You are on page 1of 11

Proceedings of ASME Turbo Expo 2009: Power for Land, Sea and Air

GT2009
June 8-12, 2009, Orlando, Florida, USA

GT2009-59256

PREDICTED ROTORDYNAMIC BEHAVIOR OF A LABYRINTH SEAL AS ROTOR


SURFACE SPEED APPROACHES MACH 1

Manish R. Thorat Dara W. Childs


Research Assistant Leland T. Jordan Professor of Mechanical
Turbomachinery Laboratory, Texas A&M Engineering, Turbomachinery Laboratory,
University Texas A&M University, College Station,TX-77843

ABSTRACT sound. For the present example, observable discrepancies arose


Prior one-control-volume (1CV) models for rotor-fluid when Rsω = 0.26 c0.
interaction in labyrinth seals produce synchronously-reduced
(at running-speed), frequency-independent stiffness and INTRODUCTION
damping coefficients. The 1CV model, consisting of a leakage Labyrinth seals are used widely in turbomachinery to
equation, a continuity equation, and a circumferential- reduce leakage. However, they are known to cause lateral
momentum equation (for each cavity) was stated to be invalid rotordynamic instabilities.
for rotor surface speeds approaching the speed of sound. A tooth-on-stator/smooth-rotor labyrinth is considered here.
However, the present results show that, while the 1CV fluid- Several models have been developed to predict the
mechanic model continues to be valid, the calculated rotordynamic coefficients for this seal. Kurohashi et al. [1] and
rotordynamic coefficients become strongly frequency Iwatsubo [2] developed a 1CV analysis. Childs and Scharrer [3]
dependent. modified the 1CV model to include angular area derivative in
A solution is developed for the reaction-force components the continuity and momentum equations. Wyssman et al. [4]
for a range of precession frequencies, producing frequency- developed the first two-control-volume (2CV) model using a
dependent stiffness and damping coefficients. They can be box-in-box model to account for the through flow and vortex
used to define a Laplace-domain transfer-function model for the flow in the labyrinth cavity. An alternative 2CV model was
reaction-force/rotor-motion components. Calculated developed by Scharrer [5]. However, experimental results by
rotordynamic results are presented for a simple Jeffcott rotor Picardo [6] showed that the 1CV model by Childs and Scharrer
acted on by a labyrinth seal. The seal radius Rs and running yielded better predictions than Scharrer’s 2CV approach.
speed ω cause the rotor surface velocity Rsω to equal the speed Analysis using these models produced synchronously-reduced,
of sound c0 at ω=58 krpm. frequency-independent results based on the assumption that the
Calculated synchronous-response results due to imbalance lowest acoustic (circumferential direction) natural frequency in
coincide for the synchronously-reduced and the frequency- the seal cavity is much higher than the rotor speed. Picardo’s
dependent models. measured results produced frequency-independent coefficients.
For an inlet preswirl ratio of 0.5, both models predict the If the Rsω is comparable to c0, frequency dependency of the
same log decs out to ω≈14.5 krpm. The synchronously-reduced coefficients is expected. The present analysis uses the model of
model predicts an onset speed of instability (OSI) at 15 krpm, Childs-Scharrer but adopts the analysis procedure by Thieleke
but a return to stability at 45 krpm, with subsequent increases in and Stetter [7].
log dec out to 65 krpm. The frequency-dependent model
predicts an OSI of 65 krpm. The frequency-dependent models NOMENCLATURE
predict small changes in the rotor’s damped natural frequencies.
The synchronously-reduced model predicts large changes. A cross-sectional area of control [L2]
The stability-analysis results show that a frequency- volume
dependent labyrinth seal model should be used if the rotor B Tooth height [L]
surface speed approaches a significant fraction of the speed of C, c direct and cross coupled damping [F t/L]

1 Copyright © 2009 by ASME


c0 speed of sound [L/t]
FX, FY seal reaction forces [F]
fr ,fθ radial and circumferential [F/L]
dynamic stiffness coefficients
H seal Clearance [L]
j −1
K,k direct and cross coupled stiffness [F/L]
kr Jeffcott rotor stiffness [F/L]
L pitch of seal strip [L]
m& leakage flow rate [M/L t]
P pressure [F/L2]
R Gas constant [L2/Tt2]
Rs radius of seal [L]
s complex variable for Laplace [1/t] Fig. 2 Radial View of Labyrinth Seal Cavity and
transform Control Volume
T Temperature [T]
t time [t] Continuity and Circumferential momentum equations are
U circumferential flow velocity [L/t] derived for this control volume based on the analysis by Childs
u0(0) U(0)/Rsω, Preswirl ratio and Scharrer [3]. The gas is ideal and temperature is constant.
θ circumferential coordinate
ρ density [M/L3] Continuity Equation:
τ shear stress [F/L2]
ν kinematic viscosity [L2/t] ∂ ρi AU ∂
ω rotor speed [1/t] ( i i
) + ( ρi Ai ) + m& i +1 − m& i = 0 (1)
∂θ Rs ∂t
Ω Excitation frequency [1/t]

