NASA Conference Pu"blication 2250

Rotordynamlc
Instability
Problems In
High-Performance

Turbomachinery 1982 -

Proceedings of a workshop
held at Texas A&M University
College Station, Texas
May 10-12, 1982

NI\S/\
 NASA Conference Publication 2250

Rofordynami
Insfa iii
Problems in
High-Performan
Turbomachinery 1 -

Proceedings of a workshop sponsored

by Texas A&M University, College Station,
Texas, and the U.S. Army Research Office,
Durham, North Carolina, and held at
Texas A&M University
College Station, Texas
May 10-12, 1982

National Aeronautics
and Space Administration

Scientific and Technical

Information Branch

1982
 PREFACE

An appreciation of this workshop's proceedings is enhanced by a look

b ackward at the proceedings of the first workshop, which was held at Texas A&M
University on May 12-14, 1980. The initial workshop was organized in response
to severe instability problems that had been encountered in the development of
space shuttle main engine turbopumps and various commercial multistage com
p ressors. The first workshop proceedings emphasized a feeling of uncertainty
in predicting the stability characteristics of turbomachinery based on per
plexing and, in many cases, unexplainable results obtained in field experien
ces with unstable turbomachinery. Further, many proposed analyses and models
for forces, particularly fluid forces, developed in turbomachinery were
clearly speculative and were largely unsupported by test data.

The present workshop proceedings demonstrates that many of the uncertain

ties that were present two years ago have been substantially reduced. To a
considerable extent, papers presented herein report the results of programs
that have been established to systematically resolve the problems of predic
ting stability characteristics of high-performance turbomachinery. The em
phasis on experimental validation of the forces that influence rotordynamics
1S particularly. encouraging.

This second workshop was organized to address the general problem of

rotordynamic instability by gathering those persons with immediate interest,
experience, and knowledge for a discussion and review of both past stability
problems and present research and development efforts. The intent of the
workshop organizers and sponsors is that the workshop and these proceedings
provide a continuing impetus for an understanding and resolution of these
problems.

Chairmen:

Dara W. Childs and

John M. Vance
Texas A&M University

Robert C. Hendricks
NASA Lewis Research Center

111
 DEDICATION

The proceedings of this workshop is

respectfully dedicated to the memory of
Lawrence P. Ludwig, Chief of the
Mechanical Components Branch, NASA Lewis
Research Center. Mr. Ludwig died on
September 11, 1981, after an extended
illness. He is survived by his widow,
two daughters, and a son.

Larry's research on seals over a

period of two decades resulted 1n
advances in sealing technology that
significantly affected the design of a
broad range of machines from aircraft
t urbine engines,
_ rocket engine turbo
pumps, and rdtorcraft transmissions to commercial pumps and com
pressors. He was an internationally recognized expert on sealing
technology and the holder of numerous patents on seals. He was
coinventor of the lift pad or self-acting seal, which has advanced
the speed and pressure ratio capability of face seals in advanced
turbine engines, turbopumps, and compressors, significantly im
proving their efficiency. He is the inventor of two new shaft
riding seals that extend both the speed capability and life of
transmissions. One of these seals, now standard F1uipment on both
a commercial and a military helicopter, has great._ reduced main
tenance costs and downtime.

With the evolution of self-acting seals into the applications

stage he concentrated his efforts on leading research on gas path
sealing for turbine engines. The first of several promising con
cepts was engine tested and should significantly reduce turbine
engine fuel consumption.

Without the multiplicity of improvements in seal technology

for which Larry was responsible, high-speed turbomachines would be
far less advanced than they are today. He was awarded the NASA
Exceptional Scientific Achievement Medal.

v
 CONTENTS

Page

FIELD EXPERIENCES WITH UNSTABLE TURBOMACHINERY

ROTORDYNAMIC INSTABILITY FIELD PROBLEMS

J. C. Wachel . . . 1

FIELD EXPERIENCES WITH RUB INDUCED INSTABILITIES IN TURBOMACHINERY

David G. Goggin. . 20

ANALYSIS OF A TURBINE ROTOR CONTAINING A TRANSVERSE

CRACK AT OAK CREEK UNIT 17
G. W. Rogers, C. A. Rau, Jr., J. J. Kottke, and R. H. Menning. 33

ASPECTS OF COULOMB DAMPING IN ROTORS SUPPORTED

ON HYDRODYNAMIC BEARINGS
P. G. Morton . 45

MECHANICAL INSTABILITY MECHANISMS AND MODELS

INSTABILITIES OF GEARED COUPLINGS - THEORY AND PRACTICE

R. Gordon Kirk, Roy E. Mondy, and Richard C. Murphy 58

SHAFT VIBRATIONS IN TURBOMACHINERY EXCITED BY CRACKS

Burkhard Grabowski 81

DETERMINATION OF BALL BEARING DYNAMIC STIFFNESS

R. F. Beatty and B. F. Rowan . 98

ROTOR VIBRATION CAUSED B Y EXTERNAL EXCITATION AND RUB

Osami Matsushita, Michiyuki Takagi,
Katsuaki Kikuchi, and Makio Kaga . . . . . . . 105

SEAL FORCES IN TURBOMACHINERY

LINEAR FORCE AND MOMENT EQUATIONS FOR AN ANNULAR SMOOTH SHAFT SEAL
PERTURBED BOTH ANGULARLY AND LATERALLY
J. Fenwick, R. DiJulio, M. C. Ek, R. Ehrgott, H. Green,
and S. Shaolian . . . . . . . . . . . . . . . . . . . . . 130

MEASUREMENT OF INTERSTAGE FLUID-ANNULUS DYNAMICAL PROPERTIES

M. L. Adams, E. Makay, and I. A. Diaz-Tous . . . . . . . . . 147

TESTING OF TURBULENT SEALS FOR ROTORDYNAMIC COEFFICIENTS

Dara W. Childs and John B. Dressman . . . . . . . . . 157

vii
 A HIGH-REYNOLDS-NUMBER SEAL TEST FACILITY: FACILITY DESCRIPTION
AND PRELIMINARY TEST DATA
Dara W. Childs, Clayton Nelson, Ted Noyes, and John B. Dressman. 172

VIBRATION AND-DESTABILIZING EFFECTS OF FLOATING

RING SEALS IN COMPRESSORS
Mark F. Emerick. 187

FLOW INDUCED FORCE OF LABYRINTH SEAL

Takuzo Iwatsubo, Naoto Motooka, and Roji Kawai 205

CIRCUMFERENTIAL PRESSURE DISTRIBUTIONS IN A MODEL LABYRINTH SEAL

Y:- M. M. S. Leong and R. D. Brown 223

MEASUREMENT AND EVALUATION OF SWIRL-TYPE FLOW IN LABYRINTH

SEALS OF CONVENTIONAL TURBINE STAGES
L. Hauck . . . . . . . . . . . . . . . . . . . . . . 242

EXPERIMENTAL MEASUREMENT OF ALFORD'S FORCE

IN AXIAL FLOW TURBOMACHINERY
John M. Vance and Frank J. Laudadio . . . . . . . . . . . . . . . . . . . 260

GENERAL THEORY OF STABILITY

HEURISTIC EXPLANATION OF JOURNAL BEARING INSTABILITY

Stephen H. Cranda ll. 274

INFLUENCE OF PARAMETER CHANGES TO INSTABILITY BEHAVIOR OF ROTORS

Claus P. Fritzen and Rainer Nordmann 0- 284

STABILITY EVALUATION OF ROTOR/BEARING SYSTEM BY PERTURBATION TESTS

Donald E. Bently and Agnieszka Muszynska . . . . . . . 307

WORKING FLUID FORCES IN TURBOMACHINERY

AERODYNAMIC STIFFNESS OF AN UNBOUND ECCENTRIC WHIRLING

CENTRIFUGAL IMPELLER WITH AN INFINITE NUMBER OF BLADES
Paul E. Allaire, Lyle A. Branagan, and John A. Kocur . , . . . . . . . 323

UNSTEADY FLOW PHENOMENA IN INDUSTRIAL CENTRIFUGAL COMPRESSOR STAGE

L.IBonciani, L. Terrinoni, and A. Tesei . . . . . . . . . . . 344

COMPARATIVE STUDY OF SUBSYNCHRONOUS ROTATING FLOW PATTERNS

IN CENTRIFUGAL-COMPRESSORS WITH VANELlSS DIFFUSERS
P. Frigne and R. Van Den Braembussche . . . . . . . . . . . 365

EXPERIMENTAL MEASUREMENTS OF HYDRODYNAMIC STIFFNESS

MATRICES FOR A CENTRIFUGAL PUMP IMPELLER
D. S. Chamieh, A. J. Acosta, C. E. Brennen,
T. K. Caughey, and R. Franz. 382

viii
GAS TURBINE ROTOR/CASE STRUCTURAL RESPONSE TO ROTATING STALL:
EXPERIMENTAL DOCUMENTATION AND ANALYTICAL APPROACH
Philip J. Haley . . . . . . . . . . 399

CONTROL OF ROTOR INSTABILITIES

INVESTIGATION OF SQUEEZE-FILM DAMPERS IN FLEXIBLE SUPPORT STRUCTURES

R. Holmes and M. Dogan. 415

VIBRATION LIMITING OF ROTORS BY FEEDBACK CONTROL

David W. Lewis, JamesW. Moore,
Philip L. Bradle,
y and Paul E. Allaire 434

EVALUATION OF SHEAR MOUNTED ELASTOMERIC DAMPER

E. ZorziandJ. Walton. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 440

l.X
 ROIDRDYNAMIC INSTABILITI FIELD PROBLBVIS

J. C. Wachel
Southwest Research Institute
San Antonio, Texas 78284

SUMMARY

Vibration data obtained during several rotor instability investigations is

presented to illustrate the effect of changes in system parameters on overall rotor
stability. The data includes the effects of bearing and seal changes as well as
those due to variations in speed and pressure ratio. Field problems indicate that
the stability of rotors is often highly sensitive to fairly minor variations in
bearing and seal parameters. Measured field data is valuable in normalizing analyt
ical computer models so that effective solutions can be obtained.

INTRODUCTION

Rotor instability vibrations in compressors and turbines have occurred more

frequently in recent years and have caused severe failures and costly downtime for
several large projects. Rotor instabilities can occur in flexible shaft units which
operate above their first critical speed. The whirling instability frequency is
usually near one of the shaft critical speeds and can be caused by many factors,
including hydrodynamic cross coupling of bearings and seals, internal friction, aero
dynamic cross coupling, and torsional coupling. The whirling motion can be subsyn
chronous or supersynchronous, and may be forward or backward precession; however, in
general most problems are subsynchronous and have forward whirl.

During the past few years, vibration data has been collected on several compres
sors that have experienced severe shaft instabilities. These compressors differed
in manufacture, shaft diameter, weight, bearing span, critical speeds, and running
speed. The spectral characteristics of shaft vibrations were observed as the
compressors/turbines approached the onst of instability; i.e., before the machine
experienced the high level vibrations normally associated with unbounded instabil
ities. On most units that have instability problems, a trace of vibration at some
instability frequency normally exists at all times; however, it is not possible to
verify the severity of the instability from vibration measurements at one operating
condition. The threshold of instability can be fully defined only from testing over
the full performance range of the machine, and even this approach is not always com
pletely adequate. Some units have run satisfactorily for several years before
serious instability trip-outs occurred. After one year of satisfactory operation,
one compressor failed eight times in the next three years from instabilities.
Because the stability margin on some units is so delicately balanced, its character
istics can be drastically changed whenever small changes are made in factors such as
pressure ratio, flow, bearing clearance, oil temperature, unbalance, alignment, etc.,
or upsets in the process such as liquid slugs, surge transients, or electrical
trip-outs.

