formulas for the dependence of particle Young's modulus on temperature were
based on the assumption that the particles lost their elastic properties at the softening temperature. Therefore, the particle Young's modulus vanished at this temperature. Equation
4.3 leads to a particle capture velocity of innity which means that the particles stick to the
surface with any impact velocity.

7.4 Tuning the Sticking Model


Young's modulus [GPa]

Figure 7.3 shows the variation of


Young's modulus versus temperature as
500
given by Ralls et al. [54]. The materials shown in the gure represent two of
Al O
400
the ash components. Aluminium oxide
300
Al2O3 can represent 15-25 % by weight
Mg
O
of the ash as found by Richards et al.
200
[56], Anderson et al. [5], Logan et al.
100
[40], Ahluwalia et al. [3] and Spiro et
al. [66]. The ash contains also magne0
0
400
800
1200
1600
sium oxide MgO with dierent percentTemperature [K]
ages as found by these authors. Figure 7.3
shows that for these materials, Young's
modulus decreases with temperature but Figure 7.3: Variation of Young's modulus with
is not markedly temperature dependent. temperature, Ralls et al. [54]
The temperature dependence of other ash
components could not be found in the literature.
2

In the present work, the ash particles were considered to have a constant Young's modulus
when their temperatures were much lower than the softening temperature. For this purpose,
the particle Young's modulus at room temperature of wood ash particles was measured at
the Institute of Materials Science and Testing, Vienna University of Technology. Table 7.2
shows the results obtained. The table indicates that the measured values changed over a
wide range. This could be attributed to the particle size. Ash particles existed in a spectrum
of size distribution. The analysis of the particles might change depending on the particle
size. Ahluwalia et al. [3] measured the density of ash particles in function of their diameters.
They found that the large particles had low densities because they consisted mostly of char
remaining from the incomplete combustion.

7.4 Tuning the Sticking Model

85

Young's Modulus [GPa] 18.4  5.73


Table 7.2: Measured Young's modulus for ash particles
The particle sticking properties were
used to represent the surface properties
10
assuming that there existed already a
monolayer built on the deposition surface.
10
The monolayer had the same properties as
ash particles but at the surface tempera10
ture. This assumption was appropriate
because Richards et al. performed their
10
experiments for a time long enough to
build a monolayer on the surface. Fig10
0
5
10
15
20
25
ure 7.4 shows the sticking coecient verTime [min]
sus time as measured by Richards et al.
The gure indicates that the rate of particle deposition was time-dependent at the Figure 7.4: Sticking coecient versus time,
beginning of the experiments because the Richards et al. [56]
surface properties were changing by deposits. After 5 minutes, the sticking coecient was relatively constant. This period was the time taken to build a monolayer on
the surface. As the monolayer was built, the surface properties were not changing and the
sticking coecient was time-independent.
Sticking coefficient [-]

-1

-2

-3

-4

The Poisson ratios for the particles and the surface were assumed to be temperatureindependent. A constant value as 0.3 was used in the calculations for both the particles and
the surface.
The particle trajectories were calculated. The particle velocities and temperatures at
impact on the surface were calculated as well. The particle impact temperatures were used
to determine the Young's modulus for the particles as postulated above. Then, the capture
velocities were calculated. The particles were considered to stick if their normal impact velocities were smaller than the capture velocity. The mass of the stuck particles was accumulated
and the sticking coecient was calculated as the ratio between the mass of the stuck particles
to the total mass of particles transported to the surface. The dependence of particle Young's
modulus on temperature was assumed using a rst order and a third order relation. The temperature at which the Young's modulus starts to change from the measured value at room
temperature was also used as a tuning parameter. Figure 7.5 shows the results obtained using
both relations. The gure indicates that the third order approximation for the dependence of
particle Young's modulus on temperature gave better tting for the experimental data over
a wide range of surface temperature than the rst order approximation.
The third order relation was considered (for Tp>1100 K ) as:

Ep = 120(1589 ; Tp)3

(7.1)

7.4 Tuning the Sticking Model

86

The rst order relation was considered (for Tp>1300 K ) as:

Ep = 3  105(1589 ; Tp)
where:

(7.2)

Ep particle Young's modulus [Pa]


Tp particle temperature [K ]

Sticking coefficient [-]

0.06
0.05

Experimental data
Third order relation
First order relation

0.04
0.03
0.02
0.01
0.00
1000

1050

1100

1150

1200

1250

1300

Surface temperature [K]

Figure 7.5: Comparison between the numerical calculations and the experimental data for
the sticking coecient

Chapter 8

Ash Particle Deposition on Turbine


Blades
In the previous chapters, the particle transport model and the particle sticking model were
investigated using experimental data. The deposition model is employed in this chapter to
predict the deposition rate on turbine blade surfaces when low-grade fuel is used. Experimental studies for particle deposition on turbine blade surfaces are almost non-existent except for
the isothermal results of Parker and Lee [51]. They measured particle deposition on turbine
blades by placing an adhesive tape on the blade surface. Therefore, their study was based on
the assumption of perfect sticking of the particles at the surface and they measured actually
particle transport to the surface. As a consequence, their data could be used to investigate
the particle transport model as done before in this work.
The study presented in this chapter is aimed at predicting the performance of turbine
blades after some operating hours when low-grade fuels are used and to compare it with
the performance of the clean blades. Unfortunately, corresponding experiments supporting
this study are not available. Although the deposition results obtained in this chapter do
not include comparisons to experimental data, it is expected that they represent the real
case in gas turbines because the elements of the deposition model were investigated in the
previous chapters and showed good agreements with the data. In addition, all the data used
to perform these calculations such as inlet particle concentration and size distribution were
obtained from the literature for the conditions expected in gas turbines when low-grade fuels
are used.
In this chapter, a turbine blade prole was chosen for which ow eld measurements are
available. The ow eld was solved numerically and compared to the experimental data. The
prole loss of the clean blade was expressed using the total pressure coecient and compared
to the experimental data of the clean blade. The deposition rate on the blade was calculated
in three time intervals. The total pressure coecient of the fouled blade was calculated and
compared to that of the clean blade.

8.1 Selection of the Blade Cascade


The well documented experimental data for the transonic turbine inlet guide vane performed
in the von Karman Institute (VKI) was used in this study with the ow eld measurements
for the clean blade given by Arts et al. [7]. This data was selected because it satises the
following requirements:
87

8.1 Selection of the Blade Cascade

88

 It represents stator blade measurements. Previous work showed that the problem of

particle deposition is more serious in stator blades than in rotor blades. Ragland et al.
[53] studied the problem of ash deposition in a wood-red gas turbine. They measured
the deposition thickness on the nozzle blades and on the rotor blades. They found that
the deposition thickness on the nozzles was greater than that on the rotors.
 These experiments satisfy the velocity used in modern gas turbines.
 They provide the inlet and the outlet boundary conditions given as pressures and
temperatures and the velocity is developed according to the blade shape.
 Finally, the distributions of the surface velocity and the downstream total pressure are
available. Therefore, the eect of deposits on these parameters could be investigated.

x/cax= 1.487

Downstream pressure measurement

x/cax= 1.433

Freestream inlet pressure and temperature

The geometry of the VKI prole is shown in gure 8.1 and the cascade parameters are
given in table 8.1. In the experiments, the blades were mounted in a linear cascade made of
5 proles, i.e. 4 passages. The central blade was instrumented for measurements.

Figure 8.1: Geometry of the VKI prole, Arts et al. [7]

Chord length, [mm] 67.647


Pitch, [mm]
57.5
Throat, [mm]
14.93
Table 8.1: Cascade parameters for the VKI prole, Arts et al. [7]

8.2 Performance of the Clean Blades

89

Arts et al. [7] measured the freestream total pressure and temperature, static pressure
and turbulence intensity 55 mm (x=cax=-1.487) upstream of the leading edge plane. They
installed also wall static pressure tappings downstream of the cascade in a plane parallel to
the trailing edge plane and located 16 mm (x=cax=1.433). Blade velocity distributions were
obtained from 27 static pressure measurements performed along the central blade prole and
referred to the upstream total pressure. The downstream total pressure was obtained by
using a fast traversing mechanism transporting a Pitot probe over 2 pitches.

8.2 Performance of the Clean Blades


This part aims to study the aerodynamic performance of the VKI transonic inlet turbine
guide vane. The ow eld through the cascade was numerically solved and compared to the
experimental data. The downstream total pressure coecient was used to evaluate the loss
through the cascade. The pressure coecient was calculated and compared to the experimental data. A grid sensitivity study was performed to verify that the ow eld calculated
is grid independent.

8.2.1 Solving the Flow Field


8.2.1.1 Computational Grid

One of the most demanding conditions for an accurate CFD solution is the selection and the
generation of the computational grid. Boyle and Ameri [11] studied the eect of the grid
orthogonality on the prediction of the ow eld by employing dierent grid geometries. They
used the VKI cascade with the data of Arts et al. [7] as performed in the present study. They
found that the principle eect of dierent grid geometries was in the pressure distribution
behind the vane. Therefore, orthogonality of the grid lines was carefully considered in the
present work to accurately predict the prole loss.
For structured grids, three dierent types of meshes are currently in use. The H-type
grid is appropriate for thin blades. However, H-grids suer from a lack of mesh orthogonality
in regions with large contour deections. C-type and O-type grids oer a much better ow
eld discretization. The C-type grid could not be generated by the pre-processor used in this
study (preBFC). In addition, the O-type grid with cyclic boundary conditions could not be
used in Fluent 4. Furthermore, block structured grid is not available. These limitations on
grid generation by preBFC and Fluent 4 make the grid generation dicult.
By choosing the grid topology shown in gure 8.2, a structured body-tted and twodimensional H-type grid was generated. The computational domain and the internal region
are shown in gure 8.3. The outer edges of the domain were mapped to the inlet boundary,
the outlet boundary and to the cyclic boundaries as shown in gure 8.3. The blade surface
was mapped as the interior region. The corner points of the interior region was chosen on
the blade surface in order to satisfy a high grid quality in almost all the domain. Figure 8.2
shows that one of the corners was chosen on the blade suction surface to increase the number
of the grid points on the blade suction surface to compensate for the dierence in length
between the pressure surface and the suction surface.

