94 GT 126
94 GT 126
94 GT 126
94-GT-126
or printed in its publications. Discussion is printed only if the paper is published in an ASME Journal. Papers are available from ASME for 15 months
ABSTRACT
G,
L
m.
m
N
p
p
P
NOMENCLATURE
A
Cd
Cp ,,
Cw
cavity radius
Op/ 2 p(V z + U 2 )
r
ro
R
Re ,z
Ro
u
Ue
Urn
udi
U
radius
outer radius of cavity
radius ratio r/r a
axial Reynolds number pu e r o /p
rotation number tar o /u 5
axial velocity
average axial velocity for annulus flow
continuity derived average velocity
(in r-z plane) for seal
velocity in r-z plane
representative of disc pumping
mean axial velocity in potential
flow solution
Presented at the International Gas Turbine and Aeroengine Congress and Exposition
The Hague, Netherlands June 13-16, 1994
v
V
w
z
rl
77 C
ry e
0
p
p
4P
w
Subscripts
a
c
in
out
1
2
?le)/(77c
lie)
1 INTRODUCTION
In modern gas turbines, ingestion of the hot mainstream gas into the space between a rotating turbine
disc and a stator can severely affect disc temperatures.
As discussed more fully by Campbell (1978), the designer is faced with the problem of suppressing ingestion while minimising component weight and use of
cooling air. Although previous research has given some
insight into this problem, there is still a need for detailed experimental data and the development of mathematical models for the prediction of hot gas ingestion
and its effects. In this paper experimental and theoretical results are presented for the model problem of
ingestion into the cavity between plane rotor and stator discs with a simple axial clearance rim seal and an
asymmetric external flow.
Much of the previously published work on rim sealing
has concentrated on ingestion due to disc pumping.
For example, Bayley and Owen (1970) considered the
simple axial seal in the absence of external flow and
obtained a correlation for the minimum sealing flow
rate needed to prevent ingestion. Other workers (e.g.
Phadke and Owen (1988), Daniels et al (1990), Chew
et al (1991) and Bhavnani et al (1992)) have further
examined the influence of disc pumping, for various
seal geometries, and empirically based models are now
.._..._.._
....-.........e
seal gap
trace
probe
2 EXPERIMENTAL APPARATUS
A general outline of the rig, which is basically similar
to that used by Dadkhah et al (1991) and Green and
Turner (1992), is shown in Figure 1. This comprised
a rotor-stator system enclosed by an annular channel
through which the mainstream flow was drawn by a
large centrifugal fan located well downstream of the
seal gap; the total pressure and temperature upstream
of the NGVs were therefore atmospheric.
The central rotor-stator assembly was supported inside the external mainstream flow annulus by eight
aerodynamic struts upstream of the NGVs and eight
downstream of the rotor. No influence of the wake
from these struts was detected. The rotor was driven
by a radial inflow turbine at speeds up to 16000 rpm.
The stator and rotor faces were completely flat and
the wheelspace gap and the axial rim seal clearance
were maintained constant at 20 mm and 2 mm respectively (G = 0.1, G, = 0.01). The rotor and stator
shroud diameters were 400 mm and the constant diameter external flow annulus, of height 30 mm, contained
18 NGVs (20 pitching) individually moulded from an
epoxy/metal compound and mounted at one of four
alternative loations upstream from the seal edge, as
shown in Figure 2. The centrally fed cooling/sealing
flow was supplied from a separate source and was capable of delivering 0.15m 3 /s at a pressure of 1100 mm
W.G., and the coolant mass flow rate was measured by
an orifice plate.
2.1 Pressure
Static pressure tappings were made in the
wheelspace on three radial lines at nine radius ratios,
the most outer one being at R = 0.95. A single circumferential row of static pressure taps were made in
the mainstream exactly on the line of the trailing edge
of the nozzle guide vanes when fitted in position 2 to
measure the outlet static profile at the hub. This corresponded to a position 12.9 mm upstream of the edge of
the rim seal. All pressures, including those across the
wheelspace coolant flow orifice plate, were recorded using a micro-computer controlled Scanivalve.