SUBSCRIPTS Circumferential Momentum Equation:


0,1 zeroth and first order
2
i Seal cavity index ∂ ρi AU ∂
( i i
) + ( ρi AU
i i)+ m
& i +1U i − m& iU i −1
r,θ radial and circumferential ∂θ Rs ∂t
R,S rotor and stator (2)
Ai ∂Pi
=− + (τ Ri aRi − τ Si aSi ) Li
Rs ∂θ
THEORETICAL MODEL
where aR and aS are dimensionless lengths upon which shear
The labyrinth seal model is based on the control volume stress acts.
shown in Figs. 1 and 2. Rotor and stator shear stresses are modeled using a Blasius
shear-stress model. To account for the curvature effects, the
shear stress terms are modified according to Martinez-Sanchez
STATOR et al. [8].
| Rsω − U i | Dhi Dhi
τ Ri = τ Ri × (1.0 + 0.075( )0.25 )
Pi-1 Control Volume Pi+1 ν 2 Rs
Ui-1 Pi | U i | Dhi Dhi
Ui Ui+1 τ Si = τ Si × (1.0 + 0.075( )0.25 )
ν 2 Rs
m& i m& i +1 where, Dhi is the hydraulic diameter defined by
2( H i + Bi ) Li
Dhi =
ROTOR ( H i + Bi + Li )

Leakage Model:
Fig. 1 Axial View of Labyrinth Seal Cavity and Control Pi −12 − Pi 2
Volume m& i = μ1i μ 2 H i (3)
RT

2 Copyright © 2009 by ASME


where, μ1i is the Chaplygin’s flow coefficient [9], μ2 is the Because of the separate sine and cosine terms, and the two
Neumann’s kinetic energy carryover coefficient [10]. independent parameters a and b in these two equations, two
independent equations are developed from each equation to
PERTURBATION ANALYSIS solve for the four unknowns K, k, C, c.
The solution from this approach is assumed to be valid if
A perturbation analysis is used to solve the governing the lowest acoustic natural frequency in the cavity is much
equations. The eccentricity ratio ε is the perturbation parameter higher than the rotor speed. However if the rotor speed
for small motion about a centered position. The zeroth-order approaches the lowest cavity natural frequency, the frequency-
equations define the leakage flow rate, plus steady-state independent solution breaks down. The natural frequency
circumferential velocities and cavity pressures at each cavity. considered here is that of a standing wave in the circular
The first-order equations define the perturbations in pressure annulus of a labyrinth seal. To determine the acoustic
and circumferential velocity of flow due to rotor motion. frequency, the circular annulus can be considered to be an
open-ended pipe. If the flow velocity within the pipe is
First-Order Perturbation Equations: negligible compared to the sonic velocity, and continuity is
enforced at the two ends of the pipe, the natural frequencies are
Continuity Equation iπ c0 2iπ c0 ic0
ωni ( pipe) = ⇒ ωni (cavity ) = = ; i = 1, 2,...
∂P G U ∂P G P ∂U1i L 2π Rs Rs
G1i 1i + 1i 0i 1i + 1i 0i + G3i P1i
∂t Rs ∂θ Rs ∂θ Hence, the running speed that equals the 1st acoustic
(4) resonance is:
∂H1 G2iU 0i ∂H1
+G4i P1i −1 + G5i P1i +1 = −G6i H1 − G2i − c
∂t Rs ∂θ ω = ωn1 = 0 ⇒ Rsω = c0 (10)
Rs
Circumferential Momentum Equation Note that the limiting condition ω = ωn1 is the same as the
∂U1i X1iU 0i ∂U1i A0i ∂P1i rotor surface velocity approaching the acoustic velocity;
X 1i + +
∂t Rs ∂θ RS ∂θ (5) i.e., Rsω = c0 . Isothermal sonic velocity, c0 = RT is
+ X 2iU1i − m& 0U1i −1 + X 3i P1i + X 4i P1i −1 = X 5i H1 considered in the present analysis, as the 1CV model is
isothermal.
Here, U1i , P1i are the perturbed circumferential velocities and
Circular Orbit Solution
pressures in the ith cavity, and H1 is the first order clearance Using an approach developed by Childs and Kim [11] for
perturbation. Due to insufficient space, the coefficients G1i and liquid annular seals, Thieleke and Stetter [7] developed a
X1i are not given in this paper; however they can be referred to solution for Eqs.(4-5) assuming a circular orbit solution of the
in ref. [3]. form:
X =a cosΩt , Y =a sinΩt (11)
Elliptic-Orbit Solution Substituting this solution into Eq.(6) gives
For small motion about a centered position, labyrinth seal FX = − Ka cos Ωt + caΩ sin Ωt
reaction forces are typically modeled as (12)
FY = ka cos Ωt − CaΩ sin Ωt
Resolving these components into radial term that is parallel to
⎧ FX ⎫ ⎡ K k ⎤ ⎧ X ⎫ ⎡ C c ⎤ ⎧⎪ X& ⎫⎪ the rotating vector a and to a circumferential term that is
−⎨ ⎬ = ⎢ ⎥⎨ ⎬+ ⎢ ⎥⎨ ⎬ (6)
⎩ FY ⎭ ⎣ −k C ⎦ ⎩Y ⎭ ⎣ −c C ⎦ ⎩⎪Y& ⎭⎪ perpendicular to a nets:
Fr
= f r = − ( K + cΩ )
Starting with Iwatsubo, the solution for the four unknowns a
(13)
k, K, c and C, was obtained by assuming an elliptical orbit of Fθ
the form: = fθ = k − C Ω
a
X =a cosωt , Y =b sinωt (7) Equation (13) provides two equations for the four
Where the rotor is assumed to be precessing at the running unknowns; however, using two precession frequencies provides
speed ω. The clearance function associated with this solution is: enough equations. Providing more than two frequencies creates
ε H1 = − a cos ωt cos θ − b sin ωt sin θ (8) a circumstance where f r (Ω), fθ (Ω) are curve fitted to solve for
A separation-of-variables approach is used for Eqs.(4-5) for the unknowns. Thieleke and Stetter did not consider
this clearance excitation. Substituting Eq.(7) into Eq. (6) yields: circumstances where Ω → ωn1 ; i.e. ,where the frequency-
FX = −( Ka + cbω ) cos ωt − (kb − Caω ) sin ωt independent model of Eq.(6) would become invalid. The
(9)
FY = (ka − Cbω ) cos ωt − ( Kb + caω ) sin ωt