8.2 Performance of the Clean Blades

Figure 8.2: VKI grid topology

Figure 8.3: Computational domain and the internal region

Figure 8.4: VKI computational grid

90

8.2 Performance of the Clean Blades

91

Figure 8.4 shows the grid generated for the turbine blade prole used in this study. It
had a total number of cells of about 19000. The grid had the inlet boundary 1.35 cax ahead
of the blade leading edge and the downstream boundary at 1.16 cax behind the blade trailing
edge. The cyclic boundary conditions were used to represent the periodic ow.
Figure 8.5 shows the grid near the blade leading edge and gure 8.6 shows the grid near
the blade trailing edge. These gures indicate that the grid topology used in this study gave
a reasonable mesh around the blade. The distance between the wall and the adjacent cell
was chosen to satisfy the limits of the dimensionless wall distance y + .

Figure 8.5: Computational grid near


the blade leading edge

Figure 8.6: Computational grid near


the blade trailing edge

8.2.1.2 Boundary Conditions


The pressure boundary conditions were used to dene the uid condition at the inlet and
at the exit of the computational domain. The total pressure was used at the inlet while the
static pressure was used at the exit. Since the Mach number in this case study is high, the
ow was considered as compressible ow. Therefore, the total pressure was calculated from
the following relation using the standard air properties:

po = 1 + ; 1 M 2  ;1
p
2

where:

po total pressure
p static pressure
M Mach number

(8.1)

8.2 Performance of the Clean Blades

92

Table 8.2 shows the boundary conditions at the inlet and at the exit of the cascade as
given by Arts et al. [7] for two dierent isentropic exit Mach numbers Mis . The turbulence
characteristic length l was calculated from the relation given in the Fluent manual [22] for
duct ows as l = 0:07L, where L is a characteristic dimension of the duct and was chosen as
the blade pitch.
Isentropic exit Mach number, [;]
0.875 1.02
Total inlet pressure, [Pa]
147500 159600
Total inlet temperature, [K ]
420
420
Free stream turbulence, [%]
1
1
Turbulence characteristic length, [m] 0.004 0.004
Static outlet pressure, [Pa]
89600 82300
Wall temperature, [K ]
298
298
Incidence angle, [o]
0
0
Table 8.2: VKI inlet boundary conditions

8.2.1.3 Numerical Calculations

The two-dimensional compressible ow eld was solved in Fluent by using the RNG k-"
model with the two-layer zonal model for the near-wall region. The ow eld was calculated
for two isentropic exit Mach numbers 0.875 and 1.02 by applying the boundary conditions
given in table 8.2. The Quadratic Upwind Interpolation (QUICK) was used as discretization
scheme to provide higher order accuracy. Fluent [22] noted that this interpolation scheme
is appropriate when the ow is turning and/or is not aligned with the grid lines as in the
present case. Kim et al. [34] modeled turbulent ow over complex geometries by using
dierent discretization schemes. Their results indicated that the QUICK scheme was able to
predict the details of the ow eld which were observed in the experiments and could not
be predicted by other schemes. The values of y + at the wall adjacent cells were checked for
the near-wall modeling. The wall adjacent cells were inside the viscous sublayer with the
dimensionless wall distance y + < 5.

8.2.1.4 Comparison with Experimental Data


Figures 8.7 and 8.8 show the Mach number contours calculated for the two isentropic exit
Mach numbers mentioned above. Figure 8.8 shows that a normal shock was predicted along
the rear part of the suction side when the isentropic exit Mach number was 1.02. During the
experiment, a similar normal shock was observed at the same location.
The surface Mach number was calculated from the static pressure at the blade surface.
Figures 8.9 and 8.10 show comparisons between the numerical calculations and the experimental data of Arts et al. [7]. The gures indicate that very good agreements were obtained
between the numerical calculations and the data for both isentropic exit Mach numbers. The
gures also show that for the lower exit Mach number, the velocity distribution was rather
at with a peak adverse pressure gradient starting at s=c  0:7. For the higher exit Mach
number, the ow accelerated up to s=c  1:0. A shock was then observed. The pressure
distributions along the pressure side varied smoothly.

8.2 Performance of the Clean Blades

93

9.61E-01
9.29E-01
8.97E-01
8.65E-01
8.33E-01
8.01E-01
7.69E-01
7.37E-01
7.05E-01
6.73E-01
6.41E-01
6.09E-01
5.77E-01
5.45E-01
5.13E-01
4.81E-01
4.49E-01
4.17E-01
3.85E-01
3.53E-01
3.20E-01
2.88E-01
2.56E-01
2.24E-01
1.92E-01
1.60E-01
1.28E-01
9.61E-02
6.41E-02
3.20E-02
0.00E+00
Y
Z

Mach No.
Max = 9.614E-01 Min = 0.000E+00

Dec 05 2000
Fluent 4.47
Fluent Inc.

Figure 8.7: Mach number contours for the isentropic exit Mach number of 0.875
1.21E+00
1.17E+00
1.13E+00
1.09E+00
1.05E+00
1.01E+00
9.65E-01
9.25E-01
8.85E-01
8.44E-01
8.04E-01
7.64E-01
7.24E-01
6.84E-01
6.43E-01
6.03E-01
5.63E-01
5.23E-01
4.83E-01
4.42E-01
4.02E-01
3.62E-01
3.22E-01
2.81E-01
2.41E-01
2.01E-01
1.61E-01
1.21E-01
8.04E-02
4.02E-02
0.00E+00
Y
Z

Mach No.
Max = 1.206E+00 Min = 0.000E+00

Dec 05 2000
Fluent 4.47
Fluent Inc.

Figure 8.8: Mach number contours for the isentropic exit Mach number of 1.02

8.2 Performance of the Clean Blades

94

Isentropic Mach number [-]

1.50

Isentropic exit Mach number = 0.875

1.25
1.00
0.75
0.50
0.25
0.00
0.0

Experimental data
Numerical calculation
0.2

0.4

0.6

0.8

1.0

1.2

1.4

Distance along the blade surface (s/c) [-]

Figure 8.9: Comparison between the numerical calculation and experimental data for the
surface Mach number distribution, M2is = 0:875

Isentropic Mach number [-]

1.75

Isentropic exit Mach number = 1.02

1.50
1.25
1.00
0.75
0.50
0.25
0.00
0.0

Experimental data
Numerical calculation
0.2

0.4

0.6

0.8

1.0

1.2

1.4

Distance along blade surface (s/c) [-]

Figure 8.10: Comparison between the numerical calculation and experimental data for the
surface Mach number distribution, M2is = 1:02

8.2 Performance of the Clean Blades

95

8.2.2 Calculation of the Losses

Arts et al. [7] measured also the pitchwise variation of the total pressure at 0.433 cax behind
the vane by using a traversing mechanism transporting a Pitot probe over two blade spacings.

8.2.2.1 Downstream Total Pressure Coecient

The two-dimensional compressible ow eld was solved using the RNG k-" turbulence model
with the inlet and outlet boundary conditions given by Arts et al. and as shown in table 8.3.
Isentropic exit Mach number, [;]
0.84
Total inlet pressure, [Pa]
184900
Total inlet temperature, [K ]
409.2
Free stream turbulence, [%]
1
Turbulence characteristic length, [m] 0.004
Static outlet pressure, [Pa]
116500
Wall temperature, [K ]
298
Incidence angle, [o ]
0
Table 8.3: VKI boundary conditions for loss calculations

8.2.2.2 Grid Sensitivity Study


Previous calculations of Boyle and Ameri [11] for the same cascade showed that the grid shape
and the number of cells had signicant eects on the downstream total pressure coecient.
Therefore, two grids were used to conrm the grid independent solution: the previous grid
with about 19000 cells and a ne grid with about 27000 cells. The downstream total pressure
coecient based on the exit velocity was used to conrm the grid independent solution. The
total pressure coecient C was dened as

C = p0o:25; Upo21
2 2

where:

(8.2)

po1 total inlet pressure


po2 total exit pressure
2 isentropic exit density

and U2 is the exit velocity calculated from the isentropic exit Mach number and isentropic
exit temperature.
Figure 8.11 shows the comparison between the total pressure coecient calculated using
both grids and the experimental data. For clarity of presentation, the location of minimum
pressure was taken as the origin. The gure shows that the numerical calculations agree well
with the experimental data. The comparison between the calculations performed using both
grids conrmed that the solution was grid independent.

8.3 Temperature Eects on Particle Transport and Sticking

96

Total pressure coefficient [-]

0.15
Experimental data
Numerical calculation (19000 cells)
Numerical calculation (27000 cells)

0.10
0.05
0.00
-0.05
-0.10
-0.15
-0.20
-0.25
-0.4

ss

ps

0.0

0.4

0.8

1.2

1.6

Distance along blade pitch y/S[-]

Figure 8.11: Comparison between the numerical calculations and experimental data for the
pitchwise variation of the total pressure coecient

8.3 Temperature Eects on Particle Transport and Sticking


Combustion of low-grade fuels in gas turbines produces large amounts of ash. Using cooled
blades in this case would have a disadvantage and an advantage. The temperature dierence
between the hot gas and the blades increases ash particle transport to the blade surfaces by
thermophoresis. On the other hand, it reduces the particle-sticking coecient by solidifying
molten phases in the ash particles. As shown in section 6.4, thermophoresis increases the
transport of submicron particles and particles of few microns.

8.3.1 Temperature Eect on Particle Transport

The isotropy based turbulence models used in this study to solve the ow eld prevented
performing the calculations for the eect of thermophoresis on ash particle transport to turbine blade surfaces. In addition, these models are fully turbulent models. Using these models
does not lead to accurate prediction of the thermophoretic particle transport in the laminar
ow regions.

8.3.2 Temperature Eect on Particle Sticking

The eect of temperature dierence between the hot gas and the blade surface on ash particle sticking was studied and simulated here to obtain the sticking coecient over the blade
surface. The eect of particle size and material properties on the sticking coecient was
investigated.

8.3.2.1 Flow Field Numerical Calculations


The VKI transonic inlet guide vane was used to study the eect of surface temperature on
the sticking coecient on turbine blades. The compressible ow eld was numerically solved

8.3 Temperature Eects on Particle Transport and Sticking

97

for the design isentropic exit Mach number of 0.875 with the same inlet and outlet pressure
boundary conditions given in table 8.2. The inlet temperature was chosen as 1573 K . This
temperature is similar to the temperature expected in the rst stage of a gas turbine. The
ow eld was solved for two dierent constant surface temperatures TW , 1473 K and 1273 K .
Although the assumption of constant surface temperature does not consider the real case in
a gas turbine, it gives some indication on how the sticking coecient changes with particle
size and material. The inlet turbulence parameters were considered as given in table 8.2.

