Three-Dimensional CFD

Rotordynamic Analysis of Gas

Labyrinth Seals
J. Jeffrey Moore
Dresser-Rand Co.,
Paul Clark Dr.,
Olean, NY 14760
e-mail: jeff moore@dresser-rand.com

Labyrinth seals are utilized inside turbomachinery to provide noncontacting control of

internal leakage. These seals can also play an important role in determining the rotordynamic stability of the machine. Traditional labyrinth seal models are based on bulk-flow
assumptions where the fluid is assumed to behave as a rigid body affected by shear stress
at the interfaces. To model the labyrinth seal cavity, a single, driven vortex is assumed and
relationships for the shear stress and divergence angle of the through flow jet are developed. These models, while efficient to compute, typically show poor prediction for seals
with small clearances, high running speed, and high pressure.* In an effort to improve the
prediction of these components, this work utilizes three-dimensional computational fluid
dynamics (CFD) to model the labyrinth seal flow path by solving the Reynolds Averaged
Navier Stokes equations. Unlike bulk-flow techniques, CFD makes no fundamental assumptions on geometry, shear stress at the walls, as well as internal flow structure. The
method allows modeling of any arbitrarily shaped domain including stepped and interlocking labyrinths with straight or angled teeth. When only leakage prediction is required,
an axisymmetric model is created. To calculate rotordynamic forces, a full 3D, eccentric
model is solved. The results demonstrate improved leakage and rotordynamic prediction
over bulk-flow approaches compared to experimental measurements.
DOI: 10.1115/1.1615248

Introduction
Labyrinth seals are widely used in turbomachinery to minimize
secondary leakage inside turbomachinery, including gas turbines
and compressors. Accurate leakage prediction is vital for reliable
performance prediction. The large gas pressure of high-pressure
centrifugal compressors develops forces inside the many labyrinth
seals that can affect the rotordynamic stability of the machine.
This research attempts to improve the prediction accuracy of these
forces by employing 3D, CFD techniques solving the general
Reynolds Averaged Navier-Stokes RANS equations along with
the energy equation, equation of state, and turbulence model. The
approach does not rely on empirical wall and interface constants
that may change for varying applications and geometry. Furthermore, the exact geometry of the seal may be modeled allowing
optimization of the profile of the teeth. The obvious drawback of
CFD compared to bulk-flow is increased computational requirement but is manageable with modern workstations.

Literature Review
Several authors have developed bulk-flow approaches to model
the dynamic forces of labyrinth seals including Iwatsubo 1,
Childs and Scharrer 2, Kirk 3, and Marquette, et al. 4. Use of
bulk-flow has been moderately successful for liquid labyrinth
seals but yields only fare predictions for gas labyrinths. Its primary advantage is efficient computational times.
Other approaches have utilized the rapidly maturing computational fluid dynamic analysis for modeling annular seals including
Dietzen and Nordman 5 and Arghir and Frene 6. These authors
employ a coordinate transformation from the 3D eccentric domain
into a 2D axisymmetric one. This technique requires a zeroth order solution and first order calculations at the different whirl frequency ratios. The procedure is efficient, but the coordinate transContributed by the Technical Committee on Vibration and Sound for publication
in the Jrgen Lund Special Issue of the JOURNAL OF VIBRATION AND ACOUSTICS.
Manuscript received June 2003. Guest Associate Editor: R. Gordon Kirk.
*Childs, D.W., 1993, Turbomachinery Rotordynamics, John Wiley & Sons, New
York, p. 345.