It follows, therefore, that the threshold of stability can likewise be improved

by small changes in these same parameters, but the exact improvement required to make
an unstable system stable is sometimes difficult to predict.
 The most sensitive elements which influence rotor stability include the fol
lowing: (1) hydrodynamic cross coupling in fluid film bearings, seals, and laby
rinths, (2) aerodynamic cross coupling forces, (3) hysteretic or internal friction
damping, (4) pulsations, (5) pulsating torque and axial loads, (6) asymmetric
shafting, (7) fluid trapped in rotor, (8) stick-slip rubs and chatter, (9) dry fric
tion whip.

To properly calculate the stability margin of a rotor, the mathematical model

must be able to simulate all possible destabilizing components. The logarithmic dec
rement evaluation of rotor system damping is useful for predicting rotor stability.
Field experience shows that while this technique provides proper direction in
designing for stability, uncertainty still exists in quantitatively predicting the
onset of instability and defining the contribution of individual influencing
parameters.

When instability vibrations occur in installed machinery, better estimates of

t he possible effects of system changes can be made if measured field data is avail
able for normalization of the mathematical model. The normalization procedures com
pensate for unknown dimensional variations which affect bearing and seal properties
and adjust for actual aerodynamic loading. This paper will present measured field
data gathered over the past nine years on several machines which exhibited instabil
ities (ref. 1). The data analysis techniques presented were used to define rotor
stability thresholds and the effects of modifications to seals, bearings, shafts, and
process variations.

CASE 1. INSTABILITY OF A STEAM TURBINE

On startup of a steam turbine after a complete turn-around in which new pressure

pad bearings were installed, the measured shaft vibrations indicated the turbine
first critical speed was at 1800 rpm as shown in figure 1 which gives the amplitude
and phase angle response versus speed. When the unit speed approached 4800 rpm, a
s ubsynchronous instability at 1800 cpm suddenly appeared (fig. 2).

A stability analysis revealed that the calculated logarithmic decrement for this
rotor with tight bearing clearances was only 0.04. Investigations were made into
possible field modifications to improve the stability that could be implemented in a
short time. Calculations showed that if the pressure pad bearing clearance was
increased and the bearing length reduced, then the logarithmic decrement increased to
0.2 with the critical frequency remaining near 1800 cpm.

The bearings were then modified and installed. Vibration data with the new
bearings is shown in figure 3. The turbine speed could be increased to 5100 rpm
without any instability occurring and the unit has continued to operate without
instabilities. This case illustrates that some instability problems can be solved by
fairly simple modifications.

CASE 2. STEAM TURBINE INSTABILITY IN A METHANOL PLANT (ref. 2)

This case deals with a 13,000 hp, 10,600 rpm, three stage steam turbine. The
rotor had pressure pad bearings at a bearing span of 236 cm (60 inches). The bear
ings were later changed to 5 shoe tilted pad bearings in an attempt to eliminate the
half speed problems which occurred at maximum speed. Data taken during the t urbine
startup with the new bearings (fig. 4) revealed a vibration component at one-half
speed when the speed reached 7200 rpm, thus showing that the change to tilted pad
bearings was not sufficient to eliminate the half speed vibrations. During subse-

2
quent runs, two subharmonic criticals at 4500 and 7000 cpm were excited as well as
the half speed component (fig. 5).

During another run a subsynchronous frequency occurred at approximately 0.3

times the running speed. Figure 6 shows that a large amplitude component occurred at
2500 cpm when the running speed was 9000 rpm. Both oil pumps were running during
this test and one tripped out for a few seconds. The turbine running speed suddenly
dropped 200 rpm. The instantaneous frequency analysis shown in the upper trace of
figure 6 shows that this upset moved the subs ynchronous component from 2500 to 4400
cpm. Frequency analyses made later showed that the subsynchronous vibration compo
nents were also different on the horizontal and vertical probes of the turbine
(fig. 7). These two instability frequencies were near calculated damped instability
frequencies.

Several modifications were implemented that reduced the magnitude of the insta
bility. These included changing the seal design, increasing the bearing clearances,
and strengthening the bearing housing. This case illustrates that changing to tilted
pad bearings may improve the stability characteristics of a rotor but does not neces
sarily eliminate instabilities. With the tilted pad bearings for this rotor, how
ever, only slight changes were required to control the instabilities.

CASE "3. INSTABILITY OF GAS REINJECTION COMPRESSOR (ref. 3)

This case deals with a much discussed reinjection compressor which experienced
excessive nonsynchronous vibrations on startup. Field vibration data will be pre
sented which shows the influence of oil ring seals, aerodynamic cross coupling, and
speed on instability frequencies and amplitudes. These areas are of major concern to
rotordynamists; however, very little experimental data is available in the open tech
nical literature. The 22,000 horsepower, eight stage compressor with back-to-back
impellers (fig. 8) was rated at 8500 rpm, had a design suction pressure of 24.1 MPa
(3500 psi), and discharge pressure of 63.4 MPa (9200 psi). The calculated first
critical speed of the rotor was 3800 cpm for a bearing span of 206 cm (81 inches).
Floating oil seals were located a few inches inboard of the bearings. The compressor
originally could not be brought to design speed and pressure without tripping out on
high vibrations (fig. 9). The units were monitored by shaft vibration probes which
automatically shut down the unit whenever the vibrations exceeded 64 m (2.5 mils);
however, due to the monitor's finite response time and suddenness of the instability
trip-outs, vibration amplitudes equaling total bearing clearance were experienced.

The frequency of the non synchronous instability was 4400 cpm which was higher
than the calculated rigid bearing critical speed of 4200 cpm. This can occur if the
floating oil seals lock up and carry some load, thereby effectively reducing the
bearing span. In the computer simulation of this shaft, an effective oil seal stiff
ness of 286,000 N /cm (500,000 lb/in) was required to calculate an instability
frequency of 4400 cpm. Using this stiffness for the oil seals the calculated log
decrement reduced to 0.08 compared to 0.3 calculated for the original rotor,
neglecting the effect of the seals. Therefore, the calculations indicated that the
seals significantly reduced the stability of the unit.

To improve the rotor instability, two circumferential grooves w ere cut into the
sealing surface of the seals, the pressure balance of the rings was improved, and the
coefficient of friction of the sliding surfaces was reduced. The compressor was
still unstable, as can be seen in figure 10. A nonsynchronous instability occurred
at 4700 cpm; however, instabilities above running speed at 9500 and 10,500 cpm were
also excited. As the unit speed reduced, the instability component at 10,000 cpm

3
remained. The rotor was found to be sensitive to the rate of acceleration; there
fore, by slowing down the startup procedure, it was possible to operate in the normal
speed range.

To more fully define the stability limits, data was obtained throughout the
entire performance map. For a constant speed of 7600 rpm, figure 11 shows how the
aerodynamic loading affects the amplitude of the instability component at 5160 cpm (a
forward precessional mode). As the suction pressure increased, the amplitude of the
instability increased but remained within bounds until a limiting pressure was
reached. The frequency of the instability component moved from 4400 to 5160 cpm as
pressure was increased. To show the effect of speed on the instability the suction
pressure was held constant at 10.3 MPa (1500 psi) and the speed increased. The
instability amplitude increased almost linearly with speed (fig. 12).

After these tests were made, several seal designs were tested (fig. 13); how
ever, there was little improvement in the overall rotor stability. The type of seal
design greatly affected the frequency of the nonsynchronous instability and the
threshold speed. One seal design studied had large radial clearances and only one
land, less than 5 mm (0.2 inch) long. The test was primarily to study the effect on
the instability frequency since the seal oil leakage was excessive. Shortening the
seal length should reduce its load-carrying effect, thus reducing the instability
frequency. However. test results showed that nonsynchronous instabilities occurred
at frequencies above running speed (10,000 and 11,000 cpm) similar to the data pre
sented in figure 10.

Another test using a different seal design also showed instabilities above run
ning speed. These instability frequencies appeared to be a function of suction pres
sure as shown in figure 14. During this test the compressor speed was 7523 rpm and
suction pressure 8.28 MPa (1200 psi). Figure 14 gives the frequency analysis showing
13 m (0.5 mil) at running speed and 25 m (1.0 mil) at 10,500 cpm and a trace at
4500 cpm. As the suction pressure was reduced, the higher frequency component low
ered to 8900 cpm and then separated into two components, 8900 and 9300 cpm.

Some specifications require that the first critical speed be greater than 0.6
times the running speed to help prevent instabilities. Figure 15 shows that an
instability component at 0.8 times running speed occurred when the running speed was
4000 rpm or slightly above the first critical speed. In this data, the instability
occurred when the ratio of running speed to first critical was 1.25, showing that a
ratio of running speed to first critical of less than 2:1 does not necessarily ensure
that a rotor will be stable. The majority of the instability trip-outs were at
s peeds where the ratio of running speed to first critical speed was less than 2:1.

The data presented shows that the stability frequency characteristics were dram
atically changed by changing only the oil seals; however, no change in seal design
made this system stable. This indicates that the seals were not the predominant
destabilizing factor.

Major efforts were then expended to reduce other destabilizing factors. These
changes included aerodynamic changes to the impellers and diffusers, seal modifica
tions, shortening the bearing span, and changing the bearings to 5 shoe, nonpreloaded
tilted pad bearings. Log decrement calculations indicated that these changes repre
sented a significant improvement in the rotor stability.

Even after these modifications were installed, instabilities still occurred. On

startup, an instability component (5200 cpm) was present with a fluctuating ampli-

4
tude. The instability frequency then shifted up to5800 cpm, and the amplitude
jumped to greater than 152 (6 mils) in approximately one second. The instability
component was particularly sensitive to pressure ratio across the machine, which con
firmed that the aerodynamic destabilizing effects were of major importance and over
s hadowed other improvements that were made.

The stability of the unit was markedly improved when a damper bearing was
installed in series with the inboard bearing. Squeeze film damper bearings employ an
oil film in the space between the outside of the bearing and the case, to which oil
is continuously supplied. Stiffness of the damper bearing is usually supplied by a
mechanical support such as a s quirrel cage cylinder with ribs, welded rod support,
corrugated metal ring, or o-rings. 16 gives the frequency analysis of the
Figure
s haft vibrations and two probes monitoring the damper bearing for 8450 rpm with a
suction pressure of 23.1 MPa (3350 psi) and a discharge pressure of 56.9 MPa (8250
psi). Instability frequencies were still present, but the added damping from the
damper bearing prevented the amplitudes from becoming unbounded. The improvement in
the stability characteristics of this rotor illustrates the potential of damper bear
ings in high pressure applications.

At the time of this study , the mathematical techniques for predicting instabili
ties were not as developed as today's procedures and the application of a damper to
an industrial compressor involved some tuning to obtain the optimum stiffness and
damping for an individual rotor. This can be amply illustrated by the fact that a
damper bearing installed in a second identical unit was not successful in eliminating
instability trip-outs (fig. 17). After some tuning of the damper bearing, the sta
bility was significantly improved.

In the process of evaluating the performance of the damper bearing, the oil tem
perature was varied to determine if it had a significant effect. At an oil tempera
ture of 510 C (1240 F), the frequency analysis of shaft vibrations showed instability
components at 2800 and 4800 cpm along with the running speed component. Some sur
prising results were noted as the temperature was lowered. At a temperature of
approximately 48.90 C (1200 F), the 6 m (0.25 mil) component at 2800 cpm disappeared
(fig. 18). Again this points out that very small changes can be significant to the
stability of a rotor.