8.3.2.2 Parameters Aecting the Sticking Coecient


The thermal response time relates the time required for a particle to respond to the changes
in temperature in the carrier uid. Therefore, it is important in studying the eect of surface
temperature on the sticking coecient. Assuming the temperature is uniform throughout
the particle and radiation is unimportant, the equation of particle temperature is:
p
mpCpp dT
dt = hcAp (T ; Tp )

(8.3)

Nu = hckDp = 2:0 + 0:6Rep2 Pr 31

(8.4)

dTp = Nu 12kf (T ; T )
p
dt
2 pCppDp2

(8.5)

The convective heat transfer coecient is evaluated using the correlation given by Fluent
[22] and Crowe et al. [17] as:
1

Therefore

For low Reynolds numbers, the ratio Nu=2 approaches unity. The second term in the
right hand side represents the thermal response time T dened as:

or

 C D2
T = p12ppk p
f

(8.6)

pCppDp2
T = 12
Cp Pr

(8.7)

Thus the equation for particle temperature becomes:

dTp = 1 (T ; T )
p
dt T

(8.8)

8.3 Temperature Eects on Particle Transport and Sticking

98

This equation indicates that particle temperature and, therefore, the sticking probability
depend upon the thermal response time. The thermal response time itself is sensitive to the
particle size, density and the specic heats of the particle material and the continuous phase.

8.3.2.3 Sticking Coecient Calculations

Samples of about 8200 particles of an assumed uniform diameter were injected with velocities and temperatures equal to the uid velocity and temperature, respectively. The particles
were uniformally distributed over the blade pitch. Dierent combinations between ash density and specic heat were assumed for the particle properties. Ahluwalia et al. [3] measured
ash particle density from coal combustion. They found that the ash density changed from
about 500 kg=m3 to 1700 kg=m3 depending upon the particle size. The specic heat of the
ash was considered in the range between the specic heat of clay (about 700 J=kgK ) and
the specic heat of coal (about 1400 J=kgK ). This data was obtained from Chapman [14].
Table 8.4 shows the particle properties assumed in this study.
Case Particle density Ash specic heat Particle size
[kg=m3]
[J=kgK ]
[m]
1
1700
710
0.1-10
2
500
710
0.1-20
3
1700
1420
0.1-10
Table 8.4: Dierent particle properties used to calculate the sticking coecient
The particle trajectories were calculated allowing heat transfer between the particles and
the uid to take place. The particle impact temperature at the surface was used to calculate
the particle Young's modulus from equation 7.1 and the particle capture velocity from equation 4.3. Then the sticking condition was applied to the particle at the surface. The average
sticking coecient on the blade surface was calculated as the ratio between the total stuck
mass to the total transported mass to the blade surface. The average sticking coecient is
drawn versus the particle size for dierent ash properties in gures 8.12 to 8.14.
The gures indicate that the sticking coecient for very small particles is almost 1. The
gures show also that the sticking coecient decreases as the particle size increases for all
ash properties. The gures indicate that blade cooling could reduce the sticking coecient
up to about an order of magnitude.
Figure 8.15 shows the average sticking coecient drawn versus the particle thermal response time. The gure shows also the data when tted into universal curves at constant
surface temperatures. Three regions could be distinguished from these distributions.

 Perfect sticking region. This region represents the particles with very small thermal

response times. The particles in this region are very small and carry small kinetic
energy. Therefore, their capture velocities are high. Cooling the blade surface in this
case does not aect the sticking coecient signicantly.
 Cooling aected-region. The particles in this region impact on the surface with velocities near to the capture velocities. Cooling the blade surface in this case freezes molten

8.3 Temperature Eects on Particle Transport and Sticking

99

Average sticking coefficient [-]

100

10-1

10-2

p= 1700 kg/m3,
Cpp = 710 J/kg K
Wall temperature = 1473 K
Wall temperature = 1273 K

10-3
0.01

0.1

10

100

Particle diameter [m]

Figure 8.12: Sticking coecient versus particle diameter, case 1


Average sticking coefficient [-]

100

10-1

p= 500 kg/m3,
Cpp = 710 J/kg K
10-2
Wall temperature = 1473 K
Wall temperature = 1273 K
10-3
0.01

0.1

10

100

Particle diameter [m]

Figure 8.13: Sticking coecient versus particle diameter, case 2


Average sticking coefficient [-]

100

10-1

p= 1700 kg/m3,
Cpp = 1420 J/kg K
10-2
Wall temperature = 1473 K
Wall temperature = 1273 K
10-3
0.01

0.1

10

100

Particle diameter [m]

Figure 8.14: Sticking coecient versus particle diameter, case 3

8.3 Temperature Eects on Particle Transport and Sticking

100

phases in the particles and increases the Young's modulus of the particles. From gure
4.5, the capture velocity decreases with increasing Young's modulus. Consequently,
the probability of particle sticking decreases. The value of the average sticking coecient depends upon particle impact temperature and therefore on the particle thermal
response time.
 Cooling unaected-region. In this region, the particles are large and therefore, the
capture velocities are small. The impact velocities are higher than the capture velocities for both surface temperatures considered in this study. Therefore, the sticking
probabilities are small.
The gure indicates also that the sticking coecients could be reduced considerably if
the particle thermal response time is in the range from 2 s to 300 s.
Average sticking coefficient [-]

100

TW = 1473 K
10-1

TW = 1273 K
co

10-2

oli

ng

p= 1700 kg/m , Cpp= 710J/kg K


3

p= 500 kg/m , Cpp= 710J/kg K


3

p= 1700 kg/m3, Cpp= 1420J/kg K


10-3
10-8

10-7

10-6

10-5

10-4

10-3

Thermal response time [s]

Figure 8.15: Universal distributions for the sticking coecient on the blade surface

8.3.3 Resultant Deposition Rate

The calculations performed in section 8.3.2 for the sticking coecient showed that the particles with very small thermal response times had sticking coecients close to 1. These
particles were submicron particles. It was found in chapter 6 that thermophoresis increases
particle transport for submicron particles. As a consequence, it is expected that using cooled
blades when the particles are in the submicron size range would increase particle transport
by thermophoresis while it would not reduce the sticking coecient signicantly. Therefore,
the deposition rate would increase. On the other hand, the results obtained in section 8.3.2
showed that the sticking coecient for supermicron particles could be reduced considerably
by using cooled blades. The reduction of the sticking coecient not only depends upon the
particle size but also on the specic heats of both the ash and the carrier uid. The eect
of thermophoresis on particle transport of supermicron particles is small. Consequently, the
deposition rate would decrease with blade cooling. In actual applications, the particles exist
in a spectrum. Therefore, it is expected that the resultant deposition rate depends upon the
particle size distribution and the mass fraction of submicron and supermicron particles.

8.4 Ash Particle Deposition on Turbine Blades

101

8.4 Ash Particle Deposition on Turbine Blades


In this part, ash particle deposition on turbine blade surfaces was predicted by using the
deposition model which includes the transport model, the sticking model and the detachment
model. The VKI transonic inlet guide vane was used in this study. The numerical deposition
calculations performed in this section were based on the following assumptions:

 All the particles had the same material properties.


 The ash particles had the same value of Young's modulus as measured at the Institute

of Materials Science and Testing, Vienna University of Technology when their temperatures were much lower than the softening temperature. At high temperatures close
to the softening temperature, the ash particle Young's modulus was calculated from
the relation obtained by tuning of the sticking model using the experimental data of
Richards et al. [56].
 The surface sticking parameters were those of the particles. This assumption was
appropriate because the time required to build a monolayer on the blade surface is very
short in comparison to the time required to build a deposition layer which signicantly
changes the ow eld.
 The temperature of the particles at the surface were lower than the softening temperature of the ash. Therefore, the ash particles were frozen or semi-molten at the surface.
Sticking of molten ash particles forming slag was not considered. In gas turbines, slagging should be avoided because it increases the deposition rate very much and produces
deposition layers which could not be removed easily.
 The heat transfer from the hot gas to the blade surface was not aected by the deposition layer.

8.4.1 Flow Field Numerical Calculation

The compressible ow eld was solved using the same pressure boundary conditions mentioned above for the isentropic exit Mach number of 0.875. The inlet temperature was chosen
as 1500 K to represent a typical inlet gas temperature in gas turbines. In addition, it was
close to the gas temperature of the experiments of Richards et al. [56], which had been used
in this study to tune the sticking model as shown in the previous chapter.
The surface temperature boundary condition was obtained from a typical distribution at
the midheight of a cooled blade designed to operate with gas temperature of 1500 K as given
by Cohen et al. [15]. The temperature versus the dimensionless axial distance x=cax was
determined from the typical temperature distribution and was applied on the VKI transonic
guide vane as shown in gure 8.16. Table 8.5 summarizes the boundary conditions used in
this study.

8.4 Ash Particle Deposition on Turbine Blades


Y

102

cax

cax

X
X

c
(a)

(b)

Figure 8.16: (a) The VKI transonic inlet guide vane, (b) Typical temperature distribution at
midheight of a cooled blade designed to operate with a gas temperature of 1500 K , Cohen
et al. [15]
Isentropic exit Mach number, [;]
0.875
Total inlet pressure, [Pa]
147500
Total inlet temperature, [K ]
1500
Free stream turbulence, [%]
1
Turbulence characteristic length, [m]
0.004
Static outlet pressure, [Pa]
89600
Wall temperature, [K ]
1000-1300
Incidence angle, [o ]
0
Table 8.5: Flow eld boundary conditions prior to deposition calculations

8.4.2 Particle Specications and Inlet Boundary Conditions

The ash particle size distribution at the inlet to the blades depends upon the type of the fuel,
the combustion temperature and the cleaning technique used to reduce particulate loading.
Sethi and Partanen [62] characterized the products of combustion in an atmospheric pressure, wood-red combustor. They analysed the particulates and determined the particle size
distribution. They found that the ash particles were in the size range 0.5-10 m. They found
also that the mean particle size changed according to the combustion temperature.
Logan et al. [40] and Richards et al. [56] studied ash particle deposition from coal fuels.
They used an electrically heated tube furnace to burn coal and to provide deposition conditions representative of coal-red gas turbines. They found that the ash particles were in the
size range 1-20 m, with the peak in the mass distribution at approximately 5 m.
Kang et al. [32] reported that ash particle size distribution with peaks at 8 m and 0.25
m was produced in pulverized coal combustion.

