Professional Documents
Culture Documents
Under supervision of
O.Univ.-Prof.Dipl.-Ing.Dr.techn. Hermann Haselbacher
Head of the Institute of Thermal Turbomachines and Powerplants
Presented by
DISSERTATION
Modeling Particle Deposition on
Compressor and Turbine Blade Surfaces
ausgefhrt zum Zwecke der Erlangung des akademischen Grades eines
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
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
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
30
31
4 Particle-Wall Interaction
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
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
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Assumptions . . . . . . . .
Experimental System . . . .
Numerical Calculations . . .
Tuning the Sticking Model .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
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]
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]
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]
Greek Characters
0
k , "
mol
p
t
ij
"
0
0
t
e
[m]
[;]
[;]
[;]
[;]
[;]
[;]
[s]
[m]
[;]
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
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.
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.
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.
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
[]
C = p0o2:5;Up2o1
2
where:
po1
po2
U2
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
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.
Particle transport
Particle sticking
Particle detachment
- Inertia force
- Turbulent diffusion
- Brownian diffusion
- Electrostatic force
- Thermophoresis
Fluid force
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.
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
Figure 1.3: Dimensionless particle transport velocity versus dimensionless particle relaxation
time, Liu and Agarwal [38]
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
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].
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]
17
Part I
18
Chapter 2
The mass conservation equation for incompressible as well as compressible steady ows is
given by:
@ (u ) = 0
@xi i
where
(2.1)
Conservation of momentum in the ith direction for a steady ow neglecting body force is
described by:
19
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
@ (u h) = @ k @T + u @p + @ui
i @x
ij @x
@xi i
@xi f @xi
i
j
where
(2.4)
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
Diusion
(2.5)
21
= + 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.
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]:
(2.11)
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:
(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.
u0iu0j
(2.13)
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
(2.16)
23
24
(2.17)
@ (u k) = @ t @k + G ; "
k
@xi i
@xi k @xi
(2.18)
(2.19)
e = 1 + C pk"
(2.21)
25
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)
(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:
(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)
(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
26
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)
(2.29)
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)
p
Mt = c k
s
and
(2.31)
p
cs speed of sound (=
RT )
R gas constant
ratio of specic heats
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+
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
Figure 2.1: Universal laws of the wall, Schlichting and Gersten [61]
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.
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)
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)
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)
cl = C;3=4;
A = 70;
A" = 2cl
(2.38)
Part II
30
Chapter 3
31
32
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)
(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
Figure 3.1: Drag coecient for spherical particles versus Reynolds number, Schlichting and
Gersten [61]
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.
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)
c = 8RT
1
2
(3.6)
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)
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.
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]
35
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.
Lift forces on a particle are caused by the velocity gradient in the uid or by particle rotation.
FS = 1:615Dp2 kr (u ; up)
where kr is the local velocity gradient.
(3.9)
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.
36
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:
kp
rT
Cs
Ct
Cm
(3.10)
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.
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.
37
(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.
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)
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.
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:
where:
(3.17)
Nu Nusselt number
Pr uid Prandtl number (Cp=kf )
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.
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
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.
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.
43
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)
vcr
Es
s
Ep
p
Dp
p
44
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.
103
102
rebounding
rebounding
101
100
10-1
10-2
105
sticking
sticking
Dp = 10 m
Dp = 1m
106
107
108
109
1010
45
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)
46
FD D2p Fpo a
(4.8)
u+ = y +
(4.9)
y + = u y
u+ = uu ;
(4.10)
For smooth surfaces, the dimensionless velocity at the center of a stuck spherical particle
Dp+
+
u =
(4.11)
uc = u Dp
u
2
(4.12)
uc = D2p u2
(4.13)
(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)
47
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)
;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.
10-6
10-7
10-8
10-9
Silicon particles
Ash particles
10-10
10-8
10-7
10-6
10-5
10-4
Figure 4.7: Variation of the contact radius with the particle diameter
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.
10-4
10-5
10-6
10-7
10-8
10-9
10-10
10-8
10-7
10-6
10-5
10-4
Figure 4.8: Variation of the sticking force for silicon particles versus particle diameter
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)
49
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
Figure 4.9: Variation of the critical shear velocity versus particle diameter
Chapter 5
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.
