Professional Documents
Culture Documents
r
rebound (cut-off) angle
C
cutting wear coefcient
D
deformation wear coefcient
molecular viscosity
M
momentum relaxation time scale
p pressure gradient in the carrier uid
coordinate angular velocity vector
steamused as the sealing source for the gland seals is extracted fromthe main steamsupply,
and upon leaving the sealing devices is discharged to the intercondenser to be condensed.
During turbine operation, the most likely factor to inuence the sealing efciency is the rub
between the labyrinth seals and the rotor. This rub wears the seal, producing two detrimental
effects. First it opens the clearance, causing an increase in the leakage area. Second, the
rub will modify the form or shape of the seal strip, changing its coefcient of discharge.
Both effects tend to increase leakage ow and therefore cause a deterioration of turbine
Z. Mazur et al. / Geothermics 33 (2004) 599614 601
efciency. In the case of geothermal steam turbines, the steam can be contaminated with
particles composed of silica, sulfur, salts and other constituents. These contaminants can
directly impact turbine components (Mazur and Kubiak, 1996) and can cause solid particle
erosion damage. This effect is independent of damage related to the presence of a rub. To
remove solids from the steam, the steam is passed through a cyclonic separator before it
enters the turbine steam path. This separation is not totally effective and some solids may
remain in the main steam ow. The degradation of the rotor seal system by solid particle
erosion commonly results in a considerable reduction of the power plants performance
because seal damage impacts condenser performance.
In this paper, the ow pattern within the rotor end gland seal of a 25 MWe turbine was
investigated under a variety of conditions and modeled using a computational uid dynamics
(CFD) code. This modeling was used to identify causes of the erosion of a rotor low-pressure
end gland seal, as well as possible means of reducing erosion through operational guidelines.
2. Background
Twenty-ve MWe turbines have been in continuous operation at the geothermal eld
for 1 year. Each turbine has one ow rotor arranged in seven stages, as shown in Fig. 1.
Fig. 1. General view of the rotor of a 25 MWe geothermal turbine.
602 Z. Mazur et al. / Geothermics 33 (2004) 599614
Fig. 2. Photograph and schematic diagramshowing where erosion of the rotor end low-pressure gland seal occurred.
Small solid particles and water droplets entrained in the high velocity geothermal steam are
thought to have produced erosion of the low-pressure end gland seal. The arrow on Fig. 1
shows the location of the rotor end gland seal. As the gland seal surface gradually eroded,
steam usage and air ow leakage increased. Details of the rotor end low-pressure gland seal
design and the location of the eroded (denoted by arrow) rotor seal surface are shown in
Fig. 2.
A schematic diagram of the steam and air ow paths to the high and low-pressure
glands of the 25 MWe geothermal turbine is shown in Fig. 3. The low-pressure end of
the unit (steam exhaust) is under vacuum. In this case, there is a positive pressure (atmo-
spheric to vacuum) that can cause ow of air into the turbine casing. Non-condensable
gases in the air will reduce the vacuum, and in doing so, increase the condensing pressure
Fig. 3. Schematic diagram of steam and air ows to the high and low-pressure glands of the 25 MWe geothermal
turbine.
Z. Mazur et al. / Geothermics 33 (2004) 599614 603
and degrade the efciency of the cycle. In addition, any free oxygen in the air will re-
act with other water-wet surfaces and accelerate corrosion. The sealing system at both
the low and high-pressure points must be able to minimize both steam leakage outward
from sections above atmospheric pressure and air leakage inward to sections below atmo-
spheric pressure under all operating conditions. This sealing system requires a regulating
device that controls and limits the steam and air ow to an acceptable level. At reduced
load, a partial vacuum can exist throughout the entire turbine, extending upstream into the
high-pressure section of the turbine. At startup, the entire turbine will be under vacuum.
Under these conditions, steam for sealing must be provided at all sealing locations. The
steam seal leak-off is maintained at a pressure above the main condenser and is vented
to the intercondenser. The lowest pressure leak-off in the system is at the gland steam
condensers, which are maintained at pressures marginally above the condenser pressure.
In these condensers, the steam is condensed and any air that is present is removed and
expelled from the system by steam-jet air ejectors to prevent its contaminating the working
uid.