8.4 Ash Particle Deposition on Turbine Blades

103

Whitlow et al. [78] studied the deposition from combustion gases produced by residual
oil. They observed that particles 4 ; 8 m in diameter predominated in the gas stream with
some fraction in the 0:1 ; 4 m and 8 ; 12 m ranges.

Capture efficiency [%]

Hot gas clean-up devices are used to reduce particulate loading at the inlet to the
100
turbine. Figure 8.17 shows typical particulate collection eciencies for a conventional
80
cyclone and for a high eciency cyclone
as documented by Sethi and Partanen [62].
Both devices can eciently remove the par60
ticles in the size range of > 5-10 m. The
Conventional cyclone
High efficiency cyclone
conventional cyclone can remove about 70%
40
of the 5 m particles while the high eciency
0
10
20
30
cyclone can remove about 90% of these partiParticle size [m]
cles. Therefore, using the cyclone separators,
it is expected that the majority of the parti- Figure 8.17: Particulate collection eciencies
cles after the cyclone will be smaller than 5 of cyclones, Sethi and Partanen [62]
m. Sethi and Partanen [62] concluded that
the uncontrolled dust loading is in the 200300 ppmw range. They concluded also that the conventional or the high eciency cyclones
can reduce the dust loading by a factor of about two.
Ash density is one of the important factors aecting particle movement by the inertia
force. Ash density depends upon the analysis of the ash. Ahluwalia et al. [3] measured ash
particle densities versus diameter. They found that the ash density changed from 1000-1700
kg=m3 for the particles smaller than 4 m.
Based on the previous data from the literature, the particle specications at the inlet to
the cascade were assumed in this study as given in table 8.6.
Inlet particle size distribution, [m] 0.25-5
Inlet particle concentration, [ppmw] 100
Inlet particle density, [kg=m3]
1700
Ash spec heat, [kJ=kgK ]
710
Table 8.6: Assumed particle specications at the inlet to the cascade

8.4.3 Numerical Deposition Calculations

Hamrick and Schiefelbein [26] presented the results of a program to develop biomass as an
alternative fuel for gas turbines. The results indicated that after approximately 26 hours
of operation, the gas turbine power output started to decline due to deposition and the gas
turbine was cleaned. They used wood-fuel which contained 1.7% ash. This ash content is
relatively high in comparison to other wood types. Ragland et al. [53] tested the wood-red
gas turbines. They used wood which contained 0.8 % ash. Sethi and Partanen [62] developed
a wood-red gas turbine. The wood types which they burned contained 0.4-0.6% ash.

8.4 Ash Particle Deposition on Turbine Blades

104

The rate of particle deposition not only depends upon the ash content but also on the
total plant performance including plant design and operating conditions.
Based on the observations of Hamrick and Schiefelbein, it was decided to calculate particle deposition in 36 operating hours and to predict the blade performance after this period.
This period was divided into three calculation periods of 12 hours each. The 12 hours period
was chosen to solve the time-dependent particle deposition and to limit the computational
time within acceptable limits. It was assumed that the ow eld was time-independent during the calculation period. After each calculation period, the deposition distribution on the
blade surface was calculated. The fouled blade prole was predicted and the ow eld was
solved. The calculations were then repeated for the next period.
A sample of about 13000 particles was used with the peak in the mass distribution at 2
m. The particles were uniformally distributed over the blade pitch and were injected with
velocities and temperatures equal to the uid inlet velocity and temperature, respectively.
The particle trajectories were calculated allowing heat transfer between the particles and
the uid to take place. Each particle was calculated 100 times to account for the eect of
turbulence on the particle trajectory and to build a deposition layer on the surface.
The sticking model was applied to the impacted particles at the surface by calculating
the particle capture velocity. The detachment model was applied to the particles stuck at the
surface by comparing the critical wall shear velocity with the uid wall friction velocity. If a
particle was detached from the surface it continued the trajectory until it left the domain or
hit the surface at another location.

8.4.4 Deposition Results

In the earlier calculations performed in this study, time was not important because the deposition was represented as fractions of the particles injected at the inlet. In this part of the
study, the time was considered to obtain the deposition thickness after a certain operating
period.
Using the specied inlet particle concentration and air ow rate per unit blade height,
the particle mass ow rate per unit blade height was obtained. The total mass of particles injected in the 12 hours period in then calculated. The mass of the sample of particles injected
to the cascade was calculated as well. The deposition distribution on the VKI transonic inlet
guide vane was determined after 12 hours by calculating the deposition from the injected
sample of particles and simply multiplying the results by a factor corresponding to the mentioned period.
Figure 8.18 shows the predicted deposition distribution on the VKI guide vane after 12
hours. The gure shows that the deposition occurred mainly on the blade pressure surface.
A small amount of deposition was predicted on the blade suction surface. Most of the particles were reaching the surface by the eect of inertia force because they were supermicron
particles. The maximum deposition thickness occurred near the blade midchord. The gure
shows also the distribution of the stuck particles on the blade pressure surface. It is clear
that particle detachment from the surface occurred on the blade pressure surface near the
trailing edge. Figure 8.19 shows the calculated uid wall friction velocity on the blade pressure surface. The maximum wall friction velocity was predicted near the blade trailing edge.
The high friction velocity in this region caused particle detachment from the surface.

8.4 Ash Particle Deposition on Turbine Blades

105

6.0

Mass [kg/m ]

Stuck mass
Deposited mass
4.0

ps
2.0

ss
0.0
0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

Distance along blade surface (s/c) [-]

Figure 8.18: Comparison between the stuck mass and the deposited mass

Fluid wall friction velocity [m/s]

100
80
60
40
20
0
0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

Distance along blade surface (s/c) [-]

Figure 8.19: Calculated uid wall friction velocity on the blade pressure surface

8.5 Fouled Blade Prole

106

8.5 Fouled Blade Prole


The following approaches were investigated to generate the prole of the fouled blade:
 In the rst approach, the volume of each uid
cell adjacent to a wall was compared to the volume of the deposited particles. When the volume of the deposited particles was larger than
the volume of the uid cell, the uid cell was
changed into wall cell. For this purpose, a userdened subroutine was modied to collect the
deposited particle in each cell, to obtain the volume of the cell and to include the condition for
changing the type of the cell. When this userdened subroutine was tested, the fouled blade
shape was obtained (gure 8.20). But when the Figure 8.20: Simulation of the fouled surow eld was solved for the fouled blade prole, face by lling uid cells
it was very dicult to obtain a converged solution because the blade surface was irregular. In
addition, the fouled surface had new adjacent
cells with new values of y + . These values were always greater than the previous values which
might not satisfy the condition for the near-wall modeling.

 The moving boundary concept was also tested. A user-dened subroutine was used to update

the node coordinates when the command was given to solve the ow eld. This subroutine
changed the grid coordinates for the wall adjacent nodes by values equal to the deposition
thickness. The main shortcoming of this method was the requirement of changing the coordinates of entire grid points. A common problem was the overlapping of the grid lines in the
near-wall region which was mainly caused by the dense grid near the wall required for the
near-wall modeling.

 Finally, it was decided to generate the fouled

blade prole outside of the code using deposition data. The deposited mass in each cell
was transferred to an output le and it was
converted to a deposition thickness. The new
coordinates were calculated for the fouled prole. Using the pre-processor, a new grid was
generated. This method took time because the
fouled blade prole and the new grid were generated each time the calculations were repeated.
Therefore, this method could not be repeated at
short intervals. On the other hand, it had the Figure 8.21: Generating a new grid for the
advantages of grid smoothness and the control fouled surface
of the near-wall distance to full the requirements of the near-wall modeling (gure 8.21).
When this method was tested, a converged solution for the fouled blade prole was obtained. Therefore, this method was used to generate
the fouled blade prole.

8.5 Fouled Blade Prole

107

Figure 8.18 shows that there were some oscillations in the deposition rate. At midchord of
the blade pressure surface, the deposition was low followed by a sharp increase in deposition
rate. These oscillations were caused by the relatively small number of particles calculated.
The total mass of the particles calculated was smaller by several orders of magnitude than
the actual mass injected. Increasing the number of calculated particles would reduce these
oscillations and obtain a smooth surface but, on the other hand, it would increase the computational time. In addition, simulating the entire mass of the particles injected during the
time period was impossible. Consequently, it was necessary to use an approach to limit the
number of particles calculated and at the same time to obtain a suciently regular surface to
generate the fouled blade prole and the new grid. Therefore, the average deposition in each
cell was calculated as the average of the deposition in the cell and in four neighbouring cells,
two of them upstream of the cell and two downstream of the cell. The average deposition
was used to calculate the fouled prole.
The apparent density is dened as the mass of the particles per unit volume. It depends
upon the particle size distribution and on the type of packing. The ratio of the apparent
density to the material density lies from =6 to 1. The value =6 corresponds to the packing
when the particles are large and all have the same diameter and each particle occupies a cube
which has an edge equals to the particle diameter. The value 1 corresponds to the limit for
packing of very small particles with diameters approaching zero.
Figure 8.22 shows a cube which has an edge of
equal size of the diameter of a large particle. The
cube is occupied by the large particle and eight identical small particles in the eight corners of the cube.
For this packing arrangement, the ratio between the
apparent density and the material density was found
to be about 0.6. In real systems, it is expected that
this ratio will be greater because the particles exist
in a spectrum and more volume of the cube could
be lled by the particles. Therefore, the ratio between the apparent density and the material density
was considered in this study as 0.75.

8.22: Assumption of particle


Figure 8.23 shows the distribution of the deposi- Figure
arrangement
at the surface
tion thickness on the blade surface calculated from the
average deposition and using the apparent density.
The fouled blade prole was generated based on the assumption that the deposits were
concentrated at the grid points. Therefore, each grid point was moved perpendicular to the
blade surface by a distance equal to the deposition thickness at this point. Figure 8.24 shows
that this concept is correct if the neighbouring cells have the same width.
Figure 8.23 shows that the deposits on the blade suction surface were very small. In fact,
it was about two orders of magnitude smaller than the deposits calculated on the blade pressure surface. Therefore, it did not change the blade shape signicantly. As a consequence,
the coordinates of the fouled blade were calculated for the pressure surface and for the blade
leading edge only. Figure 8.25 shows the predicted prole of the fouled blade after 12 operating hours.