This case history illustrates that many factors influence the onset, frequency
and amplitude of instability vibrations, and that sophisticated mathematical models
are required to simulate the instability phenomena measured in these machines.

This instability problem was controlled primarily by increasing the shaft diam
eter to raise the first critical speed, thus significantly increasing the ability of
the shaft to withstand the large aerodynamic loading effects.

Aerodynamic loading effects are the most predominant destabilizing components in

many high pressure systems. In the design stage the designer needs to estimate the
level of equivalent aerodynamic loading so that the rotor will have an adequate sta
bility margin. The author has consulted on several instability problems and has
developed an empirical formula for estimating the level of aerodynamic loading
(ref. 4).

B(hp)(Mol Wt) PD
K
xy Dhf
Ps

5
 aerodynamic loading, N/m (lb/in)
xy
K

B cross coupling constant, 16 (105)

hp power, kW (hp)

Mol Wt molecular weight

D impeller diameter, m (in)

h restrictive dimension in flow path, m (in)

f speed, Hz
3
density of fluid at discharge conditions, kg/m (lb/cu ft)
PD
3
density of fluid at suction conditions, kg/m (lb/cu ft)
Ps
=

When this formula was applied to several rotors that had instabilities, it
appeared to give overall levels of aerodynamic loading near that required to cause
t he logarithmic decrement to be negative. This equation is presented so that it may
be further evaluated.

CASE 4. INSTABILITY OF ATOMIZER SHAFT

An atomizer was driven by a 250 hp, 3600 rpm, electric motor through a variable
s peed transmission over a speed range between 10,000 and 13,500 rpm. A slurry was
p umped by positive displacement pumps to the atomizer which wa s mounted on top of a
large cone-shaped tank. The slurry was then sprayed into the tank and was instantly
dried by hot air which was blown directly at the atomizer wheel. The heated air was
forced through the heater and into the tank by a forced draft fan. The dried powder
was then drawn out of the tank by an induced draft fan into a bag house where the
pow der was collected. Upon startup, high vibrations were experienced on the atomizer
s haft, resulting in several shaft failures near a carbon bushing. A field investiga
t ion was made to determine the causes of the high vibration and failures and to
develop a solution.

During the tests the atomizer shaft vibrations were measured with a proximity
probe mounted inside the atomizer housing approximately 15.2 cm (6 inches) above the
carbon bushing (fig. 19). Torsional vibrations were measured with a CEC torsiograph
mounted on a special stub shaft attached to the motor shaft. Pulsations in the
liquid feed line and the tank near the atomizer wheel were measured with pressure
transducers. Vibrations on the atomizer, gearbox, and motor were measured with
accelerometers.

Several tests were made, including running the unit without condensate or
slurry, r unning with water alone, running with fans on but atomizer off, running with
heater only at a low feed rate, and shock excitation runs to determine natural
frequencies.

When the shaft was run dry, two instability frequencies at 12 and 26 Hz were
indicated in the real time spectral analysis of the proximity probe measurements of
the carbon. When the unit was run with condensate water, the amplitudes at the
instabilities increased (fig. 20). The vibrations at the forward whirl mode occurred
first and built up until the shaft touched the carbon bushing which caused a backward
whirl mode to be excited at 12 Hz. Measured pulsations at the slurry input showed
pulsations at 26 Hz which could excite the forward mode at 26 Hz. The sensitivity of

6
the instability modes can be seen in figure 21. When the vibrations were measured
over the range of flow rates, the vibrations became severe as shown in figure 22.

Based upon the analysis of the data, it was determined that the excitation for
the instability was the pulsative fluid flow. The atomizer was designed for a feed
r ate of 50 cubic meters per hour; however, the maximum flow rate was only 19 cubic
meters per hour. The atomizer was creating a vacuum in the feed header line even at
f ull flow rates. This caused the liquid to enter the atomizer in slugs rather than a
s mooth flow. The liquid slugs entering the atomizer wheel acted as destabilizing
forces which caused the forward whirl.

Several modifications were implemented to improve the inlet flow characteris

tics. The distributor flow area was reduced approximately 70% by an orifice
r estrictor and the inlet feed tube diameters were reduced and connected together
through a " Y " connection.

These modifications on the inlet flow lines and the new distributor improved the
inlet flow characteristics and reduced the instability vibration amplitudes; however,
t he shaft continued to rub against the carbon bushing under all flow conditions.
P lu gs were then installed in 12 of the 24 discharge holes in the atomizer wheel to
f urther restrict the flow rate.

After these modifications were installed, the vibrations were low ( fig. 23) and
t he shaft no longer rubbed against the carbon bushing. The atomizers have run suc
c essfully without failures since 1977.

This case history illustrates that if the excitation source for an instability
can be identified and reduced, the instability can be controlled without shaft
modifications.

REFERENCES

1. Wachel, J. C., Case Histories of Rotordynamic Installations in the Field, Short

Course on Rotordynamics of Turbomachinery , Texas A&M University, College
Station, Texas, May 18-20, 1981.

2. Wachel, J. C., Turbine and Compressor Vibration, Ammonia Plant Safety, vol. 15,
1973, American Institute of Mechanical Engineers, pp. 69-76.

3. Wachel, J. C., Nonsynchronous Instability of Centrifugal Compressors, ASME Paper

7 5-Pet-22.

4. Wachel, J. C. and W. W. von Nimitz, Ensuring the Reliability of Offshore Gas

Compressor Systems, Society of Petroleum Engineers of AIME, Journal of
Petroleum Technology, Nov. 1981, pp. 2252-2260.

7
 '"

Figure 1. - Vibration and phase angle measured during shutdown.

.! dP ,
. . I, II" I 11-1 I "H I'
i .. , ;'
, , , ,
, ' , " , I

"
" , I"
11 II I
" "

, , '-++

I:R: ..
', ' 1 " 1 .
" , " " ,
I I I , I ' . ' "

Figure 2. - Subsynchronous instability excited on turbine.

8
 Figure 3. - Elimination of instability by bearing modification.

Q)

E
b
.....

o 3 6 9 12 15

CPM x 10-3

F igure 4. - Spectral time history of half speed vibrations on

turbine with tilted pad bearings.

9
 1
o 3 . 6 9 12. 15

CPM x 10-3

Figure 5. - Spectral time history of turbine showing

subharmonic vibrations in normal speed range

VIBRATION DATA
D PULSATION DATA
mi'''di,
p"fd..
T D STRAIN DATA
D NOISE OATA
pin f,n.ldi,
dBfdi,

Test Poml Tl- 'i[fL. 04 62.

Unit Speed -87f!JO 'pm l)HJ:A;N.
a.:i ""'t.IP;"
Unit
t:iJoAM.
! Ve,' Scale
LI/Zz.
Time

I
i HortZ. Scale 0- 2()OtJO UIII Date .

----_. __ . . . -._----- ... . -. _-

!
.

i
A - . - - . . . _ . " -

K \
4:1. A. 11.,. ! . .

o
1\ -

-Runninl --

Speed

, Note Change in Sub -:

1'1
,
. . . . . . . . . . - - -----.---

i
!
-harmonic FrequencY,

! \
;
,

. . . --

,
I 1 mil -
25.4 11m
E i
'" !
!
--
,

I -\
I

,
i I I

j .1
-1 :.3.M!. t--
r
o zooo 4000 6000 8000 10000 IZOOO 14000 16000 18000 Z1IOOO
RPM

Figure 6. - Subharmonic vibrations of turbine.

10
 \ I I
o
VIBRATION DATA
PULSATION DATA
-----"-_.- - -

_7JLgA/1?-
ma,/di.
",ildi.
I

Test COl.lg
04- Z9'"
Cod.

- -I-
Test Poml

+-r
____

Unit Speed _-----./t>....!l_. rpm Unit 5Y;V6.S

VerI. Scale 0.5 _L_ Time 12 -40
,P
-<

__

. I HOfll. SciJle 1l:......- 'pm Date 2.. -14- Zle.

t
.r
1 mil = 25.4 lJm

V
- J\
lPI'>-H..
---r--- )\
-h .- ZZL
r--
..,-/ . . -- .--.

- -
\.
- Note Difference in Frequency

-- . --.
on Vertical and Horizontal

-I--- 77Ll-H
Probe.

IIIIIr!!:':
1 I I
'I
....
) l.
o zooo 4000 6000 8000 10000 IZOOO
RPM

Figure 7. - Subharmonic turbine vibrations measured at 10400 RPM.

Vibration Proximity Probes

(Horizontal, Vertical, Axiall

Coupling

t '- ter
Labyrinth

Gas Labyrinth Impeller

Figure 8. - Typical high pressure compressor design.

11
 o 50 10000
eque cprn

Fi gure 9. - Tripout of compressor as speed is increased.

1(j)
S
......

E-!

o 1 00 20000
cpm

F igure 10. - Compressor instability with improved seal design.

12
 0.00

0.70

0.60

0.50

0.20

0.1l

I pd .. 6.895 kPa

suction Pressure, pslg

Figure 11. - Instability amplitude versus suction pressure.

1.00

0.80

0.60

0::
I

tJ)
..J 0.40
if
W-
0
'"
f-
::;
Q. 0.30

<t

0.20 1 mil .. 25.4 lJlll

0.0 '------'---'---'
o 8000 8100 8200 8300 8400

COMPRESSOR SPEED, RPM

F igure 12. - Instability amplitude versus compressor speed.

13
Mod. 1 (1) Teflon O-rings
(2) Teflon eoating on sealing surfaces
(3) Two circumferential grooves
(4) Machine shoulder to balance pressure

D
Mod. 2 (1) Original seals
(2) No O-rings
(3) Teflon coating on sealing surfaces
(4) Three 011 dams at 1200 intervals

Mod. 3 Basic Mod. 1 seal with two downstream lands removed

Mod. 4 (1) Original seals

(2) Three per Bea1 assembly
(3) Upstream edge beveled
(4) Teflon O-rings

Mod. 5 (1) No O-rings

(2) Circumferential groove
(3) Reduced unbalanced pressures using peripheral lugs
(4) Teflon coating of sealing surfaces

l
Mod. 6 Identical to Mod. 5 except that only two seals were
used per end

t:=:::
fI:J
a.>
Mod. 7 Basic Mod. 1 seal with axial grooves replacing circum-
oronti.1 groove. S
. ...

E-<

o 6000 12000
Frequency, cpm

Figure 13. Oil seal designs tested. Fig ure 14. Variation of instability
frequency with suction pressure.

14
 o 10000 20000
Frequency, cprn

Figure 15. - Instability excited as compressor goes through first critical.

.t
Damper Bearing Horizontal

o..J
c
:::>
!::
-'
0-
:E
j Outboard Vertical

----
Outboard Horizonta I
co::

I
co::
IX
<0

J "-
Inboard Vertical
>

O
,
=!
.;!:: _
Inboard Horizontal

i
O C=== )\j

o 2000 400G 6000 8000 10000

Frequency, cpm

Figure 16. - Frequency analysis of shaft and damper bearing vibration

with diSCharge pressure of 8250 PSI (56.9 MPa).

15
 1

o 5000 10000
Frequency, cpm

Figure 17. - Instability tripout of rotor with damper bearing installed.

tC1)
S
.

E-i

a 5000 10000
Frequency, cpm

Figure 18. - Elimination of instability component as oil temperature reduced.

16
 Figure 19. - Cutaway view of atomizer.

Backward
Whirl

'" - ._--_._.- .. .. ---- - ----- ----------

I I
-

3
------ ______ 0-- ------------_.-

'"
Forward
"
.... Whirl Atomizer
....
13 Sub synchronous Vibrations Speed
Increased when Condensate
,j
Added.
]
..... 2
,..;
'"

"
0 Motor
.....
'"' Speed
'"
.. 1---- .- -- -..-
.0 1
.....