3 Copyright © 2009 by ASME


example in the following section, explains the inadequacy of 3.0E+07

using the frequency independent model when Ω → ωn1 2.0E+07

fr ,Radial Dynamic Stiffness


1.0E+07
Test Case for a Long Labyrinth Seal

Coefficient (N/m)
The test conditions, seal geometry and operating conditions 0.0E+00
used in this analysis are obtained from the tests conducted by ‐1.0E+07
Picardo [6]. The input data are shown in Table 1. Air is used in
Picardo’s tests. The mass flow rate for the given operating ‐2.0E+07
conditions at rotor speed of 20200 RPM and preswirl ratio of ‐3.0E+07
0.578 is 0.373 kg/sec [6].The predicted exit axial flow Mach
‐4.0E+07
number is 0.29
‐5.0E+07
Table 1 Geometric, Operating, and Input Data
‐6.0E+07
Reservoir Pressure 70.149 bar
0.0 1.0 2.0 3.0 4.0 5.0
Sump Pressure 36.694 bar
Temperature 288.559 K Non-Dimensional Frequency
Radial Clearance 0.198 mm Fig. 3a Radial Dynamic Stiffness Coefficient Vs
Seal Radius 57.340 mm Excitation frequency, (Preswirl ratio=0)
Tooth Pitch 4.293 mm
Tooth Height 4.293 mm 0.0E+00
Rotor Friction Constant 0.079
Rotor Friction Exponent -0.250 fθ ,Circumferential Dynamic ‐1.0E+07
Stator Friction Constant 0.079 Stiffness Coefficient (N/m)
‐2.0E+07
Stator Friction Exponent -0.250
Compressibility Factor 1.000 ‐3.0E+07
Ratio of Specific Heats 1.400
‐4.0E+07
Kinematic Viscosity 0.00001510 m2/s
Gas Constant 286.900 J/kg K ‐5.0E+07
Number of Teeth 20
Tooth Location Stator ‐6.0E+07

‐7.0E+07
Speed Influence on Acoustic Resonance Location
Figures 3a and 3b show fr and fθ versus non-dimensional ‐8.0E+07
excitation frequencies for ω =15.2 krpm. The excitation 0.0 1.0 2.0 3.0 4.0 5.0
frequency Ω is normalized with respect to the rotor speed ω. Non-Dimensional Frequency
The plots show the resonant peaks of the 1st acoustic mode of
the labyrinth cavity. As ω increases, the dynamic-stiffness Fig. 3b Circumferential Dynamic Stiffness Coefficient
coefficients at the rotor speed are strongly influenced by Vs Excitation frequency, (Preswirl ratio=0)
resonance. Thus, the fr and fθ coefficients can no longer be
defined by linear expressions of Eq.(13). Figure 4, shows the predicted change in the acoustic
resonant frequency versus ω. For lower rotor speeds, the
resonant frequency closely matches the predicted isothermal
acoustic frequency of 5022 rad/sec from Eq. (10). As the 1CV
model is isothermal, acoustic frequency is evaluated using
isothermal sonic speed, c0 = RT . For higher rotor speeds,
the resonant frequency increases with increasing ω. This
dependency can be attributed to (i) the circumferential flow
velocity in the annulus becoming an appreciable fraction of c0,
and (ii) convective acceleration terms that are retained here but
neglected in conventional acoustics.