8.5 Fouled Blade Prole

108

Deposition thickness [mm]

3.0
Pressure surface
Suction surface
2.0

1.0

0.0
0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

Distance along the blade surface (s/c) [-]

Figure 8.23: Predicted deposition thickness distribution after 12 operating hours

Profile of the clean blade


Profile of the fouled blade

Figure 8.24: Concept for generation of


the fouled blade surface

Figure 8.25: Predicted fouled blade


prole after 12 operating hours

A new grid was generated for the fouled blade after 12 operating hours. Figure 8.25 shows
that the deposition model predicted a thick deposit on the blade pressure surface near the
leading edge. The author thinks that this thick deposit is not realistic and will be removed
by the uid ow due to gross detachment of particles which was not considered in this study.
Therefore, it was omitted when the new grid was generated. Figures 8.26 and 8.27 show the
structured two-dimensional grid generated for the fouled blade with a total number of cells
of about 17000.

8.6 Deposition on the Fouled Blade

Figure 8.26: Computational grid near


the blade leading edge for the fouled
blade after 12 operating hours

109

Figure 8.27: Computational grid near


the blade trailing edge for the fouled
blade after 12 operating hours

The compressible ow eld was solved by using the same inlet, outlet and wall boundary
conditions as shown in table 8.5. The ow eld was solved by using the RNG k-" model with
the two-layer zonal model. The change in the heat transfer caused by the deposits was not
considered. The increase in surface roughness due to particle deposition was not considered
either. Further investigation is required to study the eect of these parameters on the ow
eld and on the mass rate of deposition. The y + values at the wall adjacent cells were investigated for the near-wall modeling and were smaller than 1.

8.6 Deposition on the Fouled Blade


The generated sample of particles was injected at the inlet to the cascade with velocities and
temperatures equal to the uid inlet velocity and temperature, respectively. The calculations
were repeated for the second and the third periods as explained in section 8.4.3.
Figure 8.28 shows the deposition distribution after each of the three periods. The coordinate origin here represents the point on the fouled blade with the minimum axial distance.
The gure indicates that the deposition distribution depended upon the prole of the blade.
The gure shows that the maximum deposition in all periods occurred on the blade pressure
surface. The location at which the maximum deposition occurred moved in time toward the
stagnation point. The maximum deposition thickness calculated in each period was smaller
than the maximum thickness calculated in the previous period.

8.7 Predicting the Performance of the Fouled Blade

110

Deposition thickness [mm]

3.0
Deposition in the first period
Deposition in the second period
Deposition in the third period
2.0

1.0

0.0
0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

Distance along the blade surface (s/c) []

Figure 8.28: Predicted deposition thickness distribution in three calculation periods

8.7 Predicting the Performance of the Fouled Blade


This part aims at investigating the
ow eld after 36 operating hours.
The surface velocity distribution and
the downstream total pressure coecient
were used in this investigation. The
fouled blade prole was calculated after
36 operating hours. Figure 8.29 shows
the predicted prole for the fouled blade.
The particle deposition model predicted a
thick layer on the blade pressure surface
and at the blade leading edge.
To investigate the ow eld for the
fouled prole, a two-dimensional structured grid was generated with about
18000 cells. Figures 8.30 and 8.31 show
the grid near the blade leading edge and
near the blade trailing edge, respectively.
The compressible ow eld was solved
for the isentropic exit Mach number
of 0.875 using the boundary conditions
shown in table 8.2. The ow eld was
solved using the RNG k-" model with the
two-layer zonal model. The y + values at
the wall adjacent cells were checked for
the requirement of the near-wall modeling and were smaller than 2.

Profile of the clean blade


Fouled profile after 12 hours
Fouled profile after 24 hours
Fouled profile after 36 hours

Figure 8.29: Predicted fouled blade prole

8.7 Predicting the Performance of the Fouled Blade

Figure 8.30: Computational grid near


the blade leading edge after 36 hours

111

Figure 8.31: Computational grid near


the blade trailing edge after 36 hours

Figure 8.32 shows the distribution of the surface velocity for the clean blade and for the
fouled blade after 36 hours. The gure indicates that in the case of the clean blade, the
acceleration of the ow was smooth on both the suction surface and the pressure surface. For
the fouled blade prole, the ow acceleration was not smooth on the pressure surface. There
was a zone of decelerating ow at about 90 % of the chord where ow separation might occur,
resulting in increased prole loss.

Isentropic Mach number [-]

1.50
Experimental data for the clean blade
Numerical calculation for the clean blade
Numerical calculation for fouled blade

1.25
1.00
0.75
0.50
0.25
0.00
0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

Distance along the blade surface (s/c) [-]

Figure 8.32: Surface velocity distribution for the fouled blade after 36 operating hours

8.7 Predicting the Performance of the Fouled Blade

112

To investigate the prole loss for the fouled blade, the ow eld was solved using the
boundary conditions given in table 8.3 for the isentropic exit Mach number of 0.84. The
total pressure coecient was calculated using equation 8.2. Figure 8.33 shows the comparison between the total pressure coecients for the fouled blade after 12 and 36 hours. The
gure also shows the calculated total pressure coecient and the experimental data for the
clean blade. The results indicate that the total pressure decreased with time. The predicted
decrease after 36 hours was higher than that predicted after 12 hours. The decrease in total
pressure of the fouled blade was caused by the thick trailing edge of the fouled blade with a
large recirculation zone.

Total pressure coefficient [-]

0.2
Experimental data for the clean blade
Calculation for the clean blade
Calculation for the fouled blade after 12 hours
Calculation for the fouled blade after 36 hours

0.1
0.0
-0.1
-0.2
ss
-0.3
-0.4

ps
0.0

0.4

0.8

1.2

1.6

Distance along blade pitch y/S[-]

Figure 8.33: Predicted total pressure coecient for the fouled blade after 36 hours

The mass averaged total pressure loss coecient C was calculated from:
S
C = ; R0 SCu dy
0 u dy
R

(8.9)

Figure 8.34 shows the mass averaged total pressure loss coecient versus time. The gure
indicates that the prole loss increased by about 60% after 36 hours.

8.7 Predicting the Performance of the Fouled Blade

113

Pressure loss coefficient []

0.10
0.09
0.08
0.07
0.06
0.05
0

12

24

36

48

Operating period [hr]

Figure 8.34: Mass averaged total pressure loss coecient

Chapter 9

Summary and Conclusions


A Lagrangian particle deposition model was developed in this study. The model is based
on the three main deposition processes as: particle transport, particle sticking and particle
detachment. A commercial CFD code was used in this study. The code allowed to model the
process of particle-wall interaction by employing a user-dened subroutine.
The eect of turbulence on the particle transport was simulated. Therefore, the eect of
turbulence modeling on particle transport to the surface was investigated. Two turbulence
models were tested: the standard k-" model and the RNG k-" model. The solution near the
wall was solved by two models: the standard wall function and the two-layer zonal model.
The eect of temperature gradient in the ow eld on particle transport by the phenomenon called thermophoresis was investigated. The eect of surface temperature on ash
particle sticking at the blade surface was also examined.
The deposition model was used to investigate the eect of ash particle deposition on
the performance of turbine blades. A blade prole was chosen from the literature and the
time-dependent particle deposition was calculated in 36 hours. The surface velocity and the
downstream total pressure were calculated for the clean and the fouled blade proles and
used in this investigation.
This study revealed that the selection of the appropriate turbulence model is important
for deposition calculations. Particle transport to the surface was signicantly aected by
both turbulence model and near-wall treatment. The standard k-" model is not the appropriate turbulence model for particle transport to turbine blade surfaces. The RNG k-" model
leads to a more accurate simulation for particle transport. The standard wall function can be
used for particle transport studies of relatively large particles which are transported to the
surface mainly by their inertia. The size of these particles can be best illustrated by using a
parameter which depends upon the particle size and density and is called particle relaxation
time. For small particles, the details of the boundary layer aect the trajectory greatly. For
these particles, the two-layer zonal model showed better results.
The calculations performed for the eect of thermophoresis on particle transport in laminar pipe ow indicated that this mechanism is an important transport mechanism for small
particles. The eect of the thermophoretic force increases as the particle size decreases.
Studying the eect of the thermophoretic force on particle transport to turbine blades
could not be performed in this study due to a deciency in the turbulence models. It is
114

Summary and Conclusions

115

concluded that predicting the thermophoretic particle transport in turbulent ows needs
anisotropy based turbulence model. In addition, the turbulence model should be able to
predict the laminar and the transition regions accurately.
The study of the eect of surface temperature on ash particle sticking at turbine blade
surface showed that particles smaller than 0:1 m always stick to the surface upon impact.
For larger particles, the sticking probability depends upon particle size, particle impact velocity and temperature. The particle thermal response time relates the time required for a
particle to respond to the changes in the uid temperature. The sticking coecient is the
mass fraction of the incident ash which remains on the surface. When the dependence of
the sticking coecient on the thermal response time was studied, three regions were distinguished: perfect sticking region, cooling aected-region and cooling unaected-region.
When the time-dependent particle deposition on turbine blade surfaces was calculated
using the deposition model, it was found that the acceleration of the ow was not smooth on
the blade pressure surface of the fouled blade. The model predicted a zone of decelerating
ow where the ow separation might occur resulting in increased prole loss. In addition, the
results indicated that the total pressure loss coecient was increased for the fouled blade.
The model predicted an increase of about 60% in the loss coecient after 36 operating hours
when the inlet particle concentration was 100 ppmw.
The ow eld was solved in this study by using isotropy based turbulence models. These
models prevented the prediction of particle transport by thermophoresis in turbulent ows.
For the future work, it is recommended to use anisotropy based turbulence models. In addition, using turbulence models which are able to predict the laminar and the transition
regions would increase the accuracy of the prediction because the thermophoretic force is
more pronounced in the laminar ow regions.
Although experimental data on particle transport is available for pipe ow and turbine
cascades, detailed experimental data on particle deposition is not available. Developing new
deposition models needs controlled and detailed experimental data to check these models.
The required experiments should emphasis on the sticking and detachment processes because
they are almost not available in the literature. They should consider the particles causing
fouling and ash deposition. In addition, the parameters aecting the sticking and the detachment processes such as Young's modulus and Poisson ratio should be measured in the
experiments.
The sticking and the detachment models considered in this study are based on the assumption of spherical particles and smooth surfaces. Non-spherical particles and rough surfaces
need to be considered in future work.