V\. \.
:>

\ -
--
a
a 50 100 150 200 250 300

Frequency, Hz

Figure 20. Instabilities excited on atomizer.

17
 " I
-
Cooling

1.0 AI ssv Immediately Decreased

H"" ..... '" .., ... After Water Off

100- J ) \

;;:
Backward Whirl I I
--j--- -
When Cooling Water Entered Wheel.
6 2.0

...>! SSV at 26 and 12 Hz Immediately

'"
Forward Increased
Whirl

i'l
... 1.0 t-
A With Cooling
Water Flow

'"

II
\. ' l J\
:-
0

1.0

50 100 150 200 250 300

Frequency. Hz

Fi gure 21. - Effect of fluid flow on instabilities.

I I I
. 12 ....laebard Vhirl
7.0 I- A 26
EI
'or-rll Whirl
..
60 Jb MDEor '-peed
(;) 210 liz am r .....

6.0
r
i\ A II
V \
s. 0

,. 0 I \/
IJ I \J
V

3. 0

V
};{

2.0
/)., /-

1.0

Figure 22. - Plot of atomizer shaft vibrations versus feed rate.

18

j j I
12 Hz Backward Whirl
A 26 Hz Forward Whirl
'" 3.0 I--
I S 60 Hz Motor Speed
Q.
.. 0 230 Hz Atomizer Speed
....
....
IS

.;
'g 2.0
....
....
....
Q.

c:
0 ......,
....
.... .,..
'" --
.. 1.0
,D
....
>

0
?
1 2 3 4 5 6 7 8

Feed Rate

Figure 23. - Atomizer shaft vibrations after modifications.

19
 FIElD EXPERIEl\CES WITH RUB INIJl.CED INSTABILITIES IN TURIDV1AGUNERY*

David G. Goggin
Radian COrporation
P.O. Box 9948
Austin, Texas 78766

SmIMARY

Rotord y nam ic instability problems are not uncommon in high speed industrial
turb omac hinery. One type of the many destabilizing forces that can occur is caused
by a rub between the stationary and rotating parts.

Descriptions are given of several cases of rub induced instab ilities. Included
in the descriptions are the conditions at onset. the w hirl frequency and direction.
and the steps taken to eliminate the problem.

INTRODUCTION

Rot ordy namic instab ility problems are not uncommon in high speed industrial
turbom achinery. Actually, a large percentage of industrial machines operating at
hig h s peed will sustain some low level subharmonic vibrations even during normal
operation. Some of these units have been observed to develop instability problems.
One mechanism that has been respon sible for some of these instab ilities is a rub
between the rotating and stationary parts.

The mechanics invol ved during a rub have been described by several authors.
Den Hartog ( ref. 1) describes rubs from the standpoint of d ry friction w hip wit h
Coulomb friction between the rotating and stationary parts prov iding t h e destabi
lizing force. Ehric h ( ref. 2) added stat or flexib ility to this m odel in an attempt
to def ine the conditions necessary for a rub to be unstable since not all rubs pro
duce instab ilities. Both analyses conclude that the instab ility should produce a
backward whirl at the rotor natural frequenc y .

Bentley ( ref. 3) proposed a n d experimentally demonstrated that several mecha

nisms, including a partial rotor-stator rub, could produce a subsy nchronous
vibration due to the periodic variation it causes in the rot or's support stiffness.
The s ub s y nchronous v ibration was stated to occur most often at exactly 1/2 the run
ning speed when operating at or s light ly above twice the first c ritical speed. The
whirl direction could be forward, backward, or even in a single plane. Although
less common, the whirl frequency was noted to also have occurred at 1/3 and 1/4
running speed. It was also stated that it was possible, but not likely , for the
s ubharmonic to occur at 1/ 5. 1/6 and 2/3, 3/4. 2/ 5, 3/5 of running
speed. This parametric exc itation mechanism was further ana ly zed by Childs ( ref. 4
and 5). The rub model used included the effects of both Coulomb friction and the

* This work was funded by The Hartford Steam Boiler Inspection and Insurance
Company.

20
periodic variation in support sti ffness. The analysis indica ted that zones of
instability exist around i nteger multiples of the first critical speed within which
a rub could cause an instability to occur. The frequenc y of the whirl would occur
at slightly more than the first critical speed causing a vibration at exactly 1/2,
1/3, of running speed. The possibility that the rub would be unstable was found
to be primarily dependent on the amount of damping of the first mode, the magnitude
of the Coulomb friction factor, and a parameter q which represents several param
e ters that essentially indicate the severi ty of the rub. An un stable rub was found
to be promoted by low damping, a high friction coefficient, or a high q factor,
i.e a severe rub.

In this paper some f ield experiences with rub induced in stabilities are re
viewed. As with most industri al problems, the pressure to resume production often
prevented a more detailed investigation. But most of the essential characteristics
involved are presented including the condition s at onset, the whirl frequenc y, the
di rection of whi rl, and the steps taken to correct the problem. Some rubs, of
course, have occurred that were not unstable, such as the one shown in figure 1 from
a 5000kw turbine generator. These are usually characterized by increases in the
higher order harmonics as well as a general increase in the low level broad band
vibration. While these types of rubs can be serious, it is us ually the unstable
rubs that have the most potential for damage and so are of deep concern t o the in
dustry.

BACK PRESSURE lURBINE ON SYNTIIESIS GAS COMPRESSORS

Figure 2 shows a schematic of a 20,000 H.P., 11,000 rpm Synthesis Gas Com
p ressor train compo sed of a condensing turbine, a bac k pressure turbine, a low pr es
s ure and a high pressure compressor. The bac k pressure turbine, or topping turbine
as it is referred to, is a two stage 12,000 H.P. drive through turbine using 10.34
MPa (1500 p s i ) inlet and 3.79 MPa (550 psi ) exhaust steam. The first and second
critical speeds are 6750 rp m and 13,000 rpm respect ively as determined f rom proxi
mity probe data during start-ups and overspeed trip tests. The 5800 kg (913 lb )
rotor is supported in a 12.7 em (5 in ) diameter bea ring at the inlet end and a 7.6
cm (3 in ) diameter bearing at the discharge end. Both are 5 s hoe tilting pad bear
ings with a load on pad ori entation. This m achine, like many others, tends to show
some low level (2.54 or 0.1 m i l ) subharmonic vibration over a broad band of
about 30-150 Hz during normal operation.

During the start-up of one of the many units like this, a vibration instability
problem developed at the discharge end of the topping turbine. The vibration
levels, as indicated by proximity probes, incr eased from 25 - 3 8 m (1.0-1.5 mils )
to 76 - 89 m (3.0-3.5 mils ) in a few seconds. Since a real time analyzer was pre
sent to monitor the start-up, it was observed that the vibration was dominated by a
large fluctuating component at 5150 rpm (85.8 Hz ) or exactly 1/2 running speed. The
suddenness with which the instability developed, the rapid fluctuations in the sub
harmonic component, and the fact that the subharmo nic was at exactly 50% of running
speed, led to the suspicion that a rub was involved. No explanation could be made
for the fact that the whi rl frequency was 5150 rpm (85 .8 H z) when the first critic al
was 6750 rpm (112.5 Hz ) . The problem was almost completely isolated to the topping
turbine discharge end bearing. Some 1/2 running s peed vibration was observed at the
inlet end bearing, but only low levels 5 m or 0.2 mil ) of this component were ob
served on other rotors in the train.

21
 A shutdown and inspection of the bearing revealed that everything was normal
including proper orientation. clearance. crush fit. alignment. etc. The bearing's
inboard oi l guard. however. was found to be heavily rubbed and was replaced with one
that was checked to insure it had the proper clearances. This was the only change
made befor e the machine was restarted. Since no subharmonic vibrations occurred
during start-up and subsequent operation. it was concluded that the rubbing oil
guard had been the source of the instability.

Another identical unit exhibited the same instability characteristics. This

unit had developed a history of vibration problems during its eleven years of opera
tion. One problem was occasional bouts of fluctuating hig h vibration at the
topping turbine discharge end bearing. It had been found that using water on the
bearing pedestals ,to change the alignment would sometimes reduce or eliminate the
vibrations. An analysis was made during one vibration excursion to determine w hat
effect the water was having. Figure 3 shows a spectra taken at 3:29 p.m. with water
being used on one pedestal. There was a running speed vibr ation of 64 m (2.5 mils )
at 10,200 rpm (170 Hz ) and a 1/2 running speed peak of 38 m (1.5 mils ) . The 1/2
running speed peak was superimposed over the normal band of low level subharmonic
noise and ther e was some electrical interference at 60 Hz. 120 Hz, 180 Hz . etc. The
water was taken off the pedestal and figure 4. taken at 3 :35 p.m shows the
resulting increase in the subharmonic. The water was then p ut on the other pedestal
and the results are shown in figure 5, taken at 3:47 p .m. The instability had
ceased suddenly and the running s peed vibration slowly decreased from 64 m (2.5
mils ) to 38 m (1.5 mils ) . During a subsequent outage. the bearings and seals at
this point were inspected and found to be severely rubbed over a bottom quarter of
thei r d iameter. It was concluded that this was another rub which had produced both
a large subsynchronous component as well as an increase in the unbalance from non
uni form heat input to the shaft. Once the rub was removed by ch anging the align
ment. the subharmonic component ceased suddenly and the running speed vibration
returned to normal slowly.

On a later occasion. this same turbine experienced high vibr ations during a
start-up. The machine was warmed up and brought up to minimum governor speed (9000
rpm ) one evening and the load was to be progressively increased o ver the next 24
hours. However. at 11:00 the next morning while operating at 10,5 83 rpm, the vibra
tion levels at the topping discharge end s uddenly increased from about 50 m
(2 mils ) to in excess of 1 2 7 m (5 mils > , the limit of the vibration monitors. By
12:00, an analysis had revealed ( fig. 6) vibration levels in excess of 152 m
(6 m ils ) with a running speed component of approximately 91 m (3.6 mils ) and a
fluctuating 1/2 running speed component of 114 m (4.3 mils ) . Table 1 details the
vibration l evels during the excursion. A check of the gap voltages showed an
increase of 0.9-1.1 volts had taken place. While some of this may have been due to
the increase in the rotor orbit shifting the rotor's mean position, it still indi
cated that the rotor was located significantly lower in the bearing than normal. A
slight increase in speed to 10.672 rpm failed to reduce the vibration level as it
had with other rub induced instabil ities. During the increase. the subharmonic
tracked the increase in running s peed so as to remain at exactl y 1/2 running speed.
The di rection of whirl during the enti re episode was forward. Water. was used on the
turbine casing supports in an attempt to lower the bear.ing r elative to the shaft and
hopefully clear the rub. When water was placed on one support pedestal, the vibra
tions increased to about 229 m (9 mils ) . But when water was quickly placed on the
other pedestal as well, the vibrations began to decrease. Within an hour. the gap
voltages had returned to normal and, as shown in figure 7. the vibration levels were
al most normal. During a subsequent inspection of the turbine. the bottom part of

22
the bearings and seals at the discharge end showed severe rub damage and some break
age. Again. the instab ility was thought to be caused by a rub brought about by a
severe misal ignment during a thermally transient condition. There was still no
explanation for the w hi rl frequency to be at 5250 rp m (87.5 Hz) when the first
critical speed was 6750 rpm (112.5 Hz).