To maintain a high efciency steam seal system, the seal ows and pressures must be
controlled within precise limits. This study sought to determine both the optimal rotor end
steam seal system ow conditions and the relationships between steam ow conditions and
rotor erosion rates to ensure both high efciency and high reliability of the seal (reduction
of erosion).
3. Modeling methodology
Numerical two-dimensional predictions have been carried out with the nite-volume
code STAR-CD using a RNG k/High Reynolds Number model for turbulence (Launder
and Spalding, 1972; Mack et al., 1999; Adapco STAR-CD, 2001). STAR-CD provides a
dispersed multi-phase ow framework. The framework is of the Lagrangian/Eulerian type
(Bracco, 1985), in which the conservation equations of mass, momentumand energy for the
dispersed phase are written for each individual particle. For ows having a comparatively
small number of dispersed particles, it is possible to solve a set of the afore-mentioned
Lagrangian equations for every element. However, if the number of particles is large, a
statistical approach is more practical. In this statistical approach, the total population is
represented by a nite number of computational parcels (samples), each of which repre-
sents a group (cluster) of particles having the same properties. The number of samples is
not arbitrary; it must be large enough that the properties of the full population are well
represented. In turbulent ows, the random walk technique (Gosman and Ioannides, 1983)
is employed to introduce the uctuating nature of the turbulent velocity eld, which results
in turbulent dispersion of the dispersed elements.
3.1. Governing equations
The motion of the dispersed phase is inuenced by that of the continuous phase (and
vice versa) via displacement and interphase momentum. The strength of the interactions
depends on the size, density and number of the dispersed particle. The momentum equation
604 Z. Mazur et al. / Geothermics 33 (2004) 599614
Fig. 4. Lagrangian particle motion parameters in the carrier uid.
for a particle of mass m
p
is (Adapco STAR-CD, 2001):
m
p
d u
p
d
t
=
F
dr
+
F
p
+
F
am
+
F
b
(1)
where
F
dr
is the drag force (Fig. 4), given by (2):
F
dr
=
1
2
C
d
A
p
| u u
p
|( u u
p
) (2)
where C
d
is the drag coefcient (dened below) and A
p
is the particle cross-sectional area.
The instantaneous uid velocity and particle velocity are denoted by uand u
p
, respectively.
The subscript p denotes the particle/dispersed phase, non-subscripted quantities refer to the
continuous phase.
F
p
, the pressure force, is given by:
F
p
= V
p
p (3)
where V
p
is the particle volume and p the pressure gradient in the carrier uid; p includes
any hydrostatic components.
F
am
is the so-called virtual mass force, i.e. the force required
to accelerate the carrier uid displaced by the particle. The expression for this is:
F
am
= C
am
V
p
d( u
d
u)
d
t
(4)
where C
am
is the virtual mass coefcient, usually set to 0.5 (Milne-Thompson, 1968).
F
b
is a general body force term, which represents the effects of gravity and acceleration
present in a non-inertial coordinate frame. In the case of a rotating frame, F
b
becomes:
F
b
= m
p
[ g + ( r) +2( u
p
)] (5)
Z. Mazur et al. / Geothermics 33 (2004) 599614 605
where g is the gravitational acceleration vector, is the coordinate angular velocity vector
and r is the distance vector to the axis of rotation.
Knowledge of the particle velocity allows its instantaneous position vector x
p
to be
determined by integrating:
d x
p
d
t
= u
p
(6)
From Eq. (2), the momentum relaxation time scale
M
can be expressed as:
M
=
m
p
| u u
p
|
|
F
dr
|
=
2m
p
C
d
A
p
| u u
p
|
=
4
p
D
p
3C
d
| u u
p
|
(7)
where D
p
is mean diameter of the particles in m.