Bibliography
[1] Abuzeid, S. ; Ahmed, A. and Goodarz, A. (1990), Point Source Particle Deposition in a
Turbulent Channel Flow, Proceedings of the ASME Intern. Computers in Engineering,
Conf. Expo., Publ. by ASCE, Boston Society of Civil Engineers Sect., Boston, MA,
USA, pp. 137-143.
[2] Abuzeid, S. ; Ahmed, A. and Goodarz, A. (1990), Numerical Simulations of Particle
Deposition from a Point Source in Turbulent Flow, Proceedings of the 36th Annual
Technical Meeting of Institute of Environmental Sciences, Mount Prospect, IL, USA,
pp. 295-302.
[3] Ahluwalia, R. K.; Im, K. H. and Wenglarz, R. A. (1989), Flyash Adhesion in Simulated
Coal-Fired Gas Turbine Environment, Transactions of ASME, Journal of Engineering
for Gas Turbines and Power, Vol. 111, pp. 672-678.
[4] Ahluwalia, R. K.; Im, K. H.; Chuang, C. F. and Hajduk, J. C. (1986), Particle and
Vapor Deposition in Coal-Fired Gas Turbines, ASME paper 86-GT-239.
[5] Anderson, R. J.; Romanowski, C. J. and France, J. E. (1984), The Adherence of Ash
Particles from the Combustion of Micronized Coal, Research Report, DOE/METC85/2007.
[6] Anderson, R. J.; Logan, R. G.; Meyer, C. T. and Dennis, R. A. (1990), A Combustion/Deposition Entrained Reactor for High Temperature/Pressure Studies of Coal and
Coal Minerals, Rev. Sci. Instrum., Vol. 61, No. 4, pp. 1294-1302.
[7] Arts, T.; De Rouvroit, M. L. and Rutherford, A. W. (1990), Aero-thermal Investigation
of a Highly Loaded Transonic Linear Turbine Guide Vane Cascade, VKI Technical Note
174.
[8] Berbner, S. and Loeer, F. (1994), Inuence of High Temperatures on Particle Adhesion,
Powder Technology, Vol. 78, No. 3, pp. 273-280.
[9] Blcs, A. and Sari, O. (1988), Inuence of Deposit on the Flow in a Turbine Cascade,
ASME paper 88-GT-207.
[10] Borman, G. L. and Ragland, K. W. (1998), Combustion Engineering, McGraw-Hill.
[11] Boyle, R. J. and Ameri, A. A. (1997), Grid Orthogonality Eects on Predicted Turbine
Midspan Heat Transfer and Performance, Journal of Turbomachinery, Vol. 119, pp. 3138.
[12] Brach, R. and Dunn, P. (1992), A Mathematical Model of the Impact and Adhesion of
Microspheres, Aerosol Science and Technology, Vol. 16, No. 1, pp. 51-64.
116

Bibliography

117

[13] Brown, D. P.; Biswas, P. and Rubon, S. G. (1994), Transport and Deposition of Particles
in Gas Turbine; Eects of Convection; Diusion, Thermophoresis, Inertial Impaction and
Cogeneration, FACT-Vol. 18, Combustion Modeling, Scaling and Air Toxins, ASME, pp.
69-76.
[14] Chapman, A. J. (1974), Heat Transfer, 3rd Edition, Collier Macmillan.
[15] Cohen, H.; Rogers, G. F. C. and Saravanamuttoo, H. I. H. (1987), Gas Turbine Theory,
Longman Scientic & Techanical.
[16] Cohn, A. (1982), Eect of Gas and Metal Temperatures on Gas Turbine Deposition,
The American Society of Mechanical Engineers (paper), Publ. by ASME, New York,
NY, USA, 82-JPGC-GT-4.
[17] Crowe, C.; Sommerfeld, M. and Tsuji, Y. (1997), Multiphase Flows with Droplets and
Particles, CRC Press.
[18] Dahneke, B. (1975), Further Measurements of the Bouncing of Small Latex Spheres,
Journal of Colloid and Interface Science, Vol. 51, No. 1, pp. 58-65.
[19] Das, S. K.; Sharma, M. K. and Schechter, R. S. (1995), Adhesion and Hydrodynamic
Removal of Colloidal Particles from Surfaces, Particulate Science and Technology, Vol.
13, No. 3-4, pp. 227-247.
[20] Diakunchak, I. S. (1991), Performance Deterioration in Industrial Gas Turbines, ASME
paper 91-GT-228.
[21] Fackrell, J. E.; Brown, K. and Young, J. B. (1994), Modeling Particle Deposition in Gas
Turbines Employed in Advanced Coal-Fired Systems, ASME paper 94-GT-467.
[22] Fluent 4.4 User's Guide (1997), Second Edition.
[23] Gkoglu, S. A. and Rosner, D. E. (1984), Correlation of Thermophoretically-Modied
Small Particle Diusional Deposition Rates in Forced Convection Systems with Variable Properties, Transpiration Cooling and/or Viscous Dissipation, Int. J. Heat Mass
Transfer, Vol. 27, No. 5, pp. 639-646.
[24] Greeneld, C. and Quarini, G. (1997), Particle Deposition in a Turbulent Boundary
Layer, Including the Eect of Thermophoresis, Proc. of the 1997 ASME Fluids Eng.
Div. Summer Meeting, Gas Particle Flows, American Society of Mechanical Engineers.
[25] Greeneld, C. and Quarini, G. (1998), A Lagrangian Simulation of Particle Deposition in
a Turbulent Boundary Layer in the Presence of Thermophoresis, Applied Mathematical
Modeling, Vol. 22, pp. 759-771.
[26] Hamrick, J. T. and Schiefelbein, G. F. (1991), Development of Biomass as an Alternative
Fuel for Gas Turbines, Pacic Northwest Laboratory, Richland, Washington 99352, PNL7673, DE91 012129.
[27] Haq, I. U.; Bu-Hazza, A. I. and Al-Baz, K. (1998), Multistage Centrifugal Compressor
Fouling Evaluation at High Power Settings, ASME paper 98-GT-53.
[28] He, C. and Ahmadi, G. (1998), Particle Deposition with Thermophoresis in Laminar
and Turbulent Duct Flows, Aerosol Science and Technology, Vol. 29, No. 6, pp. 525-546.

Bibliography

118

[29] Hildebrandt, T. and Fottner, F. (1999), A Numerical Study of the Inuence of Grid
Renement and Turbulence Modeling on the Flow Field Inside a Highly Loaded Turbine
Cascade, Journal of Turbomachinery, October 1999, Vol. 121, pp. 709-716.
[30] Huang, W.; Yu, M.; Yao, X. and Mao, J. (1997), Numerical Study of Fine Particle
Deposition in Gas Turbine Cascade, Proceedings of the International Symposium on
Multiphase Flow, pp. 250-256.
[31] Kallio, G. A. and Reeks, M. W. (1989), A Numerical Simulation of Particle Deposition
in Turbulent Boundary Layers, International Journal of Multiphase Flow, Vol. 15, No.
3, pp. 433-446.
[32] Kang, S. G.; Kerstein, A. R.; Helble, J. J. and Sarom, A. F. (1990), Simulation of
Residual Ash Formation During Pulverized Coal Combustion: Bimodal Ash Particle
Size Distribution, Aerosol Science Technology, Vol. 13, pp. 401-412.
[33] Kawagishi, H.; Nagao, S. and Kawasaki, S. (1992), Performance Evaluation of Geothermal Steam Turbine with Scale Deposits, PWR-Vol. 18, Steam Turbine-Generator Developments for the Power Generation Industry, ASME, pp. 69-73.
[34] Kim, S. E.; Choudhury, D. and Patel, B. (1999), Computations of Complex Turbulent
Flows Using the Commercial Code Fluent, Modeling Complex Turbulent Flows, Kluwer
Acdemic Publishers, pp. 259-276.
[35] Kladas, D. D. (1993), Turbine Cascade Optimization Against Particle Deposition, Ph.
D. Thesis, Department of Mechanical Engineering, University of Patras.
[36] Launder, B. E. and Spalding, D. B. (1974), The Numerical Computation of Turbulent
Flows, Computer Methods in Applied Mechanics and Engineering, Vol. 3, pp. 269-289.
[37] Lee, F. C. C. and Lockwood, F. C. (1999), Modelling Ash Deposition in Pulverized
Coal-Fired Applications, Progress in Energy and Combustion Science 25, pp. 117-132.
[38] Liu, B. and Agarwal, J. (1974), Experimental Observation of Aerosol Deposition in
Turbulent Flow, Aerosol Science, Vol. 5, pp. 145-155.
[39] Liu, H.; Ge, M. and Lou, V. (1989), Deposition Pattern of Fine Solid Particles in a Gas
Turbine Cascade, Proceeding of the 2nd Xi'an International Symposium on Multiphase
Flow and Heat Transfer, September 18-21, Xi'an, pp. 1131-1139.
[40] Logan, R. G.; Richards, G. A.; Meyer, C. T. and Anderson, R. J. (1990), A Study
of Techniques for Reducing Ash Deposition in Coal-Fired Gas Turbine, Prog. Energy
Combust. Sci., Vol. 16, pp. 221-233.
[41] Marcus, R.D; Leung, L. S.; Klinzing, G. E. and Rizk, F. (1990), Pneumatic Conveying
of Solids, Chapman and Hall.
[42] Menguturk, M. and Sverdrup, E. F. (1982), A Theory for Fine Particle Deposition in
Two-dimensional Boundary Layer Flows and Application to Gas Turbines, Transaction
of ASME, Journal of Engineering for Power, Vol. 104, pp. 69-76.
[43] Menguturk, M.; Gunes, D. and Mimaroglu, H. K. (1982), Blade Boundary Layer Effect on Turbine Erosion and Deposition, Particulate Laden Flows in Turbomachinery,
AIAA/ASME Joint Fluids Plasma, Thermophysics and Heat Transfer Conference ST.
Louis, Missouri, June 7-11, pp. 71-78.