TURBINE ON COz COMPRESSOR

Figure 8 shows a schematic of a CO compressor train driven by a 8200 H.P.,

z
9300 rpm, single extraction, condensing turbine. The 19,240 kg (3030 lb) rotor was
supported on 15.2 cm (6 in) journals with sleeve type bearings. The f irst and
second critical speed were approx imately 4900 rpm and 10.150 rpm res pectively as
determ ined from proximity probe data during start-ups and overspeed trip tests.
This unit had been in operation over nine years with no reported vibration problems
except for occasional temporary bows during start-ups. The machine's susceptibility
to thermal bows was due to short warm-up and start-up periods that were necessitated
by plant design. Due to the trouble free operation of the unit. it had only been
overhauled once five years after installation. The train was located out oors with
a protective roof over it.

A problem was first noted one February when the turbine thrust position monitor
showed a temporary alert condition after a sudden change in wind direction. Future
changes in wind direction caused the problem to recur with enough regularity that a
portable blower was used to keep a con stant ai r flow directed at the thrust bearing
housing. This worked until March. when a rainstorm occurred which resulted in al l
the turbine vibration alarms going off as well as a thrust alert condition. The
condition lasted for about one minute and subsided as quickly as it started.

During a brief outage in the summer, the bearings were inspected and found to
be acceptable. But the bearing's oil seals were replaced a t the governor end due to
excess ive leakage. These seals were noted to be tighter than normal with 50-75
(2-3 mils) d iametral clearance. During the start-up, another vibration excursion
occurred. The turbine had reached minimum governor s peed and the s peed was being
increased slow ly as needed. At 9 250 rpm, the vibration levels suddenly increased
from about 25-38 m (1.0-1.5 m ils) to about 76 m (3 mils ) and fluctuated rapid ly
between 25-38 m (3 and 4 mils). The thrust monitor showed an alert condition also.
The condition lasted two to three minutes and it was found that a slight increase in
speed caused the v ibrations to return to normal.

During a rainstorm several d ays later, another vibration excursion occurred

lasting less than a minute with symptoms very similar to the previous excursion.
The problem subsided when the speed was increased fr om ap proximately 9300 rpm to
9500 rpm. Several more vibration excursions occurred during rapid changes in wea
ther conditions over the next few weeks.

Due to the transient nature of the problem, the plant connected thei r tape
rec order to the proximity probe monitors and the operators were instructed to turn
the recorder on during any sudden changes in weather that might precipitate another
vibration excursion. Local thunderstorms resulted in three vibration excursions in
the next 24 hours.

23
 TYpical frequency spectra from just prior to and just after the start of a
vibration excursion are shown in figures 9 and 10 respectively while operating at
93 60 rpm (156 Hz). The transition time between figure 9 and figure 10 was less than
0.1 second. The overall vibration level had been 33 m (1.3 mils) with 15.7
(0. 62 mils) at running speed and approximately 5 (0.2 mils) at 62 Hz or 2/5 of
running speed. During the excursion. the vibration level increased to approx imately
76 m (3 mils) with the majority of this increase due to the 62 Hz component. The
amplitude of the sub synchronous component averaged about 53 m (2.1 mils). but was
fluctuating wildly over a range as wide as 13 to 89 (0.5 to 3.5 mils) but typi
cally from 38 to 64 m (1.5 to 2.5 mils). The orbit developed into a "double orbit"
from a relatively ci rcular one and the whirl di rection was noted to be forward.
Various other frequencies noted during the excursion were found to be sum and dif
ference frequencies of the two dominant frequencies of 62 Hz (subsynchronous vibra
tion) and 156 Hz (-running speed). These are shown in more detail in figure 11. A
very na rrow band analysis verified that the subsynchronous component was 2/5 of
running speed to within 0. 1 Hz.

The problem was thought to be due to a transi ent rub condition since:

1) the condition was closely related to weather conditions.

2) the w ide fluctuations in the subharmonic component.
3) that slight increases in speed caused the instability to
cease.

The uni t had a gr eat deal of hot p1p1ng a ttached. all of which wa's routed through
the roof and exposed to the weather. Significant piping strains associated with
transients was thought to have contributed to the units sensitivity to changes in
weather. The fact that the 62 Hz component was present a t a very low level prior to
the excursion was not thought to be very significant since a review of the vibration
records showed that it was almost always present wit h levels between about 1 to 5
(0.0 5 to 0.2 mils). Also. as stated earlier. it is not uncommon for high speed
units to display a low level subharmonic even during norm al operation. The reason
the whirl occurred at 62 Hz (3720 rpm) when the first critical was known to be at
approx imately 81.7 Hz (4900 rpm) remains unexp lained.

The possibility that the instability was an oil whirl condition precipitated by
a change in alignment. as has been known to occur. was considered but discarded for
several reasons. Fi rst. changes in alignment would probably not cause the almost
instantaneous change in subharmonic vibration level. Also. w hile subsynchronous
components due to oil whirl instabilities can fluctuate. they have not previous ly
been observed to fluctuate so rapid ly and over so wide a range. And finally. the
fact that slight increases in speed were sufficient to eliminate the instability was
not at all a characteristic of oil whi rl. The newly installed oil seals were
obviously suspected as the source of the rub. However. due to the violence of the
instability and the fact that these seals had probably rubbed at other times in the
unit's nine year hist ory. without causing an instability. it was decided that a ful l
dismantle inspection was i n order.

This dismantle inspection revealed that:

1) S tanding at the t urbine governor end. the governor end of the

rotor was misal igned 510 ( 20 mils) low and 254 m
(10 mils ) to the right and the coupling end was 10 1 m
(4 mils) to the right.

24
 2) A s l id ing key at the turbine go vernor end meant to maintain
axial alignment while sliding to accommo date thermal growth.
had f rozen preventing the case f rom mov ing axially.
3) The bearing oil seals were light ly rubbed.
4) The rotor was heavily rubbed a t the center by the interstage
labyrin ths.

It was concluded that the frozen footing was primarily responsible for the
rubs. The case was apparently bowing in the center quite severely during sharp
thermal transients as it tried to acco m modate the change in casing growth and pipe
strain. This als o was thought to be the reason for the rub during start-up since
the case again co uld n ot shift to acco mmodate thermal growth. Although other rubs
had probab ly occ urred at s ome time in the mac hine's history. there had been none
this severe as indicated by the depth of sco r ing on the rotor. This explained why
this problem had not occurred before.

The rotor was realigned and the frozen footing was freed. The machine has
s ince operated for a year and a hal f with n o further problems except for the usual
prob lem of thermal bo ws during startups.

CONCLUSION

Some cases of rub induced instab ilities have been des c ribed. These have been
noted to appear and cease very s uddenly. ca n sometimes be controlled during opera
tion by changing the alignment. and have sometimes been eliminated with minor in
creases in speed. On one occasion the whirl frequency was noted to track an in
crease in rotor speed so as to remain at 1/2 running speed. The whirl frequencies
are usually at 50 percent of running speed and the whirl direction has often been
forward. However. whirl frequenc ies have been observed at other fractions of run
ning speed. Also. whirl frequencies have been observed at speeds Significantly
lower than their first c ritical speed. S ome rubs have produced sum and difference
frequencies based on the whirl frequency and the running speed frequency.

REFERENCES

1. Den Hartog. I. P.: Mechanical Vibrations. Fo urth Edition. McGraw Hill Book
Company. New York. 1956. pp. 292-293.

2. Ehrich. F. F.: The Dynamic Stability of Rotor / Stator Radial Rubs in Rotating
Machinery. I. of Eng. Indus No v. 1969. pp. 1025-1028.

3. Bentley, D. E.: Forced Subrotative Speed Dynam ic Action of Rotating Mac hinery.
A.S.M.E. Paper No. 7 4-PET-16.

4. Chil ds. D. W. : Rub-Induced Parametric Exc itation in Rotors. A.S.M.E. Pape r No.
7 8-WA/DE-14.

5. Childs. D. W Fractional-Frequency Rotor Motion Due to Nonsymmetric Clearance

Effects. A.S.M.E. Paper No. 81-GT-14S.

25
 TABLE. I - SUMMARY OF VIBRATION LEVELS (milsp_p)

Point #3 Point #4

TllfE VERTleAL HORIZONTAL VERTICAL HORIZONTAL

IsX 1X Total IsX 1X Total IsX 1X Total IsX 1X Total

10:45pa -
0.5 0.6 -
0.35 0.4 -
0.8 0.95 -
0.6 0.75

11:43_ 0.5 0.6 1.7 0.8 0. 2 1.3 4.3 2.3 75 4.3 3.6 75

12:02pa 2.5 3.8 5.8 4.8 3.6 6.0

12:10pa 2.5 3.5 5.8 4.5 3.5 6.0

12:13pa 4.5 8.5 4.5 8.0

12:17pa 3.0 9.0 5.6 8.5

12:19pa 8.3 8.2

1:45pa 0.17 0.25 0.48 0.45 0.8 1.05 0.7 1.6 1.79 0.1 0.4 0.6

26
 1 X

0.10

Light Rubbing

0.05 2 X

..II'.
111
II)
c:a.
,...
u 0.0
II)
11.1 0 200 400 600 800
-
r:: Hz
or4
>,J

>-

or4
U
0
....
II)
>
lID
.9 0.10 X
11.1
111
U

Normal Operation
0.05

2 X

0.0
o 200 400 600 800
Hz

Figure 1. - Turbine case vibration on a 5000 kw turbine-generator.

27
 D LOCAliON OF OATA POJNT

pRAWING ! gCAJJQH Of pATA POIHTS

Cg"QH!t!HG TURBINe lope I KG 111881 HE CPMPRf5CPR CgMPpESSQR

1500 INLET
'550 , EXHAUST

Figure 2. Schematic of the Syn Gas Compressor.

170 Hz
1 X II.S.
2.5 .. U. 3:29 p ....
Water on WIst pedestal
60 Hz
Electrical
Interferenc.

85 Hz
li X II.S.
1.5 IOU. aVB.
bouncing
0.5-1.5 ,.u.

2 X R.S.
320 Hz 440 Hz
O+- ____________________u-__________
__

o 300 liDO 50
Hz

Figure 3. Topping turbine discharge end vibration -

with water on west pedestal.

28
 85 Hz
\ x R.S.
3.25 mils

170 Hz
60 Hz 1 X R.S.
2.5 mils
3,35 p.m.
Just after water wa.
taken off pedes cal.

1.5 X R.S.
0.5 mils

S.

300 oo oo
Hz.

Fi gure 4. - Topping turbine discharge end vibration -

water taken off pedestal.

170 Hz
1 X R.S.
2.5 U.

60 Hz

),47 p.m.
Few minutes after water
was put on east pedestal.
Amplitude at R. S. slowly
decreased for next hour
to 1. 5 mUs.

2 X R.S.

o to--------:--------L:--------------3jOO----------40O---e----------'5
0
Hz

Figure 5. - Topping turbine discharge end vibration -

with water on east pedestal.

29
 5 \ x R.S.
4.3 mils fluctuating

4 1 X R.S.
3.6 mils
p.
I
p.
<II
....
"" 3
a

60 Hz
1

0
0 800

Figure 6. - During vibration excursion.

Figure 7. - At conclusion of vibration excursion.

30
 Discharge
Suction

CPHQ'HSUM lUI".

SIHGU EXTWACTlOII COMP'USOII COJIIIR[SSOI CCMI'IESSOIt CQllltR[SSoa

II H. P.
9310 RPM

Figure 8. Schematic of the CO2 Compressor.

Prior to Ixc::ur.lon.
Pt. '1
0.75
. . 1. 3 .11. Over.ll
p-p

0.5

.
l


'"
........
:.i
ci M'"

",ci
. .
,;
o.a ... ==
...