Journal of Vibration and Acoustics

formation is only valid for constant radius seal geometry, and

most analyses ignore the variation of turbulence quantities around
the circumference. Recent work by Kim and Rhode 7 eliminate
the coordinate transformation, making it valid for varying radii.
However, the approach has only been validated against liquid
seals to date. Using these 2-D approaches, only axisymmetric seal
geometry may be modeled, preventing the modeling of swirl
brakes for example.
Three-dimensional approaches including Tam, et al. 8, Nordmann and Dietzen 9, Rhode, Hensel, and Guidry 10, and
Athevale 11 have been developed which solve the 3D Reynolds
Averaged Navier Stokes equations of a whirling rotor. This approach is the most general and most compuationally intensive.
However, these authors only make correlations to liquid seal results. Moore and Palazzolo 12 have utilized and compared the
approach to Laser Doppler Anemometer LDA data and force
measurements for liquid labyrinth seals. Moore and Palazzolo
13 have also validated the approach for predicting rotordynamic
force coefficients of shrouded pump impellers. Recent work by
Kwanka, Sobotzik, and Nordmann 14 compare the 3D approach
to gas labyrinth measurements showing similar trends as shown in
the measurements. Unfortunately, only test pressures to 100kPa
were used.
The CFD code SCISEAL 15 used in the present study utilizes
a 3D whirling method developed by Athavale, et al. 11. This
approach is more computationally intensive but is more general in
the class of problems that may be solved. This work represents the
first use of this code to validate against gas labyrinth seal
measurements.

CFD Code Description

The Reynolds-averaged Navier-Stokes equations shown in
Eqs. 12 using index notation are solved in strongconservative form utilizing SIMPLE-C solver. The advection
scheme used by the code is the Upstream Weighted Difference
Scheme UWDS. The code allows solutions for both rotating and
stationary frames of reference, using body fitted structured grids
of multi-domains.

Copyright 2003 by ASME

Downloaded From: http://asmedigitalcollection.asme.org/ on 02/18/2015 Terms of Use: http://asme.org/terms

OCTOBER 2003, Vol. 125 427

Table 1 Constants used in the k - model

continuity

U j

0.0
t
x j
momentum

(1)

U i U iU j
p

Ui U j

eff

t
x j
xi x j
x j
xi

Ul
2
eff
k i j S iu
3
xl

2
Ul
eff
k ij
3
xl

T t h

x j
x j
Prt x j

Xr 2
V 2

2
2

(4)

(5)

The current work assumes the entire flow domain to be turbulent and implements the standard 2-equation model with wall
functions. The expression for turbulent kinetic energy (k) and the
dissipation rate , are given below along with the relation for
turbulent viscosity.
(k equation

P k t

t k
P k
k x j

Ui U j Ui 2
Uk Ul

k t
x j
xi x j 3
xk xl

equation

U i

t
xi
x j

Prt

.09

1.44

1.92

1.

1.3

0.9

u y

(6)

t
c 1 P k c 2
x j
k
(7)

c u k 2

for y 12

1
u ln Ey
K
where

for y 12

yU
,

and

(9)

u
U

(10)
(11)

(12)

In order to obtain accurate wall shear stress, care is taken to

ensure the near wall grid points are placed in the logarithmic
region. Although not shown, the code also solves the energy equation and equation of state to model the compressible flow. A
SIMPLE-C algorithm is utilized for pressure-velocity coupling.
The upwinding scheme consists of first order upwinding blended
with second order central difference.
The CFD code SCISEAL was developed by CFD Research
under contract from NASA Glenn. Additional details may be
found in the input manual 15.

Rotordynamic Solution Technique

where H is the total enthalpy, h is the static enthalpy, and T is the

static temperature. The last term in brackets is also a consequence
of the eddy viscosity assumption for the turbulent fluctuating energy correlation, u i h. This form of the energy equation is
widely used for compressible flow. For rotating domains, the
Rothalpy (I), is used instead of total enthalpy and the velocity is
defined in the relative frame of reference.

k U ik

t
xi
x j

(3)

H U iH p

Ui U j

U i eff

t
xi
t x j
x j
xi

Ih

c 2

(2)

is the rotational speed of the model and r is the distance

where
vector from the axis of rotation. This term is zero for stationary
flow components.
Since the flow is compressible, the energy equation along with
an equation of state is used.
energy

c 1

Law of the wall formulations model the sharp velocity gradients near the wall:

Note that the first term in Eqs. 1 and 2 vanishes for the current
work since only the steady terms are considered. The 2/3 i j k
term is a consequence of the eddy viscosity model of the Reynolds stress tensor. Also, eff represents the sum of the kinematic
viscosity and the turbulent viscosity ( t ). Equation 2 contains a source term accounting for the centripetal and Coriolis
effects.