4 Copyright © 2009 by ASME


8000 Data and Curve Fit; 5 poles, 4 zeros D(s)
160
Acoustic Frequency (rad/sec)

7000 Curve Fit


Data

Gain dB
6000 140

5000 120

4000 100 1 2 3 4
10 10 10 10
3000 Freq. [rad/sec]
2000 -100

Phase [Deg]
1000 -200

0
-300
0 10000 20000 30000 40000 50000 60000 70000
Rotor Speed (RPM) -400 1 2 3 4
10 10 10 10
st Freq. [rad/sec]
Fig.4 Predicted 1 acoustic damped natural
frequency versus running speed. (Preswirl ratio=0)
Fig. 5 Magnitude and Phase Plot of Calculated and
Curve-fitted D transfer function (Zero preswirl, Rotor
Transfer-Function Models for Reaction Force Components Speed 15200 RPM)
If Rsω approaches c0, fr and fθ cannot be modeled by the
frequency-dependent model of Eq. (6). A similar situation
arises in honeycomb/hole-pattern stator seals where the Data and Curve Fit; 5 poles, 4 zeros E(s)
150
apparent acoustic velocity for flow within the seal can be Curve Fit
reduced due to the effect of gas compressibility within the 140
Data
Gain dB

holes/cells, dropping the lowest acoustic frequency within the 130


operating region. Kleynhans and Childs [12] present solutions
that produce frequency-dependent rotordynamic coefficients for 120
these types of seals, following the approach of Bolleter et al. 110 1 2 3 4
[13]. Their transfer-function model is 10 10 10 10
Freq. [rad/sec]
⎧ FX ( s ) ⎫ ⎡ D( s ) E ( s ) ⎤ ⎧ x( s ) ⎫
−⎨ ⎬=⎢ ⎥⎨ ⎬ (14) 0
⎩ FY ( s ) ⎭ ⎣ − E ( s ) D( s ) ⎦ ⎩ y ( s ) ⎭
Phase [Deg]

-100

with D and E defined as follows: -200

-300
f r + (Ω) = − Re[ D( jΩ)] − Im[ E ( jΩ)] -400 1 2 3 4
10 10 10 10
f r − (Ω) = − Re[ D( jΩ)] + Im[ E ( j Ω)] Freq. [rad/sec]
(15)
fθ + (Ω) = + Re[ E ( jΩ)] − Im[ D( jΩ)] Fig. 6 Magnitude and Phase Plot of Calculated and
Curve-fitted E transfer function (zero pre-swirl, Rotor
fθ − (Ω) = + Re[ E ( jΩ)] + Im[ D( jΩ)] Speed 15200 RPM)
Here, the ‘+’ power indicates positive excitation frequencies
and ‘-’ power indicates negative excitation frequencies. The The following D(s) and E(s) transfer functions were obtained
complex functions D(jΩ) and E(jΩ) are obtained by adding and by curve fitting fr and fθ for 0-8000 rad/sec:
subtracting terms in Eq.(15).
Analytical expressions are obtained for D and E by curve- −4.7 × 1012 s 4 − 1.1× 1016 s 3 − 1.3 × 1020 s 2 − 1.3 × 1023 s + 2.8 × 1025
D(s) =
fitting to standard polynomial forms. Figures 5 and 6 provide s 5 − 2.7 × 105 s 4 − 6.5 × 108 s 3 − 1.4 × 1013 s 2 − 1.7 × 1016 s − 1.7 × 10 20 (16)
representative results for ω = 15200 rpm and zero-preswirl. −1.24 × 108 s 4 − 2.4 × 1013 s 3 + 3.6 × 1017 s 2 + 3.6 × 1019 s − 2.3 × 10 23
E ( s) =
s 5 + 3.2 × 103 s 4 + 5.4 × 107 s 3 + 1.0 × 1011 s 2 + 6.8 × 1014 s + 4.8 × 1017

In Eq. (16), D(s) has unstable poles that are discounted in the
stability analysis.