== .." ==
,; ... ..
... '" ..
.. :!l
'"
N

.A A f\- A
o 100 200 100 400 OO
Hz

Figure 9. Turbine outboard bearing - just prior to excursion .

31
 Durin. Excuralost
Pt. 11

3.0-4.0 mll" _ Overall

p p
Fluctuations a8 wide 411 0.5-3.5 mUa
2.0

..

l.()o ..
.;

..
..
N
...
i
M
NO

. .

:II
...
.. :II
... ..
;::
A A
100 200 300 400 500
a.

Figure 10. - Turbine outboard bearing - during vibration excursion.

; 3

1
:lI 1
.
'"
N

i... s::1
...
t: f
;
-8
.3 .; >e
.. 3
:II . 3-
.. ; .:I,
!;1 .. "
:lI " A ; :lI
N >eN
..
... .:I ., ...
'"
; :lI 3
!

j
; ..
0 ;
+ .. !;1
!i j
!i 3

II
.:I
+
:lI
'" .:I 3 '"
'"
.. .. , ..
... :lI
..

Figure 11. - Turbine vibration - during excursion: Sum and difference frequencies.

32
 ANALYSIS OF A TURBINE ROTOR CXNrAINIl'G A

TRANSVERSE CR.AQ( AT OAK CREEK UNIT 17

G. W. Rogers and C. A. Hau, Jr.

Failure Analysis Associates
Palo Alto, California 94303

J. J. Kottke and R. H. Menning

Wisconsin Electric Power COmpany
Milwaukee, Wisconsin 53201

SUMMARY

Transient increases in one, two and three per revolution vibration characteristics
of a low pressure steam turbine were observed during steam temperature reduction oper
ations. Vibration and fracture mechanics analyses suggested the presence of a trans
verse shaft crack which was eventually identified by ultrasonic inspection and .
confirmed by destructive sectioning. Signature analyses of vibration data recorded
over a two-year period prior to crack identification are correlated with fatigue crack
growth, which occurred intermittently during transient temperature decreases. The
apparent increased response of the rotor to vibration is due to asymmetric stiffness
changes introduced by the growing transverse crack. The vibration response is pre
dicted to increase with increasing crack depths in excess of 10% of the shaft diameter.
Fracture mechanics analyses predict that fatigue crack growth occurred during periods
of steam temperature decrease, when high surface tensile stresses are present. These
same transient thermal stresses are shown to have retarded and prevented subsequent.
fatigue crack growth during steady operation.

INTRODUCTION

Failure Analysis Associates (FAA) was retained by Wisconsin Electric Power Company
(WEPCO) to analyze abnormal transient vibration levels observed during boiler deslag
ging operations of their Oak Creek Unit 7 3 20 MW General Electric turbine generator.
Of particular interest were the vibration levels recorded on main bearing caps of the
low pressure, 1800 rpm turbine/generator set.

Boiler deslagging was performed on this unit to dislodge slag that accumulated on
the outside wall of water and steam tubes. The procedure calls for reduction in fur
nace firing rates which causes cooling in particular tube regions. Subsequent thermal
contraction effectively dislodges mineral deposits.

The ',subject turbine/generator utilizes four journal bearings, one pair supporting
each of the turbine and generator rotors. The vibration characteristics of the bear
ing caps monitored when the turbine/generator was at a fully loaded condition were
essentially the same as those recorded during early stages of unit operation. Horiz
ontal, vertical,and axial vibrations were recorded for each bearing cap. In addition,
shaft rider sensors monitored vibration at an angle of approximately 250 from vertical.

33
 VIBRATION ANALYSIS

Figure 1a is the normal steady state vibration signature obtained with a shaft
rider probe at the number 1 turbine bearing. A strong synchronous 1800 rpm signal
and very slight two and three per revolution signals are present. Figure lb, on the
other hand, shows the signature at peak vibration levels subsequent to a thermal down
ramp of 1400F (770C). Note the increase in the one, two, and three per revolution
amplitudes.

The vibration characteristics of turbine bearings number 1 and 2 were similar.

Both bearings exhibited a significant increase in synchronous and twice-synchronous
signals during the transient temperature decrease.

A comparison of vibration data and other parameters indicated a correlation between

steam temperature and vibration amplitude. It was observed that a decrease in steam
temperature resulted in a delayed vibration level increase. Increases in steam tem
perature, on the other hand, either did not affect the vibration or reversed a vib
ration increase associated with a previous temperature decrease. As expected with
thermall y induced phenomena, there was a definite time lag between steam temperature
change and the observed change in vibration. An example of the vibrational response
associated with steam temperature variation is shown in Figure 2 . Reduction in steam
temperature cools the surface of the shaft relative to the shaft bore. This condition
produces tensile hoop stresses and tensile axial stresses of sufficient magnitude that
a transverse crack in the shaft near the surface would open. Such a crack was subse
quently identified near the mid span of the double-flow rotor under the inlet edge of
the first shrunk-on, turbine wheel of the governor end.

FATIGUE CRACK PROPAGATION ANALYSIS

Fractography of the cracked rotor shaft after removal from service showed the
presence of a fatigue crack of size and shape illustrated by Figure 3. Fracture
mechanics calculations of fatigue crack growth were performed by integration of the
material's crack growth rate law over the range of crack driving forces to obtain
the crack progression between some initial size and the final size.

The crack driving force is conveniently characterized by the crack tip stress
intensity factor K, which is proportional to nominal stress range o, the s quare
root of the crack size a, and the specific crack and part geometry. To compute the
specific magnitude of the steady (K ) and cyclic (K) crack driving force for the
s
crack shape modelled in Figure 3, we have utilized a general purpose fracture mechanics
computer program BIGIF [lJ. (K and K ) for shaft bending are
The crack driving forces
S
summarized in Table I. BIGIF was used to accuratel y compute K and K for the more
complex stress distribution near the shaft surface. This dist ibution is affected
by the stress.concentration associated with the diameter change, fillet radius, and
shrunk-on disk contact stress. However, because these localized surface stresses
affect the magnitudes of K and K only for cracks near a surface, it was not neces
sary to compute their magnI tude in order to quantify K and K for the deeper crack
g
depths between four and seven inches where vibration c anges were introduced.

The rate of fatigue crack growth in rotors made from Ni-Cr-Mo-V low alloy steels
has been studied by various investigators. The crack growth rate per cycle, da/dN

34
can be represented over a wide range of growth rates by a power law of the form

(1)

10
where C is 5.99 x 10- for Kmin/Kmax = 0 and increases with increasing K ; n is ap
proximately 2.8. At low valu:s of-tiK, the crack growth rate falls below hat predic
ted by equation (1), and eventually a threshold value of lIK !;, Kth' is ,reached below
=

which fatigue crack growth does not occur. The fatigue threshold, lIK ' is also de
h
pendent on the ratio of steady stress (0) to cyclic stress (1I0) , or t te ratio of K
to lIK. Specifically, Hopkins and Rau [2] have shown that the specific effect of t e
stress ratio, R =0
min 10max
= 1<L'11l 1n I 1<-1nax' can be estimated as

A-R
( R) = --8-- lIK ( R=O) (2)
lIKth th

where A and B are material constants and generally A B 1.6.

The nominal cyclic stresses during steady state operation are estimated by the
manufacturer to be 1.6 ksi (or lIo 3.2 ksi). With a bearing misalignment of 0.050
=

inch, the nominal cyclic stresses will be lIo 4 ksi. The steady stress from centri
=

fugal force and the small steady-state axial and radial thermal gradients should be
near zero and probably less than 5 ksi. However, during the deslagging operation, the
the steam temperature is quickly reduced by 100-150oF, and the shaft surface is cooled
relative to the bore, setting up a radial temperature gradient which introduces sur
face tensile stresses up to Os = E a liT (30xl03 ) x (6.6xlO-6) x l300F
= 26 ksi, D

where E = elastic modulus in ksi, a = coefficient of thermal expansion in inches per

inchoF, and liT change in temperature in of.
=

Utilizing BIGIF, the crack driving force, lIK, corresponding to lIo = 3.2, was cal
culated for a range of crack depths. BIGIF was also used to numerically integrate
equation (1) to obtain the increase in crack depth, a, caused by rotating bending
during the periods of the high steady stresses produced during deslagging. The results
are shown in Figure 4 for the case where C, in equation (I), is elevated by the high
mean stress during the deslag cycle to 5.7 times the value at in /ax = O.

CRACK GROWTH THRESHOLD CONDITIONS

Because the 1800 rpm turbine speed introduces fatigue cycles very quickly (1800 x
60 min/hr x 24 hr/day 2.6 million cycles per day) , crack growth could not have been
=

occurring continuously during normal operation. Otherwise the crack would have grown
to a size where the rotor would have failed. The reason that cracking does not occur
during normal operating conditions is that the II K produced is below the material's
apparent threshold considering prior loading history. In the absence of any steady
stress (0 = 0), the threshold of these rotor steels is typically between 6 and 10
ksi/in. fising a 6 ksi/in threshold value, the cyclic nominal stress at which a 6.7
inch deep crack would grow is obtained from Table I and equation (2) to be lIo 3.0 =

ksi. Therefore, since the nominal cyclic stress of lIo = 3.2 ksi with perfect bearing
alignment is greater than this value, fatigue crack growth did not occur under steady
state operation in this case because the high steady stress which caused fatigue crack

35
growth during des1ag also served as an overload, increasing the effective threshold
above which fatigue crack growth can occur. Figure 5 shows published results of
Hopkins, Rau, Leverant and Yuen [3 ] . These results indicate a major increase in the
apparent threshold stress intensity factor, K ' when preceded by an overload. The
th
larger the relative overload, the higher the apparent threshold for subsequent crack
growth. In the case of the OC7 rotor, the ratio of maximum overload stress to maxi
mum steady state stress is at least

26 + 1. 6
5 + 1.6 =
4.2
and perhaps as high as

26 + 1.6
17.
--U- =

The data of Figure 5 show that the corresponding apparent threshold for fatigue crack
growth during turbine operation after a des1ag cycle will be at least 1. 6 times higher
and may be more than five times higher than the threshold would have been without the
deslag overload cycle. Specifically, the cyclic nominal stress above which the 6.7-
inch crack can continue to grow when cr = 0, is between 1. 6 x 3 = 4.8 ksi and 5 x 3
= 15 ksi. If steady-state operation p oduces some small tensile steady stress, the
apparent threshold will be reduced by the factor computed by equation (2). Neverthe
less, the gnitude of the overload produced by the deslag was sufficient to prevent
continued propagation during subsequent steady operation.

CRACK GROWTH VS. VIBRATION INCREASE

The correlation of crack growth to changes in the vibration characteristics of the

turbine rotor involves many factors. First it has been recognized [4,5,6,7,8J that a
continuously open crack produces a twice per revolution rotor response due to a [email protected]
ing, asymmetrical stiffness analogous to an elliptical shaft geometry. An opening and
closing crack, on the other hand, may produce synchronous, two, three, four, and even
higher multiples of synchronous responses due to the form of rotor orbit.

The nature of the crack and the state of stress associated with the thermal tran
sients previously described suggest the the crack was, in fact, continuously open
during the periods of high vibrational response. Grabowski, in particular, [9] has
shown analytically that the change in shaft compliance and hence the dynamic response
due to the asymmetry is relatively insensitive to crack size until the crack depth is
approximately 10% of the shaft diameter.