X
Xr 2
XV
S iu

(8)

The flow field inside a whirling annular seal is inherently unsteady. However, for a centered seal the whirling, unsteady problem is transformed into a steady one by solving the threedimensional, eccentric flow field in the frame of reference
attached to the whirling rotor. In this rotating frame, the stator
wall moves in the opposite direction for positive whirl. The rotor
surface moves with, against, or not at all relative to the whirling
frame of reference, depending on the processional frequency ratio
PFR, defined as the ratio of rotor whirl to rotor spin. Quasisteady solutions are obtained at each PFR. A PFR equal to unity is
termed synchronous whirl, where the shroud is whirling at the
same frequency it rotates. A PFR of zero indicates a static displacement of the shroud, which then simply spins.
The choice of eccentricity is arbitrary but is typically kept near
10% of the shroud clearance to capture the linear, small motion
characteristics. This value was arrived at based on previous modeling of annular seals 12. Larger eccentricities are possible and
may be utilized to predict nonlinear characteristics.
A solution is obtained at multiple PFR values typically ranging
from 0.0 to 2.0. For gas annular seals, the impedance curves are
essentially linear no inertia coefficient, therefore a minimum of
two whirl frequencies are required to solve for the rotordynamic
coefficients. For this study, three precessional frequency ratios
PFR are used 0.0, 0.5, and 1.0 for better accuracy of the
curvefits.
After a converged solution is obtained at each PFR, a pressure
integration is performed on the rotor surface in the two orthogonal
directions yielding F t and F n , respectively. The shear stress contribution at the wall contributes less than 1% of the total impedance force and is ignored. These impedance forces may be normalized by yielding the impedance both normal and tangential
to the whirl orbit. A comparison of these impedance forces to a
linear, first order model is shown in Eqs. 13 and 14.

The constants for Eqs. 68 are defined in the Table 1.

428 Vol. 125, OCTOBER 2003

Downloaded From: http://asmedigitalcollection.asme.org/ on 02/18/2015 Terms of Use: http://asme.org/terms

F n

Fn

Kc

(13)

Transactions of the ASME

Fig. 1 2D grid for inlet, 1st tooth, and cavity

F t

Ft

kC

Fig. 4 Mesh density study meridional

(14)

Each of the above equations represent one equation and two

unknowns. The force coefficients are evaluated by determining the
impedance force at multiple processional frequencies . The
coefficients of the curve-fit yield the seals stiffness and damping
force coefficients (K, k, C, c). The generality of this approach
allows modeling of all variety of geometry including both seals
and impeller shroud passages. As demonstrated by Moore and
Palazzolo 12, the method can also predict the inertia coefficients
of liquid seals.

Grid Generation
The 2D grid for the inlet and first seal cavity are created using
a commercial grid generator. A FORTRAN code was developed
which sweeps a 2D grid transforming it into an eccentric 3D grid.
The code was further refined to automatically create repeating
geometry like labyrinth teeth. Therefore, only the inlet region, the
first seal land, and the first cavity grid need to be defined as shown
in Fig. 1. The land and cavity grids are then repeated by N number
of times, where N is the number of labyrinth teeth minus one.
This code allows 3-D grids to be made quickly.
The grid is made eccentric to the height of the land by 10% of
the clearance. Starting with the one node away from the wall, the
grid points are redistributed linearly in the radial direction. The
near-wall grid points are not disturbed so as to maintain a nearly
constant y , resulting in improved wall shear stress prediction.
Figure 2 shows 2D view of the resulting 3D full seal model for an
18 tooth labyrinth seal showing the repeating seal geometry.