5 Copyright © 2009 by ASME


JEFFCOTT ROTOR MODEL-IMBALANCE RESPONSE
transfer function model
AND STABILITY ANALYSIS 1.0E+02

Magnitude Ratio (X/e)


synchronously reduced
1.0E+01
coefficient model
Labyrinth seal 1.0E+00
1.0E-01
1.0E-02
1.0E-03
1.0E-04
1.0E-05
Bearing
1.0E-06
10 100 1000 10000 100000
Fig. 7 Simple Jeffcott Rotor Rotor Running Speed (RPM)
Fig. 8 Magnitude Plot for frequency-independent
The Jeffcott rotor model of fig. 7, acted on by labyrinth seal model (synchronously reduced coefficients) and
forces is used to consider the effect of frequency dependency frequency dependent model (Transfer Function
on synchronous response and stability analysis. Model) for u0(0) = 0.5
The rotor parameters are:
Mass (m) 100 kg STABILITY CALCULATIONS
kr
Natural frequency , 795.87 rad/sec = 7.6 krpm Speed-Dependent, Frequency-Independent Model
m Approach
This natural frequency is half the rotor speed of 15.2 krpm. The frequency-independent model is implemented
This choice was made to amplify any possible impact of the considering synchronously-reduced rotordynamic coefficients
labyrinth on rotor stability. The labyrinth seal parameters used evaluated using elliptical precessional orbit at the rotor running
here are shown in Table 1. Labyrinth seal forces are the only speed. The homogeneous version of Eq.(6) applies.
source of viscous damping in the model.
Frequency-Dependent Models
Imbalance-Response calculations Frequency dependency can be accounted by using Eq.(6)
The governing equation for the Jeffcott rotor model of figure 7 with frequency dependent rotordynamic coefficients, with the
with conventional labyrinth seal forces is: corresponding precession frequencies equal to the rotor’s
damped natural frequencies. This approach is regularly used to
x ⎫ ⎡ C c ⎤ ⎧ x& ⎫ ⎡ K + kr
⎡ m 0 ⎤ ⎧ && k ⎤ ⎧x ⎫ account for calculated frequency dependency of tilting-pad
⎢ 0 m ⎥ ⎨ && ⎬+ ⎢ ⎥⎨ ⎬+ ⎢ ⎨ ⎬
K + kr ⎥⎦ ⎩ y ⎭
bearings. Thus, the rotordynamic coefficients for forward and
⎣ ⎦ ⎩ y ⎭ ⎣ −c C ⎦ ⎩ y& ⎭ ⎣ −k (17) backward damped natural frequencies can be extracted from the
⎧⎪meω 2 cos ωt ⎫⎪ D and E transfer functions via:
=⎨ ⎬
⎩⎪meω sin ωt ⎭⎪
2
K + = Re[ D( jΩ nr + )] K − = Re[ D( jΩnr − )]

where kr is the rotor stiffness and e is the rotor imbalance. k + = Re[ E ( jΩ nr + )] k − = Re[ E ( jΩ nr − )]
Comparisons were made for the speed-dependent Im[ D( jΩ nr + )] Im[ D( jΩ nr − )]
(frequency-independent) model and the frequency and speed C+ = C− = (18)
dependent model for a range of inlet swirl ratios and rotor Ω nr + Ω nr −
speeds. Figure 8 shows the (same) calculated amplitude results Im[ E ( jΩ nr + )] Im[ E ( jΩ nr − )]
for both models. Although not shown, the phase plots also c+ = +
c− =
coincide. This outcome applies because, for response to Ω nr Ω nr −
imbalance, the rotordynamic coefficients are calculated for
forward precession at ω for both model types. Where, Ωnr is the damped natural frequency of the rotor, and
superscripts + and - indicate rotor’s forward and backward
modes.
For hole-pattern stator seals, seal forces can significantly
change the damped natural frequencies of the rotor. In such
cases, the rotordynamic coefficients are re-evaluated at the
calculated damped natural frequencies, and the procedure is

6 Copyright © 2009 by ASME


repeated until there is convergence between the assumed and 0.06 synchronously reduced coefficients
calculated natural frequency. Thus, the stability analysis
0.05