52
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.
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.
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
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.
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.
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.
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.
54
Part III
55
Chapter 6
(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:
(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)
57
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.
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 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
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.
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
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)
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
(6.7)
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.
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.
100
10-1
10-2
10-3
10-4
10-5
10-6
0.1
10
100
1000
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]
61
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.
62
249
127
167
0.67
Table 6.1: Parameters for the cascade of Parker and Lee [51]
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.
63
Figure 6.5: The grid topology used to generate the grid for the blade of Parker and Lee
11.0
0.0
1:013 105
2
0.009
Table 6.2: Inlet boundary conditions of the cascade of Parker and Ryley [52]
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
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
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
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.
66
3.2
3.1
3.0
2.9
Dp= 0.05 m
Dp= 1 m
2.8
0
20
40
60
80
100
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.
67
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
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.
68
20
15
10
0
0.0
0.2
0.4
0.6
0.8
1.0
1.2
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.
69
1.5
suction surface
1.0
0.5
pressure surface
0.0
3.0
suction surface
2.0
pressure surface
1.0
0.0
-1.2
-0.8
-0.4
0.0
0.4
0.8
1.2
Figure 6.12: Comparison between the numerical results and the experimental data of particle
transport for two dierent mean diameters
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.
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.
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
Figure 6.15: Comparison between the numerical calculations using two grids and the experimental data
72
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.
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.
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.
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).
75
Velocity [m/s]
0.5
0.0
0.0
0.2
0.4
0.6
0.8
1.0
Figure 6.20: The secondary ow eld at the inlet to the test tube
76
77
2.5
2.0
1.5
Dp = 0.1 m
1.0
0.5
0.0
0
15
30
45
60
75
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.
78
Dp = 0.1 m
Numerical calculation
Experimental data
0
0
15
30
45
60
75
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
20
Experimental data
Numerical calculations
15
10
wall temperature
= 293 K
wall temperature
= 283 K
0
0.1
10
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
79
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
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
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
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.
83
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.
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.
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.
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.
85
-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)
86
Ep = 3 105(1589 ; Tp)
where:
(7.2)
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
Figure 7.5: Comparison between the numerical calculations and the experimental data for
the sticking coecient
Chapter 8
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
x/cax= 1.433
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.
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.
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.
90
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 + .
po = 1 +
; 1 M 2
;1
p
2
where:
po total pressure
p static pressure
M Mach number
(8.1)
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
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.
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
94
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
Figure 8.9: Comparison between the numerical calculation and experimental data for the
surface Mach number distribution, M2is = 0:875
1.75
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
Figure 8.10: Comparison between the numerical calculation and experimental data for the
surface Mach number distribution, M2is = 1:02
95
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.
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
C = p0o:25; Upo21
2 2
where:
(8.2)
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.
96
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
Figure 8.11: Comparison between the numerical calculations and experimental data for the
pitchwise variation of the total pressure coecient
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.
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.
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)
(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)
dTp = 1 (T ; T )
p
dt T
(8.8)
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.
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
99
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
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
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
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
10-7
10-6
10-5
10-4
10-3
Figure 8.15: Universal distributions for the sticking coecient on the blade surface
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.
101
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.
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.
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
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.
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.
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
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.
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.
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.
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
Figure 8.18: Comparison between the stuck mass and the deposited mass
100
80
60
40
20
0
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
Figure 8.19: Calculated uid wall friction velocity on the blade pressure surface
106
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.
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.
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.
108
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
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.
109
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.
110
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
111
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.
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
Figure 8.32: Surface velocity distribution for the fouled blade after 36 operating hours
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.
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
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.
113
0.10
0.09
0.08
0.07
0.06
0.05
0
12
24
36
48
Chapter 9
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
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
41
41
42
43
44
45
47
48
49
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
82
83
84
85
8.1
8.2
8.3
8.4
8.5
8.6
8.7
8.8
8.9
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
125
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
88
92
95
98
102
103
126
Curriculum Vitae
Name:
Date of birth
Place of birth
Nationality
Religion
Marital status
1971-1977
1977-1980
1980-1983
1983-1988
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