As shown in Figure 2, the amplitude of the second harmonic vibration is near a

maximum for a period of approximately three (3) hours. It was estimated that the
crack is fully open for this period of time due to the state of transient thermal
stress. The total number of des1ag operations performed on the OC7 unit between in
stallation and the time such procedures were eliminated was approximately 1100. Using
the three (3) hour time estimate for each, the total number of high thermal stress
state cycles can thus be calculated as

1800 cycles x 60 min. 6

1100 cycles 3 hours 356 10 cycles
x x x
mi n.
(3)
Hr.
=

The calendar period over which these cycles occurred was reviewed and correlated to

36
recorded maximum increase in vibration during selected des1ag operations. The form
of the recorded data is unfiltered peak to peak displacement, in mils, as sensed on
the number 1 bearing cap in the vertical direction. A curv fit of these data as a
function of cycles under the thermal transient conditions as calculated in equation
( 3 ) is shown in Figure 6, Figure 7 shows the correlation of the maximum change in 10w
pressure turbine bearing cap vibration during des lag with predicted rotor crack growth.

The graph shows that the point at which the change in vibration becomes recognizable
corresponds to a crack depth between 3 and 4 inches: or between B% and 10% of the
shaft diameter. This correlates very well with the crack depth ( 10% of rotor diameter )
predicted analytically by Grabowski [9J as necessary for significant increase in rotor
vibration response.

At the time of this writing, sufficient data reduction to establish similar time
history amplitude curves for individual synchronous and twice-synchronous components
of the unfiltered bearing cap vibration data has not been completed. Figure 2, how
ever, shows an example of expected differences between the first and second harmonics.
Whereas the absolute magnitude of change in velocity for the two harmonics is of the
same order, the percentage change in the second harmonic is much greater than in the
first harmonic.

Vibration data for certain rotor coast downs were also recorded on magnetic tape.
Bode plots of amplitude vs. frequency for first, second, and third harmonics OD rotor
speed were generated as shown in Figure B. Note that the ordinate is a 10grithmic
scale of velocity. The first critical speed was observed to be approximately 1035 cpm.
The second and third harmonics responded at 1050 and lOBO cpm, respectively, most
probably due to the increase in stiffness in the bearing oil film with reduced rotor
speed.

MODIFIED UNIT OPERATION

Acting upon an initial belief that the rotor contained a transverse crack, but
prior to actual determination of that fact by inspection, WEPCO decided to continue
operation with revised procedures. This decision was based primarily on the fact that
steady-state, full-load operation showed no trend tlowards vibration increase and the
conclusion that the crack was propagating only during the des1ag cycles as previously
discussed.

The operating procedures were therefore modified to minimize high vibrations and
hence to minimize the magnitude of thermal down ramps. This was accomplished by limit
ing the unit to a constant 200 MW output, regardless of system demand. At the 200 MW
load, the boiler slagging is minimized, thereby eliminating the need for continual
des1agging cycles.

The procedure for taking the unit off line was also revised since it was believed
desirable to have crack closure while coasting down through critical speeds. This was
accomplished by keeping the shaft at approximately constant temperature. The method
employed brought the unit to a minimum load until vibration returned to normal after
which the load was dropped rapidly while maintaining high steam temperatures.

The result of these procedures was that the trend of continually increased vibration
level with age or des1ag cycles ceased. Also, no increase in steady-state vibration

37
was observed. These operating restrictions continued until the unit was shut down,
ultrasonically inspected, and the rotor removed for repair.

S UMMARY AND CONCLUSIONS

Some vibrational characteristics of a large steam turbine rotor containing a trans

verse crack have been identified. It was determined that the measured vibration sig
natures during thermal transient des1ag operations correspond to analytically predicted
responses for a continuously open crack. The vibration signatures also showed signi
ficant increase in the twice per revolution response of the rotor and an overall vi
bration amplitude increase with increasing crack depth.

Fracture machanics analysis, including the effects of prior "overloads" on subse

quent fatigue crack growth, was used to predict crack growth. The increase in bearing
cap vibration was correlated with the predicted crack growth. This trend compared
very well with the analytically predicted onset of rotor response to stiffness asym
metry caused by a growing transverse crack near the mid span of a rotor. It was shown
that a crack depth corresponding to approximately ten percent of the shaft diameter is
necessary to produce a significant change in the rotor vibration response.

Although there are many factors which affect rotor vibration response to cracking
(such as axial position of the crack or the phase relationship of the crack to rotor
eccentricity), it has been shown that vibration monitoring, rotor dynamics analysis,
and fracture mechanics analysis can be used to identify, predict and understand the
presence and growth of transverse cracks in large steam turbine rotors.

REFERENCES

1. Besuner, P.M., Rau, S.A., Davis, C.S., Rogers, G.W., Grover, J.L., and Peters, D.C.,
"BIGIF: Fracture Mechanics Code for Structures--Introduction and Theoretical
Background - Manual 1," EPRI NP-1830, April 1981.

2. Hopkins, S.W., and Rau, C.A., Jr., "Prediction of Structural Growth Behavior Under
Fatigue Loading," Fatigue Crack Growth Measurements and Data Analysis, ASTM STP
783, 1981.

3. Hopkins, S.W., Rau, C.A., Jr., Leverant, G.R., and Yuen, A., "Effect of Various
Programmed Overi1oads on the Threshold for High-Frequency Fatigue Crack Growth,"
Fatigue Crack Growth Under Spectrum Loads, ASTM STP 595, 1976.

4. Grabowski, B., "The Vibrational Behavior of a Turbine Rotor Containing a Transverse

Crack," Transactions of the ASME, Vol. 102, January 1980.

5. Mayes, I.W., Crack Propagation in Rotating Shafts, ASME 77-DET-164.

6. Mayes, I.W., and Davies, W.G.R., The Vibrational Behavior of a Rotating Shaft
System Containing a Transverse Crack, I. Mech. E. C168/76.

7. Okah-Avae, B.E., Analogue Computer Simulation of a Rotor System Containing a Trans

verse Crack, Simulation, Dec. 1978.

38
8. Zrebarth, H., and Baumgartner, B.J., Early Detection of Cross-Sectional Rotor
Cracks by Turbine Shaft Vibration Monitoring Techniques, 81-JPGC-Pwr-20.

9. Grabowski, B., "The Vibrational Behavior of a Rotating Shaft Containing a Trans

verse Crack -- Part 1 - Crack Models and the Vibration of a Rotor with Single Disk,
on a Massless Shaft Containing a Transverse Crack," (to be published) 1982.

39
 TABLE 1. - NORMALIZED CRACK DRIVING FORCE FOR CHORD CRACK IN CIRCULAR SHA FT
UNDER LINEAR STRESS VARIATION (BENDING)

Crack Depth K/a (AK/Aa) AK (for K

s (for
Aa " 3.2 ks i) a ,,25 ksi)
s
(inches) (ksi ITiiT (ksi Ilri)

0.99 1.395 5.02 34.88

1.98 1.940 6.98 48.50

2.8 0 2.235 8.05 55.88

4.00 2.630 8.42 65.7 5

5.60 2.990 9.57 74.75

6.70 3.260 10.43 81.50

7.92 3.510 12.64 87.75

11.20 4.060 14.62 101.50

15.80 5.440 19.58 136.00

I I I I I
0.2
-
(5.08) I-
u
III 0.1 -
en I-
(2.54) a
=

u 0
III
en
.....
c:
0.3 - -
I
(7.62)

0.2 b
u
0 - -

III
( 5.08 )
>

0.1 - -
(2.54) I,

.' J -' 1 .-L .-L

o
o 2 4 6 8 10 12
Frequency - Cycles per Minute x 1,000

Figure 1. - Vibration signatures of amplitude vs. frequency in a radial direction,

'
a) d u ring normal full,load operation, b) during peak, vibration subsequent to a
thermal down ramp.

40
 .4 r-----,----r----r_--_r__,

u
.3
QJ (7.62) u
C/l 0
...... 600
(315) u..
0
U QJ
QJ
C/l :::l
...... .2 +'
c It!
(5.08)
QJ
Steam Temperature 0-
E
QJ
b' I-

U 500 E
0 It!
(260) QJ
QJ
:> .1 +'
Vl
(2.54)

400
14 (204)
o 2 4 6 8 10 12
Elapsed Time (hrs)

Figure 2. - Vibration amplitude of first and second harmonic and inlet steam

temperature vs. time.

\

From UT
From Surface

20.72"
21.5" 13.
.,,
:: 13.75" :::::

Oak Creek #7 Shaft Diameter = 39.569"

--- BIGIF r'1odel of Crack Shape

Figure 3. - Representation,of Oak Creek #7 LP turbine shaft transverse cracking

including BIGIF model.

41
 8 20

",':
6 ....
v> Q)
ill .....
..c Q)
U E
'"
5
c

CI
'" , ....
0
.d 4 10 .....
0
..... '"
0. "-
ill ..c
CI .....
0-
-'"' 3 Q)
U CI
'"
'- -'"
u u
'"
'-
u

o L-__L-L-L-______-i__-L__-L__-L__-L______ __
__ __
-J
-
6
o 20 40 60 80 100 120 140 160 180 200 220 240 260 280 300 320 340 x10
Cycles

Figure 4. - Transverse crack growth rate due to deslag thermal transients predicted
using BIGIF transverse crack model.

/
4

UTHB
o

..
CJ
:0::
c:I
!'-2 0

:0::

Ti- AL-4V D_S. Nickel Alloy
CI
294 K 866 K
0 x R:O.l A-R:O.694
x o R:O.3 R:O.785
o R:O.5
t-R:O_887
o R:O.7

0- R:O.9

3 5 7 9 11
K OL /KB

Figure 5. - Relative change in fatigue threshold after single cycle overloads as a

function of the relative overload for the alloys an d R ratios shown.

42
 3.5 I I I I I I

-
0

3. 0 o Moni tared
I

Level
01
I

c
2.5
/
-
o
I
/o

'"
<- 2.0 i-
.n
::>-
"

) -
c I
1.51- 0 0

1.0 0
ry
I
-
-
0/

0.5 q, 60

/"
0 - -,..----r--;.,-I'"--- 1 0 I I

220 240 260 280 300 320 340 360 380

6
Cycles x 10-

Figure 6. - Maximum changes in LP turbine bearing cap vibration during deslag

as a function of total number of cycles at thermal transient stress state.

Removed
From
Service

8 r-r-r--.---.---r--,,--.------r-1 4"0

3.0 -;;
"iii
c
0
..,
"

'"
<-
'" 4 2.0 .a
:>
c

C1I
I 0>
c
'"
/ 1.0
.<:
u
/ E
"
/ E

/ "x

..,..../
---- ---- --- --- - -- --- ------- --- --- --- --- - -- ---
o 0
o 20 40 60 80 100 120 140 160 180200 220 240 260 280 300 320 340 360
6
Cycles (x10- )

Figure 7. - Correlation of maximum change in LP

during deslag with predicted rotor crack growth.

turbine bearing cap vibration

43
 1.0
u (25. 4)
QJ
III
'-
E
E
0.1
u
QJ (2.54)
III
'-
c:
.

I
0.01
1'S (0.254)
u
a

Qj
>-
0.001
en
a (0.025)
-.J

0.5 1.0 1.5 2.0 2.5

Frequency-Cycles per Minute x 1,000

Fi gure 8. - Bode plot of amplitude vs. frequency for first, second, and third
harmonics of running speed.

44
 ASPECTS OF CDULOVJB DAlVP lID IN ROfOOS

SUP:roRTED ON I-lYIECUYNAMIC BFARIIDS

P. G. M>rton
G. E. C. Power Engineering Ltd.
Lichfield Road, Stafford,
England ST17 4LN

ABSTRACT

The paper is concerned with the effect of friction in drive couplings on the
non-sychronous whirling of a shaft. A simplified model is used to demonstrate the
effect of large coupling misalignments on the stability of the system. It is
concluded that provided these misalignments are large enough, the system becomes
totally stable provided the shaft is supported on bearings exhibiting a viscous
damping capacity.