generate pre-swirl. The long straight section contains stationary

walls and consists of 3 grid blocks. The spinning rotor begins at
the step just upstream of the seal.
Several modeling approaches were used including a seal-only
grid, upstream added Up-Seal, then downstream sections added
Up-Seal-Dn. When the upstream region is included, velocity is
specified thus fixing the mass flow rate. Density is calculated by
specifying the expected inlet static pressure and temperature. For
models where only the seal is modeled, a total pressure boundary
condition used. For the seal only model, the calculated inlet
boundary conditions including preswirl are obtained from a 2-D
model with the upstream region included.
Mesh Density Study. Before comparisons are made to experiment, a mesh independence study is performed to determine
how fine a grid is required to capture the important flow physics.
The grid is refined in the meridional plane radial-axial and circumferential direction independently. A seal-only model is utilized
for this study using the following operating conditions:
16,000 rpm
75.69 mm
0.118 mm
8
3.175 mm
3.175 mm
0.152 mm
18.2 bars
7.34 bars
0 m/s

N
Shaft Radius
Radial Clearance
# Teeth
Tooth Pitch
Tooth Height
Tip Width
Inlet Tot. Pressure
Exit Tot. Pressure
IGV Swirl

Benchmark Case8 Tooth Labyrinth Seal

In order to benchmark the code, a test case for an 8 tooth,
teeth-on-stator labyrinth seal presented by Pelletti 16 was modeled. This data represents the most complete set of labyrinth seal
data in the literature using test pressures up to 300 psig 2.1 MPa.
Figure 3 shows a 2D view of the computational grid. The model
begins at the exit of the inlet guide vanes, which are used to

Fig. 2 3D labyrinth seal model

Fig. 3 Pelletti up-seal grid

An axisymmetric calculation is made including the upstream region to determine the actual conditions at the An inlet of the seal
used with the Seal-Only model. The mass-flow averaged values
are:
Seal
Seal
Seal
Seal
Seal

Inlet
Inlet
Inlet
Inlet
Inlet

Total Pressure
Swirl Ratio ( s )
Static Temp.
k

17.3 bars
0.12
297 K
477.4
1.55e8

Figure 4 shows the effect of refining the grid in the meridional

plane as shown in Fig. 3. The cross-coupled stiffness Kxy
varies less than 5% with increasing mesh density as shown in Fig.
4 and is essentially mesh independent. The direct stiffness Kxx,
however, becomes mesh independent with the medium density
grid. Therefore, the medium grid is effectively mesh independent
for both quantities and will be used in subsequent calculations.
Figure 5 shows the effect of circumferential node count using
the medium density grid from above. Mesh independence is
achieved at around 61 nodal points around the circumference.

Journal of Vibration and Acoustics

Downloaded From: http://asmedigitalcollection.asme.org/ on 02/18/2015 Terms of Use: http://asme.org/terms

OCTOBER 2003, Vol. 125 429

Fig. 7 Static pressure distribution through seal

Fig. 5 Mesh density study circumferential

However, due to the large resulting model size, 51 points was

chosen as a compromise and still gives essentially mesh independent results.
The resulting mesh independent grid used in this study for the
Pelletti seal is summarized below:
Block

Location

1
2
3
4, 6, . . .
5,7, . . .

Inlet
Inlet
Inlet
Land
Cavity

IJK
604051
124051
183151
141451
304351
Total 715,836 nodes

CFD Results Axisymmetric Seal. In addition to calculating simply forces with CFD, the entire flow field within the labyrinth seal may be viewed to give insight into the nature of the
flow. This section describes the flow field from the axisymmetric
seal model 3 nodes in circumferential direction, 1 pie slice.
Figure 6 shows the predicted velocity vectors inside one labyrinth
seal cavity. Notice the high through-flow jet, which diverges and
creates a stagnation point on the next tooth. The jet drives a large
vortex inside the cavity. Figure 7 shows the predicted pressure
breakdown across the seal. Most all of the pressure drop occurs
across the teeth. Initially, the seal clearance provided by Pelletti