Logarithmic Decrement
becomes iterative in nature. However, Labyrinth seal forces do non-synchronously reduced coefficients
not significantly change the rotor’s damped natural frequencies, 0.04 transfer function model
and in the present example the rotor’s critical speed closely
approximates the first forward and backward damped natural 0.03
frequencies. The stability analysis in this example is non-
0.02
iterative with the evaluation of rotordynamic coefficients at the
rotor’s critical speed. 0.01
K = Re[ D( j Ωcr )]
0
k = Re[ E ( jΩcr )]
Im[ D( jΩcr )] -0.01
C= (19)
Ωcr -0.02
Im[ E ( jΩcr )] 0 0.2 0.4 0.6 0.8
c=
Ωcr Preswirl Ratio
Where, Ωcr is the rotor critical speed. Fig. 9 Log-Dec versus u0(0) for Forward Critical Speed
The second frequency-dependent approach involves directly
implementing the D and E transfer-function results into the
rotor model using a state-space format. TEST CASE 2, ω = 40 krpm; Rsω = 0.7c0
Figures 9-11 present calculated log-dec results for the Figure 10 shows the following outcomes:
model versus the inlet preswirl ratio u0(0), which is varied from
• The synchronously-reduced model predicts instability for
0 to 0.8. Zero corresponds to a highly-effective swirl brake,
forward precession at all preswirl ratios up to 0.8.
and 0.8 corresponds to a high preswirl value as might be
• The frequency-dependent and transfer-function models
expected for a balance-piston seal with no swirl brake. Results
predicts instability at a preswirl ratio greater than ~ 0.66
were obtained using the following three approaches:
Although not illustrated, the improved models predict reduced
1. Speed dependent, frequency-independent model.
stability for the backward-precessing mode as compared to
2. Frequency-dependent stiffness and damping
predictions from the synchronously-reduced model.
coefficients
3. D and E transfer-function model.
Results are presented for both forward and backwards-
0.25 synchronously reduced coefficients
precessing roots.
0.2
Logarithmic Decrement

non-synchronously reduced coefficients


TEST CASE 1, ω = 15.2 krpm; Rsω = 0.26 c0 0.15
transfer function model
Figure 9 shows the following outcomes for forward precession: 0.1
• The synchronously reduced model predicts instability for
0.05
forward precession at a pre-swirl ratio of ~0.57.
• The frequency-dependent model predicts instability at a 0
pre-swirl ratio of ~0.74 -0.05
• Transfer Function Model predicts instability at a preswirl -0.1
ratio of ~0.76. -0.15
Although not presented, the improved models predict that the
backward-precessing mode is more stable than predicted by the -0.2
synchronously-reduced model. -0.25
0 0.2 0.4 0.6 0.8
Preswirl Ratio

Fig. 10 Log-Dec Vs Pre-swirl Ratio for Forward


Critical Speed

TEST CASE 3 ω = 70 krpm; Rsω = 1.2c0


Figures 11a and 11b illustrate the following outcomes:
•For the forward precessing mode, the synchronously-
reduced model predicts stability for 0 ≤ u0(0) ≤ 0.8

7 Copyright © 2009 by ASME


•The frequency-dependent model predicts instability at u0(0) dependent model predicts an onset speed of instability of 65
≈ 0.43 krpm.
•The transfer-function model predicts instability at u0(0) ≈
0.51
•For the backwards precessing mode, the improved models
predict stability for 0 ≤ u0(0) ≤ 0.8; the synchronously
reduced model predict instability for 0 ≤ u0(0) ≤ 0.8.

1.0 synchronously reduced coefficients


Logarithmic Decrement

non-synchronously reduced coefficients


0.8
transfer function model
0.6

0.4

0.2

0.0 Fig. 12a Log-dec Vs rotor speed for forward whirling


mode
-0.2 For the backwards-precessing mode, figure 12b also shows
0 0.2 0.4 0.6 0.8 about the same calculated log dec for speeds out to ~14 krpm.
The synchronously-reduced model predicts an onset speed of
Preswirl Ratio instability for the backward mode at ~50 krpm. The frequency-
Fig. 11a Log-dec versus u0(0) for forward whirling dependent model predicts a continuing increase in the log dec
mode as the running speed increases. Hence, the rotor backward
precessional mode is predicted to remain stable for the
1.0 complete speed range.
synchronously reduced coefficients
0.8
Logarithmic Decrement

non-synchronously reduced coefficients

0.6 transfer function model

0.4

0.2

0.0

-0.2

-0.4

-0.6
0 0.2 0.4 0.6 0.8
Preswirl Ratio
Fig. 11b Log-Dec versus u0(0) for backward whirling
mode
Fig. 12b Log-dec Vs rotor speed for backward
whirling mode
The historical experience with labyrinth seals is that they
Evaluation of onset speed of Instability:
produce low values of direct stiffness and have a minimal
The onset speeds of instability for forward and backward
impact on the rotor natural frequencies. Figure 13a shows the
precessing modes are evaluated for u0(0)=0.5. Figure 12a
calculated damped natural frequency for the rotor’s forward
illustrates the predicted log dec for the forward precessional
precession mode. The synchronously-reduced model shows it
mode, showing that stability calculations are about the same for
first dropping and then increasing sharply as the ω increases.
synchronously-reduced and frequency-dependent solution up to
The frequency-dependent results show a modest drop in the
~14 krpm, but diverge from this speed onwards. The frequency-