2.2 Nitrous Oxide gas concentration
The detection and quantification of mainstream flow
ingestion was made by seeding the coolant supply to
the wheelspace with approximately 450 ppm of nitrous
oxide and sampling through four 1 mm bore tubes
at four radial positions, R = 0.163, 0.411, 0.658 and
0.905. The uncertainty in the gas concentration measurements, estimated at 2%, was due principally to
the averaging of the small rapid fluctuations in concentration experienced inside the wheelspace.
13 static
pressure
tappings
from SSto 79-5
Extemal tic
3 MATHEMATICAL MODELLING
3.1 Computational Fluid Dynamics (CFD)
Earlier CFD studies for rotating disc flows have
largely been confined to axisymmetric conditions, for
which quite extensive validations against experimental
data have been carried out. The computer program
used here is a 3D extension of the code described by
Vaughan et al (1989). The numerical methods embodied in the code include a finite volume discretisation of
the differential equations, the SIMPLEC pressure correction algorithm, and a non-linear multigrid acceleration technique. The equations solved are the Reynoldsaveraged Navier-Stokes equations with a mixing length
model of turbulence. In this study incompressibility
has been assumed, as this is a reasonable approximation of the experimental conditions. As details of the
w=0
(2)
repeated here.
To model the experimental conditions, a 20 0 sector
is considered with cyclic boundary conditions in the
circumferential direction (see Figure 3). The central
inlet to the cavity was specified as having a uniform
axial velocity with zero radial and tangential velocity
components. On solid walls the usual no slip velocity
boundary conditions were applied and zero derivative
conditions were assumed at the annulus flow exit.
v = V
u = U
c> = A
0.5.^p
P
VZ
0.5Ap
cos(NO + A)exp(zN/r) (4)
P V2 + U 2
0.50p
0. +
p( V2
U2)
ordinate system
r=o.2m
(r, 8, z), p
_ r-o.2m
r=o.197m
(1)
20 points in the 9-direction, indicated that discretisation errors were sufficiently low for an initial assessment. Complete mesh independence is not claimed,
however. Convergence of the solution could usually be
obtained in about 100 multigrid cycles, requiring about
90 minutes cpu time on an Amdahl 5990 1400.
3.2 Simpler Models
Some success with relatively simple models for predicting ingestion due to both disc pumping (e.g. Chew
et al, 1991) and mainstream pressure asymmetry (e.g.
Hamabe and Ishida, 1992) has been reported previously. In this paper, we concentrate on the effects of
pressure asymmetry and include comparison of experimental results with an elementary model. The development of more complete models is currently being
investigated and may be reported later.
Considering that the length scale of the circumferential pressure asymmetries will be much greater than
the seal clearance, it might be assumed that the rim
seal flow can be treated as locally two-dimensional with
orifice theory assumed to apply. Hamabe and Ishida
(1992) have published such a method and proprietary
methods based on this approach have been available
for some time (e.g. Campbell, unpublished work). According to this model, the cavity pressure p c is assumed
constant, any effects of the tangential velocity component are neglected, and the mean velocity through the
seal, u rn , is given by:
19
= C '2 1 P,
pal
(6)
Here Cd is a discharge coefficient and p a is the annulus pressure, u rn takes the sign of p c p a , and U r ,
Cd and p a may be functions of 9. For a given annulus
pressure distribution, integration of eqn. (6) around
the circumference allows the flow into and out of the
cavity to be determined as a function of p c . Performing
calculations for a range of p, values the ingested flow
rate may be obtained as a function of the net throughflow rate for the cavity. An estimate of the sealing
effectiveness is then given by:
measurement station
1400
s Re z =5.0x10 5
Rez=4.0x10S
10o Rezo3.0
x10
1200
1000
n
a 600
400
0
0
10 15
20 25 30 35 40 45 50 55
case, may help understanding of the experimental results of Green and Turner (1992).