Seal 1 was used adjusted for centrifugal growth equaling 0.176

mm 6.9 mils. When specifying the experimental leakage, the
predicted delta-P was substantially lower. Conversations with
other researchers observed similar trends when comparing to this
data. The author believes that nominal clearances were reported
by Pelletti 16 and were not confirmed by measurement. Therefore, the clearance was reduced until the predicted leakage correlated with the experiment. The resulting clearance is 0.118 mm
4.6 mils.
Inlet guide vanes IGVs were used in an attempt to induce
preswirl in the experiment. This swirl is modeled starting with the
exit of the IGVs see Fig. 3. The swirling flow must travel an
axial distance of about 2 inches 50 mm. As shown in Fig. 8, the
predicted swirl reduces significantly before reaching the seal entrance. In fact, the step near the seal entrance, where the spinning
rotor begins, influences the swirl entering the seal more than the
IGVs.
The resulting seal inlet swirl ratio and delta-P is summarized in
Tables 2 and 3. Note, the pressure ratio is defined as the exit
pressure divided by the inlet pressure. The swirl ratio is the average swirl at that location divided by the rotor surface velocity.
Notice that even for a hypothetical negative inlet guide vane swirl
not actually tested, the resulting swirl at the seal entrance is still
positive. The explanation for this phenomena is simple. The swirl
velocity component is much larger than the axial velocity component. Therefore, the flow swirls around many times in the upstream region creating an effective flow length much longer than
the axial length. Since both upper and lower walls are stationary,
most of the swirl simply dies out before it reaches the entrance of
the seal. The spinning rotor then generates some positive preswirl.
This example demonstrates how CFD can be used to better understand and improve the design of experimental test rigs and as well
as turbomachinery.
These quantities are used in subsequent calculations.
Rotordynamic Force Prediction. To solve for rotordynamic
forces, a full 360, eccentric grid is required as previously described. Figure 9 shows the streaklines traveling through the laby-

Fig. 8 Circumferential swirl prediction upstream of seal

darkhigh swirl
Table 2 Predicted swirl and delta-P 16 krpm, Pr0.65
s Exiting
IGV

Calculated
s Entering
Seal

Calc.
delta-P

Exper.
delta-P

0.35
0.00
0.35

0.11
0.20
0.24

610,500
611,200
600,900

640,000
621,000

Table 3 Predicted swirl and delta-P 16 krpm, Pr0.403

s Exiting
IGV

Calculated
s Entering
Seal

Calc.
delta-P

Exper.
delta-P

0.00
0.35

0.12
0.22

993,600
833,700

1,097,000
1,088,000

Fig. 6 Velocity vectors in labyrinth seal cavity

430 Vol. 125, OCTOBER 2003

Downloaded From: http://asmedigitalcollection.asme.org/ on 02/18/2015 Terms of Use: http://asme.org/terms

Transactions of the ASME

Fig. 12 Tangential impedance, Pr0.65, 16 krpm, s 0, Seal

Only

Fig. 9 Streaklines through seal

rinth seal. Notice the spiraling nature of the flow. Figure 10 plots
the static pressure around the circumference variation about mean
pressure for the non-whirling case (WFR0). The plot starts at
the minimum clearance and travels around in the direction of rotation. The phase shift creates the tangential force Kxy and opposes whirl in this case negative Kxy.
Using the mesh independent grid and the inlet boundary conditions obtained from the axisymmetric calculations, the impedence forces were obtained for the two different inlet swirl values,
as plotted in Fig. 11 along with the experimental curves. This
model only contains the seal geometry and not the up or downstream regions. The predictions show the same trend as the ex-