8 Copyright © 2009 by ASME


natural frequency with increasing ω. A comparison of figures presented for a Jeffcott model acted on by a labyrinth seal,
13a and 12a suggests that differences between the log-dec using: (i) A synchronously-reduced (traditional) model, and (ii)
predictions for the models arise mainly due to erroneous frequency-dependent models. Frequency dependency can be
predictions of the damped natural frequency for the accounted for by considering frequency dependent stiffness and
synchronously-reduced model. damping coefficients or by using transfer-function models. The
transfer-function models can be integrated into a conventional
rotordynamic code by converting them into a state-space form.
The traditional speed-dependent (but frequency independent)
stiffness and damping coefficients continue to be valid for
synchronous imbalance response.
Rotordynamic calculations were performed for a Jeffcott
model acted on by a labyrinth seal. The rotor diameter at the
seal produces a predicted surface velocity equal to the speed of
sound at ω=58 krpm. Log-dec predictions from the two model
types coincide for speeds out to about 15k rpm. The frequency-
dependent models predict an onset speed of instability (OSI)
around 65 krpm. The frequency-independent model predicts an
initial OSI at ~15k rpm followed by a return to predicted
stability around 45 krpm. The stability results show that the
traditional model is not valid for rotor surface velocities
approaching a significant fraction of the speed of sound.
The new frequency dependent model needs to be validated
by experiments. Test data by Picardo [5], provides dynamic
Fig. 13a Damped natural frequencies Vs rotor speed
stiffness coefficients for a frequency range of excitation of 10
for forward whirling mode
to 150 Hz. However, to verify the predictions in this analysis,
test data are needed from rotors running at high rotor surface
velocities (approaching the speed of sound), or the excitation
frequency range must be expanded significantly beyond the
rotor running speed.

ACKNOWLEDGMENTS
The work reported here was supported by the
Turbomachinery Research Consortium of the Texas A&M
University Turbomachinery Laboratory.

REFERENCES
[1] M. Kurohashi, et al., 1980, “Spring and Damping
Coefficients of the Labyrinth Seals,” Paper No. C283/80,
Proceedings of the 2nd International Conference on Vibrations
Fig. 13b Damped natural frequencies Vs rotor speed in Rotating Machinery (Institution of Mechanical Engineers),
held at Churchill College, Cambridge University, pp. 215-222.
for backward whirling mode
[2] Iwatsubo, T., 1980, “Evaluation of Instability Forces of
Figure 13b presents calculated results for backward- Labyrinth Seals in Turbines or Compressors,” NASA CP 2133
precessing natural frequency. The frequency-dependent model Proceedings of a workshop at Texas A&M University 12-14
shows a slow drop for ω out to about 48 krpm, followed by a May 1980, entitled Rotordynamic Instability Problems in High
gradual increase. The synchronously-reduced model shows an Performance Turbomachinery, pp.205-222.
initial sharp drop ending at about 27 krpm, followed by a sharp [3] Childs, D., Scharrer, J., 1986, “An Iwatsubo-Based Solution
rise as ω increases further. for Labyrinth Seals: Comparison to Experimental Results”,
Journal of Engineering for Gas Turbines and Power, Vol.108,
pp.325-331.
SUMMARY, DISCUSSION, AND CONCLUSIONS
The 1CV model for a see-through labyrinth seal continues [4] Wyssman, H., Pham, T., and Jenny, R. (1984), "Prediction
to be used; however, the solution procedure is modified to of Stiffness and Coefficients for Centrifugal Compressor
account for frequency-dependency as the rotor’s surface Labyrinth Seals," ASME J. of Engineering for Gas Turbines and
Power, 106, 920-926.
velocity approaches the speed of sound. Calculated results are