It is also of interest to examine the conditions under which disc pumping will affect the rate of ingestion. From the model for disc pumping-driven ingestion, we can associate a dynamic head of 2 pud,,, with
the pumping flow for the axial clearance seal. Here
ud fz is a radial velocity associated with the disc pumping, and is typically of order 0.lwr. Thus we might
expect rotation to have little effect on ingestion when
0.005pw 2 r' /Op2 is small, where .Ap2 is the amplitude
of the circumferential pressure asymmetry. This order
of magnitude analysis for the effect of rotation is very
similar to that of Hamabe and Ishida (1992).
4000
3000
2000
4 RESULTS
4.1 Annulus Flow
Measured pressure distributions downstream of the
guide vanes are shown in Figure 4 for various annulus
flow rates with no rotation and no cooling flow. From
tests with the guide vanes in positions 2, 3 and 4 in
Figure 2, the pressure distribution downstream of the
vanes was determined. Figure 4 [i] shows the circumferential pressure variation just downstream of the vanes,
while Figure 4 [ii] shows the axial decay of the pressure asymmetry amplitude /p2. The non-dimensional
pressure Cpis defined as (P2 pl)/ 2 pue where P2 is
the measured pressure, and pl and u e are the static
pressure and axial velocity upstream of the vanes.
With the inlet asymmetry specified as described in
section 3.1, the CFD calculation gave a more rapid
axial decay of the pressure asymmetry than was measured on the rig. The CFD results are shown in Figure
5 which gives results at fixed annulus flow rate and rotational speed, for various levels of inlet asymmetry,
and for two different coolant flow rates. Close to the
seal the pressure field is affected by interaction with
the ejected cavity flow, so results for this region are
not included in this figure. The numerical results show
a slightly faster decay rate than eqn. (1) (which indicates a factor 0.53 in amplitude between z = 10 mm
and 17 mm on the figure). The 0-variation and the
relation between the amplitudes of the circumferential
variations were found to be roughly in agreement with
eqns. (1) to (4).
While the agreement between eqns. (1) to (5) and
the numerical results is reassuring, the comparison of
the calculated and measured decay rate for the pressure
asymmetry gives some cause for concern. The flow rate
of decay that has been measured may well be associated
with shortcomings in the vane design, leading to flow
1000
location
10
12
14
16
18
20
Figure 5: Axial variation of annulus pressure asymmetry from CFD solution; Re ; = 5 x 10 5 ; Ro = 1.6;
X u m /u e = 0.064 (Cw = 2000); 0 u,,,./u e = 0.194
(Cw = 6000)
1.00.80.60
0.4-
___
0.20 . 0 _t
10
15
20
25
30
u./u, n
[i]
1.00.80.6
-_
U
0.4 -
0.2 -
0.0 I
0
I
5
10
15
20
25
30
u./u m
'Iii'
Figure 6: Discharge coefficients without vanes for Ro =
0.0; [i] outflow; [ii] inflow
ranges 0 < Re ; /10 6 < 1.4, 3.2 < Cw/10 3 _< 11 and
0 < Re /10 6 < 1.2, 3.5 < Cw/10 3 < 14 for the outflow
and inflow cases respectively. The following equations
are shown to correlate the results reasonably well.
R Re,^510000
t Re,-510000
Re.-510000
X Re,-420000
+ Re,-290000
A Re,-510000
O Re,-410000
O Re,-290000
5
0.0
0.05
0.1
0.15
0.6-
0.5-
0.3
0.35
0.4
0.45
0.5
UI."