periment where increasing swirl increases the tangential impedance. The cross-coupled stiffness (y-intercept is being
underpredicted as is the damping slope. Both prediction and
experiment show a linear impedance curve.
Figure 12 shows a similar plot but for a pressure ratio of 0.65
( s 0). The predictions show good agreement for cross-coupled
stiffness, but under-predicts the damping slightly.
In an effort to improve the predictions, the upstream region is
included as part of the seal model Up-Seal. Figure 13 shows an
improvement for the zero swirl, Pr0.403 case. Also plotted are
results using the bulk-flow code of Kirk 3 as well as the code of
Childs and Scharrer 2. Adding the upstream to the model improves the cross-coupled stiffness prediction. The Kirk code
shows reasonable agreement, but not as good as the CFD. The
Scharrer code shows the greatest deviation. It should be noted that
the same inlet boundary conditions are used in the bulk-flow cases
as in the CFD model. Table 4 summarizes these coefficients.
The addition of the upstream region Up-Seal improves the
Kxy prediction and the Kxx somewhat. Adding the downstream
block changes the Kxy only slightly but dramatically improves the
direct stiffness calculation. Both bulk-flow codes greatly underpredict in negative direction the direct stiffness though they provide slightly better damping prediction. Overall, the Up-SealDown CFD model provides the best results.
The effect of seal clearance is shown in Fig. 14. The Seal 1
clearance 0.118 mm shown previously was increased by 82% for
Seal 3 as reported by Pelletti resulting in a clearance of 0.216 mm
8.5 mils. These values were all corrected for centrifugal growth.
The experiment shows a reduction in cross-coupled stiffness and
damping as a result of the larger clearance. CFD shows a similar
trend of lower Kxy and Cxx, but the reduction in the experimental
data is more pronounced. Some of the difference may lie in the
uncertainty of the actual seal clearance tested.

Fig. 10 Circumferential pressure distribution

Fig. 11 Tangential impedance, Pr0.403, 16 krpm, Seal-Only

Fig. 13 Comparison of seal models, 16 krpm, Pr0.403,

s 0

Journal of Vibration and Acoustics

Downloaded From: http://asmedigitalcollection.asme.org/ on 02/18/2015 Terms of Use: http://asme.org/terms

OCTOBER 2003, Vol. 125 431

Table 4 Summary of rotordynamic force coefficients, Pr

0.403, s 0
Model
CFDSeal Only
CFDUp-Seal
CFDUp-Seal-Dn
Scharrer Bulk Flow
Kirk Bulk Flow
Experiment

Kxx

Kxy

Cxx

138,000
93,600
18,200
510,300
592,200
28,340

158,000
67,000
76,300
280,000
120,000
33,150

147
69
87
205
134
493

Fig. 14 Effect of seal clearance, Pr0.398, 16 krpm, s 0, upseal model

Leakage Prediction Comparison

Since the exact seal clearance in the Pelletti tests were not
accurately known, a valid leakage comparison could not be made.
Therefore, comparisons are made to leakage data taken during
testing of a seal design from the authors company consisting of
an 18 teeth-on-stator labyrinth seal configuration Fig. 15. In ad-

Mass flow
Inlet Temp.
Centrifugal Growth
s
Fluid

0.18 lbm/s 0.082 kg/s

70F (21C)
0.000183 4.65 m
0
Nitrogen gas

Figure 16 shows the presence of two counter-rotating vortices,

the large one driven by the through-flow jet and a smaller one near
the top of the cavity. Figures 17 and 18 compare the predicted
pressure breakdown through the seal versus experiment for the
non-rotating and 15,000 rpm cases, respectively.
Again, the mass-flow is fixed and the pressure drop is predicted.
The four experimental points at each axial station represent the 4
circumferential pressure tap positions 90 apart. The non-rotating
case Fig. 17 shows good correlation but slightly over-predicts
the pressure drop. The rotating case Fig. 18 shows good agreement for total pressure drop including exit loss at the seal exit.