Bibliography

119

[44] Mezheritsky, A. D. and Sudarev, A. V. (1990), Mechanism of Fouling and the Cleaning
Technique in Application to Flow Parts of the Power Generation Plant Compressors,
ASME paper 90-GT-103.
[45] Montassier, N.; Boulaud, D. and Renoux, A. (1990), Experimental Study of Thermophoretic Deposition of Particles in Laminar Tube Flow, The 3rd International Aerosol
Conference, Kyoto (Japan), pp. 395-398.
[46] Montassier, N.; Boulaud, D. and Stratmann, F. (1990), Comparison between Experimental Study and Theoretical Model of Thermophoretic Particle Deposition in Laminar
Tube Flow J. Aerosol Science, Vol. 21, pp. 585-588.
[47] Morsi, S. and Alexander, A. (1972), An Investigation of Particle Trajectories in TwoPhase Flow Systems, J. Fluid Mechanics, Vol. 55, Part. 2, pp. 193-208.
[48] Orszag, S. A.; Staroselsky, I.; Flannery, W. S. and Zhang, Y. (1996), Introduction to
Renormalization Group Modeling of Turbulence, Simulation and Modeling of Turbulent
Flows, ICASE/LaRC Series in Computational Science and Engineering.
[49] Ounis, H.; Ahmadi, G. and Mclaughlin, J.B. (1991), Dispersion and Deposition of Brownian Particles from Point Sources in a Simulated Turbulent Channel Flow, Journal of
Colloid and Interface Science, Vol. 147, No.1, pp. 233-250.
[50] Owen, I. (1987), The Diusion of Aerosols onto a Heated Turbine Blade Particulate and
Multiphase Processes, Vol. 2, Contamination Analysis and Control, Publ. By Hemishere
Publ. Corp., Washington, DC, USA, pp. 357-368.
[51] Parker, G. and Lee, P. (1972), Studies of the Deposition of Sub-Micron Particles on
Turbine Blades, Proc Instn Mech Engrs, Vol. 186 38/72, pp. 519-526.
[52] Parker, G. and Ryley, D. (1970), Equipment and Techniques for Studying the Deposition
of Sub-micron Particles on Turbine Blades, Fluid Mechanics and Measurements in Two
Phase Flow Systems, Symp. by Thermodynamic and Fluid Mech. Group of Inst. Mech.
Eng., and Yorkshire of Inst Chem, Eng., Sep. 2425, 1969, Univ of Leeds, Engl., Inst
Mech Eng (Proc 1969-70 V. 184, pt 3c), London.
[53] Ragland, K. W.; Misra, M. K.; Aerts, D. J. and Palmer, C. A. (1995), Ash Deposition
in a Wood-Fired Gas Turbine, Transactions of ASME, Journal of Engineering for Gas
Turbines and Power, Vol. 117, pp. 509-512.
[54] Ralls, K. M.; Courtney, T. H. and Wul, J. (1976), Introduction to Materials Science
and Engineering, John Wiley & Sons, Inc.
[55] Rimai, D. S.; Demejo, L. P. and Bowen, R. C. (1994), Mechanics of Particle Adhesion,
J. Adhesion Science Technology, Vol. 8, No. 11, pp. 1333-1355.
[56] Richards, G. A.; Logan, R. G.; Meyer, C. T. and Anderson, R. J. (1992), Ash Deposition
at Coal-Fired Gas Turbine Conditions: Surface and Combustion Temperature Eects,
Transactions of ASME, Journal of Engineering for Gas Turbines and Power, Vol. 114,
pp. 132-138.
[57] Ryley, D. J. and Davies, J. B. (1983), Eect of Thermophoresis on Fog Droplet Deposition on Low Pressure Stream Turbine Guide Blades, Int. J. Heat & Fluid Flow, Vol. 4,
No. 3, pp. 161-167.

Bibliography

120

[58] Romay, F. J.; Takagaki, S. S.; Pui, D. Y. and Liu, B. Y. (1998), Thermophoretic Deposition of Aerosol Particles in Turbulent Pipe Flow, Journal of Aerosol Science, Vol. 29,
No. 8, pp. 943-959.
[59] Saman, P. G. (1965), The Lift on a Small Sphere in a Slow Shear Flow, Journal of
Fluid Mechanics, Vol. 22, Part 2, pp. 385-400.
[60] Saravanamuttoo, I. H. and Lakshminarasimha, A. N. (1985), A Preliminary Assessment
of Compressor Fouling, ASME paper 85-GT-153.
[61] Schlichting, H. and Gersten, K. (2000), Boundary Layer Theory, 8th Edition, Springer.
[62] Sethi, W. K. and Partanen, W. (1994), Corrosion and Wear Issues in a Wood-Fired Gas
Turbine System, National Association of Corrosion Engineers (NACE), Baltimore, MD,
pp. 172/1-172/13.
[63] Soltani, M. and Ahmadi, G. (1994), On Particle Adhesion and Removal Mechanisms in
Turbulent Flows, J. Adhesion Science Technology, Vol. 8, No. 7, pp. 763-785.
[64] Sommerfeld, M. (2000), Theoretical and Experimental Modelling of Particulate Flow,
VKI Lecture Series, 2000-06.
[65] Sommerfeld, M. (1992), Numerical Method for Calculating Particle Dispersion in Turbulent Flow, Sixth Workshop on Two-Phase Flow Predictions, Erlangen, March 30 - April
2, 1992.
[66] Spiro, C. L.; Kimura, S. G. and Chen, C. C. (1987), Ash Behaviour During Combustion and Deposition in Coal-Fueled Gas Turbines, Transactions of ASME, Journal of
Engineering for Gas Turbines and Power, Vol. 109, pp. 325-330.
[67] Stalder, J. P. (1998), Gas Turbine Compressor Washing State of the Art-Field Experience, ASME paper 98-GT-420.
[68] Tabako, W. (1986), Study of Single Stage Axial Flow Compressor Performance Deterioration, American Society of Mechanical Engineers, Fluids Engineering Division (Publication) FED, Vol. 37, Publ. by ASME, New York, NY, USA, pp. 95-100.
[69] Tabako, W. (1984), Review-Turbomachinery Performance Deterioration Exposed to
Solid Particulates Environment, Transactions of ASME, Journal of Fluids Engineering,
Vol. 106, pp. 125-134.
[70] Tabako, W.; Hosny, W. and Hamed, A. (1976), Eect of Solid Particles on Turbine
Performance, Transactions of ASME, Journal of Engineering for Power, pp. 47-52.
[71] Talbot, L.; Cheng, R. K.; Schefer, R. W. and Willis, D. R. (1980), Thermophoresis of
Particles in a Heated Boundary Layer, J. Fluid Mech., Vol. 101, Part 4, pp. 737-758.
[72] Tarabrin, A. P.; Schurovsky, V. A.; Bodrov, A. I. and Stalder, J. P. (1998), Inuence of
Axial Compressor Fouling on Gas Turbine Unit Performance Based on Dierent Schemes
and with Dierent Initial Parameters, ASME paper 98-GT-416.
[73] Tarabrin, A. P.; Schurovsky, V. A.; Bodrov, A. I. and Stalder, J. P. (1996), An Analysis
of Axial Compressors Fouling and a Cleaning Method of their Blading, ASME paper
96-GT-363.

Bibliography

121

[74] Venables, W. N. and Smith, D. M., An Introduction to R, A programming Analysis for


Data Analysis and Graphics.
[75] Wenglarz, R. A. and Fox, R. G. (1989), Physical Aspects of Deposition from Coal Water
Fuels under Gas Turbine Conditions, ASME paper 89-GT-206.
[76] Wenglarz, R. A. and Fox, R. G. (1989), Chemical Aspects of Deposition/Corrosion from
Coal Water Fuels under Gas Turbine Conditions, ASME paper 89-GT-207.
[77] Wenglarz, R. A.; Nirmalan, N. V. and Daehler, T. G. (1995), Rugged ATS Turbines for
Alternate Fuels, ASME paper 95-GT-73.
[78] Whitlow, G. A.; Lee, S. Y.; Mulik, P. R.; Wenglerz, R. A.; Sherlock, T. P. and Cohn,
A. (1983), Combustion Turbine Deposition Observations from Residual and Simulated
Residual Oil Studies, Transactions of ASME, Journal of Engineering for Power, pp.
88-96.
[79] Wood, N. B. (1981), A Simple Method for the Calculation of Turbulent Deposition to
Smooth and Rough Surfaces, J. Aerosol Science, Vol. 12, No. 3, pp. 275-290.
[80] Yau, K. K. and Young, J. B. (1987), The Deposition of Fog Droplets on Steam Turbine
Blades by Turbulent Diusion, Transactions of ASME, Journal of Turbomachinery, Vol.
109, pp. 429-435.
[81] Young, J. B. and Yau, K. K. (1988), The Inertial Deposition of Fog Droplets on Steam
Turbine Blades, Transactions of ASME, Journal of Turbomachinery, Vol. 110, pp. 155162.

List of Figures
1.1 Distributions of the downstream total pressure coecient, Kawagishi et al. [33] 11
1.2 Details of particle deposition process . . . . . . . . . . . . . . . . . . . . . . . 12
1.3 Dimensionless particle transport velocity versus dimensionless particle relaxation time, Liu and Agarwal [38] . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.1 Universal laws of the wall, Schlichting and Gersten [61] . . . . . . . . . . . . . 27
3.1 Drag coecient for spherical particles versus Reynolds number, Schlichting
and Gersten [61] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.2 Cunningham correction factor versus Knudsen number, Talbot et al. [71] . . . 34
3.3 Regimes of dispersed two-phase ows as a function of particle volume fraction,
Sommerfeld [64] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.1
4.2
4.3
4.4
4.5
4.6
4.7
4.8
4.9

Summary of the most common sticking mechanisms . . . . . . . . . . . . . . .