0.9-
0.7-
0.25
1.0-
0.8-
0.2
Ro-4.72
Ro.. 1.57
Ro-6.3
Ro-7.83
Ro-11.07
Ro-3.15
Ro-3.91
Ro-5.54
IIAPAIIIIII
IIUlIu
u..11I1 I
0.0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
0.5
[ii] Urn/tie
iiI
IIp.
proximated as 0.007Ro 2 << 1, and so the experimental results are consistent with the order of magnitude
arguments. Further scatter in the experimental results
might be attributed to Reynolds number effects including variation of the pressure asymmetry, mixing of the
flow in the cavity and seal discharge coefficient variations. One aspect of the measurements that is not fully
understood is the apparent levelling off of the effectiveness, at a value less than 1, as u,,,/u e increases. This
phenomena has been noticed in earlier studies with
guide vanes fitted (see Green and Turner, 1992) and
may be associated with the complex interaction of the
coolant and annulus flows. Note that the dynamic head
associated with the circumferential velocity in the annulus is quite large and disturbances of this flow could
give rise to significant, possibly unsteady, static pressure variations.
The results given in Figure 7 [ii], for the vanes in
position 2 (shown on Figure 2), include the case of zero
rotation. The absence of rotation will radically alter
the flow and mixing in the cavity, so the estimates of
sealing effectiveness for the zero rotation cases should
be viewed with caution. The results with rotation show
similar trends to those in Figure 7 [i].
With the vanes fitted in position 3 (see Figure 2),
the amplitude of the pressure asymmetry at the seal is
roughly half that with the vanes in position 1. The criterion, discussed above, for pressure asymmetry dominated ingestion therefore becomes 0.014Ro 2 << 1. It is
debatable whether the results in Figure 7 [iii] show this
predicted increased sensitivity to rotation, the scatter
iuiimH 111 I
0.0
0.05
0.1
0.15
0.2
[nil
0.25
0.3
0.35
0.4
0.45
0.5
U WUe
lower levels of sealing effectiveness than other conditions. This suggests that the ingestion is driven principally by the pressure asymmetry with disc pumping
becoming important at the extreme of the experimental
range. The criterion given in section 3.2 for rotational
effects to be negligible is that 0.005pw 2 r 2 /Op2 should
be small. For these conditions this can be roughly ap-
of rotational effects. A reduction in the level of ingestion as the vanes are moved further upstream can
be seen by comparing Figures 7 [i], [ii] and [iii]. It
may also be noted that the level of ingestion for the
case wr o /u e 11 is considerably less than would be
expected in the absence of external flow. This is attributed to the sealing effect of the crossflow as shown
by the discharge coefficients in Figure 6.
A comparison of CFD predictions with effectiveness measurements is shown in Figure 8. The nondimensional inlet pressure for the CFD calculations was
set to P = 2.5 here. This gives a similar level of pressure asymmetry just upstream of the seal to that in the
experiment with the vanes in position 1. The degree
of agreement between calculation and measurement is
perhaps as good as might be expected considering the
simplifications in the CFD model. This result is encouraging in that it suggests that it may not be necessary to capture all the details of the annulus flow to
gauge the effect of ingestion.
Ap e = 0.25pUe
10
1 01
-
NGV =3
0.9
0.8
0.8
0.7
07
0.6
0.6
0.5
05
APe = 0.75pUe
Um/Ue
a
+
NGV= ^
0.1
0.2
0.3
0.4
0.5
Um/Ue
--6 measurements Re
--- simple model
5x105, R o= 3
Figure 9 gives a comparison of effectiveness measurements with the simple model described in Section 3.2.
In these calculations eqn. (6) was used to model the
seal flow, with discharge coefficients given by eqns. (8)
and (9), and pressure asymmetries of the shape shown
in Figure 4 (for the higher annulus flow rate) were
assumed. Numerical methods were used to integrate
around the circumference of the cavity. Predictions
from the model are presented for pressure asymmetries equal to 75% and 25% of the axial dynamic head
for the annulus; these conditions are representative of
the extremes of the experimental range. The measurements shown are for a relatively low rotation number
(Ro .:s 3) and with the guide vanes fitted in positions
1 and 3. The main conclusion drawn from this figure
is that the simple model tends to overestimate the degree of ingestion. Further calculations using Dewynne
et al's (1989) result for the inflow discharge coefficient
showed some sensitivity to the choice of Cd,i,^ but confirmed this general conclusion.