Fig. 15 18 tooth labyrinth seal

Fig. 16 Velocity vectors, DR 18 tooth Laby

dition to leakage, intra-seal pressure taps were measured at three

axial locations and at four circumferential positions (90 apart.
An axisymmetric model was created including part of the upstream region. Comparisons are made at zero and 15,000 rpm.
Figure 16 shows the predicted velocity vectors inside the seal
cavity. The operating conditions and geometry are given below:

Shaft Radius
Radial Clearance
Pitch
Tooth Width

2.250
0.007
0.110
0.011

57.15 mm
0.178 mm
2.79 mm
0.279 mm

CFD is not only useful for prediction but is also a powerful tool
for design and troubleshooting. Using CFD a problem was discovered with the inlet guide vane arrangement in the experimental
test rig. The swirl generated all but vanished before reaching the
seal entrance. This result helps to explain the rather poor correlation to this data in the past.

Conclusion
This work benchmarked a CFD code used to model gas labyrinth seals. The CFD predictions show good correlation to experimental data for both rotordynamic force coefficients and leakage,
an improvement over bulk flow predictions. The addition of the up
and down stream regions in the CFD model further improves the
predictions, though in doing so increases the computational time
of the model somewhat. Leakage comparisons to in-house measurements show good agreement for both rotating and nonrotating cases.

Fig. 17 Static pressure tap measurement vs. prediction for 18

tooth DR Laby, 0 rpm

432 Vol. 125, OCTOBER 2003

Downloaded From: http://asmedigitalcollection.asme.org/ on 02/18/2015 Terms of Use: http://asme.org/terms

Transactions of the ASME


s
K

t
eff

turbulent kinetic energy dissipation rate

swirl ratioavg. circum. swirl/surface speed
von Karman constant0.4
molecular viscosity
turbulent eddy viscosity
effective viscosity t
gas density
wall shear stress
rotational speed rad/sec
whirl speed rad/sec

References
Fig. 18 Static pressure tap meas. vs. prediction for 18 tooth
DR Laby, 15 krpm

The drawback of CFD is computational time. The mesh independent grid for the 3D case contains over 700,000 nodes and
requires approximately 20 hours for one whirl frequency ratio
minimum of two required. Therefore, at the present time, CFD is
reserved for designing new seals and when the utmost in accuracy
is required. If only leakage is required ie. axisymmetric, the
model size is much smaller and solutions may be had in as little as
30 minutes. Solvers using advanced techniques such as multi-grid
reduces the computational time but with some memory penalty.
Bulk-flow continues to be used for day-to-day lateral rotordynamic stability analysis in industry. As computational power increases in the next few years, use of CFD techniques will become
a standard part of the design process for modeling gas forces in
turbomachinery.

Acknowledgments
The author would like to thank Dr. Lee Hill of Dresser-Rand
Co. for his assistance in the project. Thanks also goes to DresserRand Co. for funding this work.

Nomenclature

ij

E
Fn , Ft
Fn , Ft
K, C
k, c
k
N
P
PFR
Pr
Sij
u
ui
Ui
U
y

surface roughness parameter9.0

normal radial and tang. reaction force
normal radial and tang. impedance (F/ )
direct stiffness, damping
cross-coupled stiffness, damping
turbulent kinetic energy
speed rpm
local static pressure
processional frequency ratio/
pressure ratioPexit/Psupply
mean strain rate
velocity component tangent to wall
fluctuation velocity component
mean velocity component
shear velocity
distance from wall to first nodal point
eccentric displacement of rotor in seal
Kronecker delta