A drag coefcient, C
d
(Basset, 1888; Boussinesq, 1897; Oseen, 1910) is introduced to
account for experimental results on the viscous drag of a solid sphere. It depends on the
particle Reynolds number. The usual practice is, therefore, to obtain a drag coefcient
from correlations, derived from experiment or separate theoretical studies. The standard
correlation in the code is as follows:
C
d
=
24(1 +0.15 Re
0.687
p
)
Re
p
, Re
p
10
3
(8)
C
d
= 0.44, Re
p
> 10
3
(9)
where Re
p
is the particle Reynolds number, dened as (Adapco STAR-CD, 2001):
Re
p
| u u
p
|D
p
(10)
In turbulent ows, particle trajectories are not deterministic and two particles injected from
a single point at different times may follow separate trajectories due to the random nature
of instantaneous uid velocity. As a consequence, the particles disperse because of the
uctuating component of the uid velocity. To account for the inuence of turbulent uid
uctuations on particle motion, the method originally developed by Dukowicz (1980),
Gosman and Ioannides (1983), and Faeth (1987) has been applied in the code.
3.2. Erosion modeling
It is assumed that when a particle strikes the wall, the mass loss is distributed uniformly
over the computational cell in which the particle impinged. Approximation errors introduced
by this assumption should be reduced if the number of parcels is high enough and the grid
spacing next to the wall is kept reasonably small.
The following inlets conditions should be specied:
The total number of computational parcels (samples) introduced per unit time.
The particle size and velocity (vector) or the distributions of these quantities, if they are
not constant.
606 Z. Mazur et al. / Geothermics 33 (2004) 599614
Fig. 5. Nomenclature used for restitution coefcients.
When a particle strikes a solid wall, there will be an exchange of momentum that will
affect the particles subsequent passage through the uid. Restitution coefcients are used to
estimate this momentumexchange. The nomenclature for correlations that predict restitution
coefcient is illustrated in Fig. 5. u
Np
and u
Tp
are normal and tangential components of the
incoming velocity vector, respectively. and
r
are the particle impact and rebound angles.
Correlations are provided for both tangential and normal components of the impact and
are given by Eqs. (11) and (12) (Vittal and Tabakoff, 1987):
e
T
= 1.0 2.12 +3.0775
2
1.1
3
(11)
e
N
= 1.0 0.4159 +0.4994
2
0.293
3
(12)
The erosion model (Neilson and Gilchrist, 1968) was developed on the premise that two
different mechanisms are present in the erosion process: cutting and deformation. In effect,
the model is a simplied version of the models of Bitter (1963, Part 1 and Part 2), and is
intended to be used for the quick correlation of experimental data. The equations, as they
are implemented, are:
E
MM
= 0.001
(1/2)| u
p
|
2
cos
2
sin 2
C
+
(1/2)| u
p
|
2
sin
2
for
r
,
(13)
E
MM
= 0.001
(1/2)| u
p
|
2
cos
2
C
+
(1/2)| u
p
|
2
sin
2
for >
r
(14)
where E
MM
is the erosion rate (kg-material/kg-particles), u
p
is the particle impact velocity
(m/s), is the particle impact angle (radians),
r
is the rebound (cut-off) angle, and
C
and
D
are the cutting and deformation wear coefcients, respectively. The latter three parameters
are material-dependent and must be supplied by the user. The available parameters for AISI
4130 steel under liquid-sand erosion conditions (Wallace et al., 2000) were used because
this material has properties similar (chemical composition and hardness) to a rotor made of
low alloy CrMoV steel. It is known (Tabakoff and Eroglu, 1988) that the erosion rate is
related to the material hardness. The erosion rate is dened as the amount of mass lost due
to erosion divided by the mass of impacting particles.
Z. Mazur et al. / Geothermics 33 (2004) 599614 607
Fig. 6. Computational mesh of the low-pressure gland seal of the 25 MWe geothermal turbine.
The computational domain that contains the details of the geometry of the rotor low-
pressure end seal channel is represented by a grid, which was used for conducting the
investigation. This is shown in Fig. 6.
3.3. Boundary conditions
For the simulated conditions, particle diameter was set to 50 m, based on a microscopic
analysis of particles in geothermal steam (Mazur and Kubiak, 1996). Boundary conditions
(Fig. 7) were xed from ow conditions calculated using the thermal balance of the power
Fig. 7. Boundary conditions for the ow simulation.
608 Z. Mazur et al. / Geothermics 33 (2004) 599614
plant. A ow inlet boundary was used for the incoming seal steam ow, with temperatures
and pressures varying from 107.2 to 165