4.4 The Seal and Cavity Flow
Since details of the disc cavity flow for a rotor-stator
in axisymmetric conditions have been discussed elsewhere (e.g. Chew and Vaughan, 1988), we concentrate
here on the three-dimensional flow that dominates the
present problem in the vicinity of the seal. An exam-
this case.
10
0.8
0-6
0-4
0.2
0-20
0-10
Iil
r(m)
0-20
0.10
[ul
r(m)
[il]
Figure 10: Velocity vectors (r z plane) in vicinity
of seal; Re z = 5 x 10 5 ; Ro = 1.6; u 1 /u e = 0.064 [i]
position of maximum outflow; [ii] position of maximum
inflow
10
5. CONCLUSIONS
The combined experimental and theoretical study
has given further insight into the rim sealing problem. For most of the experimental range covered, the
asymmetric annulus flow provided the dominant driving force for ingestion. At the higher values of rotation number tested, disc pumping increased the level
of ingestion, as is consistent with order of magnitude
arguments. In these cases the presence of the annulus
flow reduced ingestion relative to the case of a cavity
in a quiescent environment. At relatively low coolant
flow rates, the measurements are consistent with the
problem being dominated by the circumferential static
pressure asymmetries created by the guide vanes. At
higher coolant flow rates interaction of the coolant flow
(ejected through the seal) and the annulus flow may be
more important. It is conjectured that this interaction
gives rise to the observed levelling off of the sealing effectiveness (at values less than unity) as the coolant
flow is increased.
A simple model, neglecting rotation and assuming
orifice theory to apply locally around the circumference of the seal, shows some of the trends observed in
the measurements, but tends to over-estimate the level
of ingestion. This contrasts with the results of Hamabe and Ishida (1992) who found remarkable agreement
between a similar model and their experimental measurements. Note, however, that Hamabe and Ishida's
experiments were for a non-swirling external flow and
their choice of discharge coefficient in the model could
be questioned. The neglect of inertial effects, associated with the swirl component of velocity, is thought
to contribute to the over-estimation of ingestion in the
present study.
In the CFD model a potential flow solution was used
ACKNOWLEDGEMENTS
The authors gratefully acknowledge support for this
work by the Ministry of Defence and Rolls-Royce plc.
We are also pleased to acknowledge the contribution of
Dr. C. M. Vaughan, who programmed the CFD code,
and our colleagues who have contributed to helpful discussions of this work. In particular, the interest of Dr.
A. M. Cargill (now deceased), Mr. D. A. Campbell,
Mr. J. A. Millward and Professor T. V. Jones has
been appreciated.
References
[1] Abe, T., Kikuchi, J. and Takeuchi, H., 1979.
An investigation of turbine disc cooling. Paper
GT30, 3rd CIMAC Congress, Vienna.
[2] Bayley, F. J. and Owen, J. M., 1970. The fluid
dynamics of a shrouded disk system with a radial
outflow of coolant. ASME J. Engng Power, Vol.
92, pp. 335-341.
11
[15] Hamabe, K. and Ishida, K., 1992. Rim seal experiments and analysis of a rotor-stator system
with non-axisymmetric main flow. ASME paper
92-GT-160.
[4] Campbell, D. A., 1978. Gas turbine disc sealing system design. Proc. AGARD Conf. on Seal
Technology in Gas Tubine Engines, AGARDCP-237.
[18] Phadke, U. P. and Owen, J. M., 1988. Aerodynamic aspects of the rim sealing of gas-turbine
rotor-stator systems parts 1-3. Int. J. Heat and
Fluid Flow, Vol. 9, pp. 98-117.