1 Iwatsubo, T., 1980, Evaluation of Instability Forces of Labyrinth Seals in

Turbines or Compressors, Proceedings of Rotordynamic Instability Problems
in High Performance Turbomachinery, NASA CP-2133, Texas A&M University, pp. 139167.
2 Childs, D. W., Scharrer, J. K., 1986, An Iwatsubo Based Solution for Labyrinth Seals: A Comparison to Experimental Results, ASME J. Eng. Gas Turbines Power, 108, pp. 325331.
3 Kirk, R. G., 1988, Evaluation of Aerodynamic Instability Mechanisms for
Centrifugal CompressorsPart II: Advanced Analysis, ASME J. Vibr.
Acoust., 110, April, pp. 207212.
4 Marquette, O., Childs, D. W., and Philips, S. G., 1997, Theory Versus Experiments for Leakage and Rotordynamic Coefficients of CircumferetiallyGrooved Liquid Annular Seals with L/D of 0.45, ASME Paper No. FED
SM97-3333, Proceedings of the 1997 ASME Fluids Engineering Division Summer Meeting, June 2226.
5 Dietzen, F. J., and Nordmann, R., 1987, Calculating Rotordynamic Coefficients of Seals by Finite-Difference Techniques, ASME J. Tribol., 109, pp.
388 394.
6 Arghir, M., and Frene, J., 1997, Rotordynamic Coefficients of
Circumferentially-Grooved Liquid Seals Using the Averaged Navier-Stokes
Equations, ASME J. Tribol., 119, pp. 556 567.
7 Kim, N., and Rhode, D. L., 2000, A New CFD-Perturbation Model For The
Rotordynamics of Incompressible Flow Seals, ASME International Gas Turbine and Aeroengine Congress and Exposition, May 8 11, Munich, Germany.
8 Tam, L. T., Przekwas, A. J., Muszynska, A., Hendricks, R. C., Braun, M. J.,
and Mullen, R. L., 1988, Numerical and Analytical Study of Fluid Dynamic
Forces in Seals and Bearings, ASME J. Vibr. Acoust., 110, pp. 315325.
9 Nordmann, R., and Dietzen, F. J., 1988, Finite Difference Analysis of Rotordynamic Seal Coefficients For An Eccentric Shaft Position, NASA CP 3026.
10 Rhode, D. L., Hensel, S. J., and Guidry, M. J., 1992, Labyrinth Seal Rotordynamic Forces Using a Three-Dimensional Navier-Stokes Code, ASME J.
Tribol., 114, pp. 683 689.
11 Athevale, M. M., Przekwas, A. J., Hendricks, R. C., and Liang, A., 1994,
SCISEAL: A 3D CFD Code for Accurate Analysis of Fluid Flow and Forces
in Seals, Proceedings of the Advanced ETO Propulsion Conference, NASA
CP3282, NASA Marshall Space Flight Center, May, Huntsville, AL, pp. 337
345.
12 Moore, J. J., and Palazzolo, A. B., 1999, CFD Comparison to 3D Laser
Anemometer and Rotordynamic Force Measurements for Grooved Liquid Annular Seals, ASME J. Tribol., 121, No. 2, pp. 307314.
13 Moore, J. J., and Palazzolo, A. B., 1999, Rotordynamic Force Prediction of
Centrifugal Impeller Shroud Passages Using Computational Fluid Dynamic
Techniques, ASME International Gas Turbine and Aeroengine Congress and
Exposition, June 912, Indianapolis, Indiana.
14 Kwanka, K., Sobotzik, J., and Nordmann, R., 2000, Dynamic Coefficients Of
Labyrinth Gas Seals A Comparison Of Experimental Results And Numerical
Calculations, ASME International Gas Turbine and Aeroengine Congress and
Exposition, May 8 11, Munich, Germany.
15 Athavale M. M., and Przekwas, A. J., SCISEAL Manual, 1995, SCISEAL: A
Computer Program for Study of Fluid Dynamic Forces in Seals, Developed
under contract by NASA Lewis Research Center NAS3-25644.
16 Pelletti, J., 1990, A Comparison of Experimental Results and Theoretical
Predictions for the Rotordynamic Coefficients of Short (L/D1/6) Labyrinth
Seals, M.S.M.E. Thesis, Texas A&M University and Turbomachinery Laboratory Report No. TL-Seal-1-90.

Journal of Vibration and Acoustics

Downloaded From: http://asmedigitalcollection.asme.org/ on 02/18/2015 Terms of Use: http://asme.org/terms

OCTOBER 2003, Vol. 125 433