Dimensionless sticking forces versus particle diameter, Berbner and Loeer [8]
The coecient of restitution versus impact velocity, Dahneke [18] . . . . . . .
Illustration for the sticking mechanism . . . . . . . . . . . . . . . . . . . . . .
Particle capture velocity versus particle Young's modulus . . . . . . . . . . .
Geometric features of a spherical particle stuck to a smooth surface . . . . . .
Variation of the contact radius with the particle diameter . . . . . . . . . . .
Variation of the sticking force for silicon particles versus particle diameter . .
Variation of the critical shear velocity versus particle diameter . . . . . . . . .

41
41
42
43
44
45
47
48
49

5.1 Deposition ow chart . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51


5.2 Deposition model in Fluent . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
6.1 Experimental system of Liu and Agarwal [38] . . . . . . . . . . . . . . . . . .
6.2 Comparison between particle transport velocity to the pipe wall calculated using dierent turbulence models and the experimental data of Liu and Agarwal
[38] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.3 Experimental system of Parker and Lee [51] . . . . . . . . . . . . . . . . . . .
6.4 Blade geometry of Parker and Lee [51] . . . . . . . . . . . . . . . . . . . . . .
6.5 The grid topology used to generate the grid for the blade of Parker and Lee .
6.6 The grid used to solve the ow eld using turbulence models with the standard
wall function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.7 The grid used to solve the ow eld using turbulence models with the two-layer
zonal model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.8 Comparison between the numerical calculation and the experimental data for
the surface velocity distribution . . . . . . . . . . . . . . . . . . . . . . . . . .
6.9 Calculated particles for dierent injected zones . . . . . . . . . . . . . . . . .
122

57
60
61
62
63
64
64
65
66

List of Figures
6.10 Comparison between the numerical results using dierent turbulence models
and the experimental data for particle transport to turbine blade surfaces . .
6.11 Comparison between particle mean impact velocities calculated using dierent
turbulence models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.12 Comparison between the numerical results and the experimental data of particle transport for two dierent mean diameters . . . . . . . . . . . . . . . . .
6.13 Secondary ows in turbine blade passages . . . . . . . . . . . . . . . . . . . .
6.14 A grid of about 17000 cells used in the sensitivity study . . . . . . . . . . . .
6.15 Comparison between the numerical calculations using two grids and the experimental data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.16 Experimental system of Montassier et al. [45] . . . . . . . . . . . . . . . . . .
6.17 Assumed dimensions of the bend . . . . . . . . . . . . . . . . . . . . . . . . .
6.18 The grid at the entrance to the test tube . . . . . . . . . . . . . . . . . . . . .
6.19 Calculated velocity at the inlet to the test tube . . . . . . . . . . . . . . . . .
6.20 The secondary ow eld at the inlet to the test tube . . . . . . . . . . . . . .
6.21 Particle trajectories in the test tube . . . . . . . . . . . . . . . . . . . . . . .
6.22 Comparison between the numerical calculations and the experimental data for
the thermophoretic transport of 0:1 m particles . . . . . . . . . . . . . . . .
6.23 Comparison between the numerical calculations using the assumption given in
case B and the experimental data for the cumulative thermophoretic transport
of 0:1 m particles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.24 Comparison between the numerical calculations using the assumption given in
case B and the experimental data for the thermophoretic transport of dierent
size particles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

123
67
68
69
69
71
71
72
74
75
75
76
76
77
78
78

7.1
7.2
7.3
7.4
7.5

Experimental set-up of Richards et al. [56] . . . . . . . . . . . . . . . . . . . .


Particle size distribution measured by Richards et al. [56] . . . . . . . . . . .
Variation of Young's modulus with temperature, Ralls et al. [54] . . . . . . .
Sticking coecient versus time, Richards et al. [56] . . . . . . . . . . . . . . .
Comparison between the numerical calculations and the experimental data for
the sticking coecient . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

82
83
84
85

8.1
8.2
8.3
8.4
8.5
8.6
8.7
8.8
8.9

Geometry of the VKI prole, Arts et al. [7] . . . . . . . . . . . . . . . . . . .


VKI grid topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Computational domain and the internal region . . . . . . . . . . . . . . . . .
VKI computational grid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Computational grid near the blade leading edge . . . . . . . . . . . . . . . . .
Computational grid near the blade trailing edge . . . . . . . . . . . . . . . . .
Mach number contours for the isentropic exit Mach number of 0.875 . . . . .
Mach number contours for the isentropic exit Mach number of 1.02 . . . . . .
Comparison between the numerical calculation and experimental data for the
surface Mach number distribution, M2is = 0:875 . . . . . . . . . . . . . . . . .
Comparison between the numerical calculation and experimental data for the
surface Mach number distribution, M2is = 1:02 . . . . . . . . . . . . . . . . .
Comparison between the numerical calculations and experimental data for the
pitchwise variation of the total pressure coecient . . . . . . . . . . . . . . .
Sticking coecient versus particle diameter, case 1 . . . . . . . . . . . . . . .
Sticking coecient versus particle diameter, case 2 . . . . . . . . . . . . . . .
Sticking coecient versus particle diameter, case 3 . . . . . . . . . . . . . . .
Universal distributions for the sticking coecient on the blade surface . . . .

88
90
90
90
91
91
93
93

8.10
8.11
8.12
8.13
8.14
8.15

86

94
94
96
99
99
99
100

List of Figures
8.16 (a) The VKI transonic inlet guide vane, (b) Typical temperature distribution
at midheight of a cooled blade designed to operate with a gas temperature of
1500 K , Cohen et al. [15] . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
8.17 Particulate collection eciencies of cyclones, Sethi and Partanen [62] . . . . .
8.18 Comparison between the stuck mass and the deposited mass . . . . . . . . . .
8.19 Calculated uid wall friction velocity on the blade pressure surface . . . . . .
8.20 Simulation of the fouled surface by lling uid cells . . . . . . . . . . . . . . .
8.21 Generating a new grid for the fouled surface . . . . . . . . . . . . . . . . . . .
8.22 Assumption of particle arrangement at the surface . . . . . . . . . . . . . . .
8.23 Predicted deposition thickness distribution after 12 operating hours . . . . . .
8.24 Concept for generation of the fouled blade surface . . . . . . . . . . . . . . . .
8.25 Predicted fouled blade prole after 12 operating hours . . . . . . . . . . . . .
8.26 Computational grid near the blade leading edge for the fouled blade after 12
operating hours . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
8.27 Computational grid near the blade trailing edge for the fouled blade after 12
operating hours . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
8.28 Predicted deposition thickness distribution in three calculation periods . . . .
8.29 Predicted fouled blade prole . . . . . . . . . . . . . . . . . . . . . . . . . . .
8.30 Computational grid near the blade leading edge after 36 hours . . . . . . . . .
8.31 Computational grid near the blade trailing edge after 36 hours . . . . . . . . .
8.32 Surface velocity distribution for the fouled blade after 36 operating hours . .
8.33 Predicted total pressure coecient for the fouled blade after 36 hours . . . . .
8.34 Mass averaged total pressure loss coecient . . . . . . . . . . . . . . . . . . .

124

102
103
105
105
106
106
107
108
108
108
109
109
110
110
111
111
111
112
113

List of Tables
2.1 The values of the constants in the standard k-" model . . . . . . . . . . . . . 24
2.2 The values of the constants in the RNG k-" model . . . . . . . . . . . . . . . 25
3.1 Dierent regimes of rareed ows with respect to particle motion, Crowe et
al. [17] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.1 Properties of silicon particles, Soltani and Ahmadi [63] . . . . . . . . . . . . . 48
4.2 Properties of ash particles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
6.1 Parameters for the cascade of Parker and Lee [51] . . . . . . . . . . . . . . . .
6.2 Inlet boundary conditions of the cascade of Parker and Ryley [52] . . . . . . .
6.3 Inlet boundary conditions for the two-dimensional ow eld calculations prior
thermophoretic particle transport . . . . . . . . . . . . . . . . . . . . . . . . .
6.4 Particle inlet boundary conditions for the two-dimensional thermophoretic particle transport . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

62
63
73
73

7.1 Inlet boundary conditions for the experiments of Richards et al. [56] . . . . . 83
7.2 Measured Young's modulus for ash particles . . . . . . . . . . . . . . . . . . . 85
8.1
8.2
8.3
8.4
8.5
8.6

Cascade parameters for the VKI prole, Arts et al. [7] . . . . . . . .


VKI inlet boundary conditions . . . . . . . . . . . . . . . . . . . . .
VKI boundary conditions for loss calculations . . . . . . . . . . . . .
Dierent particle properties used to calculate the sticking coecient
Flow eld boundary conditions prior to deposition calculations . . .
Assumed particle specications at the inlet to the cascade . . . . . .

125

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

88
92
95
98
102
103

126

Curriculum Vitae
Name:
Date of birth
Place of birth
Nationality
Religion
Marital status

Hesham Mohamed El-Batsh


1.10.1965
Kafer El-Sheikh, Egypt
Egyptian
Moslem
married and has a son

1971-1977
1977-1980
1980-1983
1983-1988

Primary school, Shebin El-Kom, Egypt


Prep school, Shebin El-Kom, Egypt
Secondary school, Shebin El-Kom, Egypt
Studying Mechanical Engineering at Menoua University, Egypt
Graduating with a Bachelor of Science
Working at Damanhour Thermal Power Plant, Egypt
University Assistant, Banha High Institute of Technology, Egypt
Master of Science
Ph. D. student,
Institute of Thermal Turbomachines and Powerplants
Vienna University of Technology, Austria

1989-1992
1992-1997
1994
1997-2001

Lebenslauf
Name:
Geboren am:
Staatsangehrigkeit
Religion
Familienstand
1971-1977
1977-1980
1980-1983
1983-1988
1989-1992
1992-1997
1994
1997-2001

Hesham Mohamed El-Batsh


1.10.1965,
in Kafer El-Sheikh, gypten
gypten
Moslem
verheiratet mit Frau Nashwa Ezzat Eissa
und einen Sohn Omar
Primary school, Shebin El-Kom, gypten
Prep school, Shebin El-Kom, gypten
Secondary school, Shebin El-Kom, gypten
Studium an der Menoua University, gypten
Studienrichtung Maschinenbau,
Abschluss mit Bachelor of Science
Mitarbeiter im Damanhour Thermal Power Plant, gypten
Universittsassistent,
Banha High Institute of Technology, gypten
Master of Science
Doktoratsstudium,
Institut fr Thermische Turbomaschinen und Energieanlagen
Technische Universitt Wien

You might also like