9 Copyright © 2009 by ASME


[5] Scharrer, J., 1987, “A Comparison of Experimental and circular whirl orbit:
uur
Theoretical Results for Labyrinth Gas Seals,” Ph.D. H 1 = x(t ) + jy (t ) = r0 e jΩt
dissertation, Texas A&M University. ur
[6] Picardo, A., 2003, “High Pressure Testing of See-Through P1i = p1i e jΩt
Labyrinth Seals”, M.S. thesis, Texas A&M University. ur
U 1i = u$1i e jΩt
[7] Thieleke, G., Stetter, H., 1990, “Experimental Investigations
of Exciting Forces caused by Flow in Labyrinth Seals”, TAMU Where r0 is the radius of the circular whirl orbit. p1i and
Instability Workshop.
u$1i are complex amplitudes:
[8]Martinez-Sanchez, M., Lee, O.W.K., Czajkowski, E., 1984,
“The prediction of Force Coefficients for Labyrinth Seals”, p1i = P1ic + jP1is
Nasa CP 2338,Rotordynamic Instability Problems in High
u$1i = U1ic + jU1is
Performance Turbomachinery, proceedings of a work held at
Texas A&M University, pp. 235-256.
[9]Childs, D., Chang-Ho Kim,1984, “Analysis and testing of The separation of variables approach yields algebraic
Rotordynamic Coefficients of Turbulent Annular Seals with equations of the form:
Different, Directionally Homogeneous Surface-Roughness [ A1 ]{ x1}i −1 + [ A2 ]{ x1}i + ⎡⎣ A3 ⎤⎦{ x1}i +1 = { x2 }
Treatment for Rotor and Stator Elements”, Nasa CP Where,
2338,Rotordynamic Instability Problems in High Performance
Turbomachinery, proceedings of a work held at Texas A&M
{ x1}i = {P1ic }
T
University, pp. 313-340. P1is U1ic U1is
[10]Gurevich, M.I., 1966, “The Theory of Jets in an Ideal T
⎧ G2iU0i GΩ X ⎫⎪
{ x2}i = ⎪⎨−
Fluid”, Pergamon Press, pp. 115-122. G
r0 + 2i r0 6i r0 0 − 5i r0 ⎬
[11]Neumann, K., 1964, “Zur Frage der Verwendung von ⎪⎩ Rsò ò ò ò ⎪⎭
Durchblickdichtungen im Dampfturbinenbau,” ⎡ 0 G4i 0 0 ⎤
Maschinentechnik, Vol.13, No.4, pp.188-195. ⎢G 0 0 0 ⎥
[12]Kleynhans, G., and Childs, D., 1997, “The Acoustic [ A1] = ⎢⎢ 04i X 0 −m& ⎥⎥
Influence of Cell Depth on the Rotordynamic Characteristics of ⎢
4i 0

Smooth-Rotor/Honeycomb-Stator Annular Gas Seals,” ASME ⎣ X4i 0 −m &0 0 ⎦
Trans., J. of Engineering for Gas Turbines and Power, Vol. ⎡ G1iU 0i G1i P0i ⎤
119, No. 4, pp. 949-957. ⎢G1i Ω − R G3i −
Rs
0 ⎥
⎢ s ⎥
[13] Bolleter, U., Leibundgut, E., Stürchler, R., and McCloskey, ⎢ GU G1i P0i ⎥
T., 1989, “Hydraulic Interaction and Excitation Forces of High ⎢ G3i −G1i Ω + 1i 0i 0 ⎥
Rs Rs
Head Pump Impellers,” in: PumpingMachinery—1989, [ A2 ] = ⎢⎢ A X 1iU 0i


Proceedings of the Third Joint ASCE/ASME Mechanical ⎢ − 0i X 3i X1i Ω − X 2i ⎥
Conference, La Jolla, CA, pp. 187-194. ⎢ Rs Rs ⎥
⎢ A0i X 1iU 0i ⎥
⎢ X 3i X 2i − X 1i Ω ⎥
APPENDIX A ⎢⎣ Rs Rs ⎥⎦

Circular Orbit Solution:


⎡ 0 G5i 0 0 ⎤
The seal clearance function can be defined as: ⎢G
òH1 = − x(t ) cos θ − y (t ) sin θ 0 0 0 ⎥⎥
⎡⎣ A3 ⎤⎦ = ⎢
5i
⎢ 0 0 0 0⎥
⎢ ⎥
The pressure and velocity perturbations can be expressed ⎣ 0 0 0 0⎦
as: The above set of equations is for a single cavity. The cavity
P1i = P1ic cos θ + P1is sin θ
U1i = U1ic cos θ + U1is sin θ
equations are assembled to form a 4NC ൈ 4NC matrix equation
Substituting the above variable definitions in the first order
continuity and momentum equations eliminates the theta
(NC= number of cavities). Pressure & velocity perturbations at
dependency. The time dependency is eliminated by considering
the entry and exit are zero. These form the boundary conditions
for the problem.

10 Copyright © 2009 by ASME


The solution is of the form:
r
P1ic = 0 P1′ic
ò
r0
P1is = P1′is
ò
The radial and circumferential forces, Fr (Ω) and Fθ (Ω) ,
on the seal are evaluated from the pressure perturbations in the
labyrinth seal cavities.
NC
Fr (Ω) = − Rs r0π ∑ p′
i =1
1ic

Similarly,
NC
Fθ (Ω) = − Rs r0π ∑ p′
i =1
1is

The radial and circumferential dynamic stiffness


coefficients are obtained from these forces:
F (Ω)
f r (Ω) = r
r0
Fθ (Ω)
fθ (Ω) =
r0

11 Copyright © 2009 by ASME

You might also like