SUMMARY

Rotating damping is well known to be a source of instability in shaft systems

and many physical and mathematical explanations of the phenomenon exist. Published
work seems mainly to be concentrated on a viscous or hysteretic damping model for
the rotor, with simple dissipative bearings. The present paper considers non-linear
damping of the Coulomb type in a simple rotor supported on 'realistic' linearized
bearings, characterized by 4 stiffness and 4 damping coefficients.

The paper examines the nature of the sub-synchronous limit cycles resulting
from interplay between the Coulomb damping within the rotor and viscous damping in
the bearings. The rotating damping in the present model is assumed to arise from
rubbing within gear type couplings at both ends of the rotor. The qualitafive
findings are however applicable to any mechanism producing friction forces depending
on rotor flexure. It is also shown that as the couplings are misaligned in any
plane, a level is reached above which the sub-synchronous oscillations collapse.
A map of these limiting misalignments is compared with the shaft orbit under
conditions of perfect alignment.

INTRODUCTION

Rotating damping has long been recognised as a source of instability in shaft

systems. Both physical and mathematical explanations of the reasons for this
instability have been advanced. A good summary has been given by Crandall ( Ref. I).
Viscous and hysteretic damping is easily incorporated in a linear mathematical
model. In many practical cases, however, the damping arises from relative motions
within the rotating structure, the force-motion relationships obeying non-linear
laws. A Coulomb damping model gives a reasonable representation of the energy
exchanges resulting from slipping mechanisms, although to accurately incorporate
such a model in a rotor/bearing system presents some problems in analysis.

45
 Intuitively, one might suppose that instability due to any saturable type of
rotating damping would be subject to limit cycles. This is only so if the system
also contains positive damping elements with a more powerful energy-amplitude
ratio. Such a system is a shaft with rotating Coulomb type damping supported on
realistic, albeit linearized, hydrodynamic bearings. Very little published work
exists on this very commonly occurring system and the present paper deals with a
particular phenomenon of considerable practical significance.

The system dealt with is a rotor supported on two hydrodynamic bearings and
driven by a gear type coupling. The rotating damping arises from movements between
mating teeth of the coupling.

THE MODEL

The paper deals with an idealised model (Fig. I). The degrees of freedom have
been limited to 4, this being the minimum necessa4Y to give a qualitatively adequate
solution. The system comprises an isotropic non-damped shaft symmetric about its
centre of span and characterised by a single flexural mode. The shaft is
supported on two linearized hydrodynamic bearings, notionally represented as 8
stiffness and damping coefficients, each varying with running speed. Each end of
the shaft moves within a coupling which has a fixed slope but is free to move in
translation. The laws governing the force-velocity relationship within the
couplings can be either of the viscous or Coulomb type. Provision is made for
altering the coupling slope in both vertical and horizontal planes.

SYMBOLS

AxXll etc. kg Elements of Inertia Matrix

- Subscripts denote
BXXll etc. Nsm 1 Elements of Damping Matrix
position in matrices.
-
EXXll etc. Nm 1 Elements of Stiffness Matrix

B Parameter concerned with viscous rotating damping (Eqn.7)

Bxx etc. Non dimensional oil film damping

BIXX etc. Dimensional oil film damping

b Dimensional viscous damping of all coupling teeth moving

axially

C m Radial clearance of bearings

Exx etc. Non dimensional oil film stiffness

l
E xx etc. Dimensional oil film stiffness

ex, ey rad Angular misalignment of couplings, Planes x, y

F N Coulomb damping force generalised (Eqn.12)

f N Coulomb damping force of all coupling teeth moving axially

g ms-Z Acceleration due to gravity

L m Half span of shaft

46
M kg Half mass of shaft
-
m kg/m 1 Mass of increment of shaft

qlX etc. Generalised co-ordinate. Shaft centre x plane

r m Radius of coupling teeth

t s Time
-
Vx, Vy ms 1 Generalised velocity of rubbing of gear teeth

Wc (Nm) Work due to Coulomb damping/cycle

Wv (Nm) Work due to viscous damping/cycle

x, y Horizontal and vertical planes

y Viscous rotating damping factor (Eqn.7)

b. Modal parameter (Eqn.4)

OX, oy Misalignment parameters (Eqn.9)

<p Mode shape of shaft

<Po m Modal co-ordinate at shaft centre

<pal e (rad) Modal slope at shaft end

X Co-ordinate vector (Eqn.2)

\Ii Modal parameter (Eqn.6)

-
A rad.s 1 Natural frequency of pinned-pinned shaft
-
w rad.s 1 Lowest natural frequency
-
rad.s 1 Running speed

EQUATIONS OF MOTION

The system co-ordinates (Fig. 2) relate to movements in the horizontal and

vertical planes. The coefficients of the equations of motion are expressed as
elements of square matrices using a subscript code where x, y determine the plane
and , 2 relate to the shaft and bearing respectively. Thus AXY12 would appear in
1
a matrix as follows:-

-]
-1- ,AXYI2
- I
.4 I ( 1)
_1-
I 1- Y

In the systems with viscous rotating damping the equations of motion are:-

AX + BX + EX = 0 (2)

47
 Dimens ional Non Dimensional Inertia
2 2
I: nt<b
AxXll AYYll I:m( 1 )
<1>0 =/).
= =

AXX22 = AYY22 = M

Axx12 = AYY21 = Axx21 =

AYY21 =
I:m(
t
o
)

I
BXX22 = B xx

l
BX Y22 = B xy g Bxy/stC

l
B YX22 = B yx g Byx/stC

I
BYY22
=
B yy g Byy/stC

br 2 1!1
2
br 2 1!1 2
BXXII BYYll B - 2 "
_

( (3)
<1>0 ) (<1>0 )
=

2
= -

/). M
2 2
EXXII = AxXIIA A
/).
2 2
EYYll = AYYIIA A

I
EXX22 = E xX g Exx/C

I
EXY22 = E xy g Exy/C

l
EYX22 = E yx g Eyx/C

EYY22 g Eyy/C

br2 1!1 2
( ) st
2M <1>0

Regarding the shaft geometry, since the equations of motion are set up in terms
of the pinned-pinned shaft modes <1>, then it is convenient to assume a simple shaft
geometry. The choices are a disc/weightless shaft combination or a uniform shaft.
The latter model has been chosen and this results in a sinusoidal modal shape for
which:-

2
/). = I:m2 /M<I>o = 0.5 (4)
\II =
I:m/M<I>o = 2/rr (5)
1
1!
rr/2L (6)
<1>0

Roots of equation 2 give both system frequencies and dampings. If the system
damping factor is plotted against the rotating damping (Fig. 3) it is possible to
find the reserve of viscous damping possessed by the system. The viscous rotating
damping factor is conveniently expressed as:-

(7)

y being given as a proportion of critical damping of the shaft vibrating in flexure.

48
 The damping B arises from the axial rubbing of the gear teeth in the coupling.
These teeth are assumed to be continuously distributed on a pitch circle of radius
r. The value b relates to the total viscous force/unit velocity obtained from
sliding the coupling axially. The values of all parameters used in plotting the
curve of Figure 3 are shown on that sheet and will be used for all further work in
the paper. The bearing coefficients relate to a specific non-circular bearing
profile of the fixed arc type.

Solutions of the linear equation of motion yield not only the system frequencies
and dampings but the orbit shape, the size being, of course, arbitrary.

The coupling damping operates on the relative velocities of teeth arising from
small vibrations in the system. It is possible to generalise these velocities in
terms of the shaft co-ordinates as follows:-

(8)

where ox, oy relate to the angular misalignments of the coupling ex, ey thus:-

oy = ey
(til) (9)

The energy reserve/cycle in viscous damping at the coupling is given by:-

(10)

Coulomb Damping

Suppose now that the damping mechanism in the coupling is of the Coulomb type
and that f is the force resisting axial movement of the teeth. The Coulomb energy/
cycle is:-

Wc (I J)

2r thal f r f
where F = ( ) =
(12)
M
- - -

'IT <Po L M

Note that strictly speaking Vx, Vy are not the same in the Coulomb case as in
the viscous case, but for reasons which will be discussed later, the error involved
in assuming both orbits to be identical ellipses is very small.

Plotting the work function (Wc - Wv) against orbit size for zero misalignment
(Fig. 4) yields the magnitude of a limit cycle at the cross-over point. The fact
that the cross-over slope is negative indicates that this is a stable cycle. The
orbit size which refers to the length of the major axis of the ellipse at the shaft
centre is proportional to f/M.

49
 Misalignments

Having fixed the orbit size for the system under consideration, the values ex,
ey are now systematically varied. Figure 5 shows that depending on the values of
these parameters the character of the work function changes. Under some circum
stances the function shows 2 cross-over points. The smaller orbit is unstable and
the larger one is stable. Certain combinations of the misalignments give a totally
stable system, there being a threshold at which a collapse from a finite orbit to
zero takes place.

It is possible to plot a map of constant orbit combinations as in Figure 6.

Here it will be seen that there exists a boundary in the ex/ey plane beyond which,
for the model assumed, instability cannot exist. Within the boundary, limit cycles
exist which are smaller than the limit cycle of the zero misaligned case.

DISCUSSION

It is first necessary to examine the validity of the model with regard to its
general applicability. The semi-rigid modelling of the shaft implies certain
constraints. This does not, however, affect the qualitative behaviour of the system,
particularly that relating to the stabilizing effects of large misalignments. It
does, however, mean that the system frequency and the orbit size will be incorrect,
but to obtain a more accurate model it is only necessary to incorporate more shaft
freedoms. Either a finite element or a modal representation could be used.

The present mathematical treatment of Coulomb damping also requires justifi

cation. It is reasonably easy, if tedious, to arrive at limit cycles by time
marching methods or other slightly more elegant algorithms. Alternatively, a
carefully constructed analogue computer circuit can solve the problem. All of these
techniques have been used by the author on the particular problem presented in the
present paper and for this case it can be shown that the orbits are very nearly
elliptical with little harmonic content. Thus the assumptions used in deriving the
energy expression (Wc - Wv) are valid. The reasons for this are fairly obvious.
The energy supplied by the Coulomb damping is a small proportion of the inertial
and strain energy in the system so that the orbits are largely controlled by these
latter energy exchanges. This circumstance results from the fact that the shaft is
very flexible, as evidenced by the small ratio of frequency: running speed (w/n
Figure 3).

The question arises as to what happens if the shaft is stiffer. As shaft

stiffness increases, so does the stability margin. This margin is expressed in the
present context as the viscous rotating damping necessary to reduce the system
damping to zero. The author has found that reasonably accurate results can occur
when this viscous damping margin is well over 15%.

Any rotating damping, whether viscous, hysteretic or Coulomb, reduces the

stability margin of a rotating system. This means that non-sychronous whirling
will start at a lower speed than that dictated solely by consideration of the oil
film. Since most rotors contain some rotating damping mechanism, then extractions
of the oil film coefficients from rotor/bearing systems may well be subject to
inaccuracies.

50
 I
More important, however, is what, in the authors view, is the erroneous
deduction from many tests and case histories, that oil film non-linearities are
responsible for the small limit cycles often experienced prior to large non
sychronous orbits. The probability is that these limit cycles arise from the
existance of friction in the drive mechanism. To be sure oil film non-linearities
operate to limit journal orbits which are large compared with the bearing clearance,
but this is a gross condition in which the oil film becomes effective around the
complete bearing circumference.

In the present mathematical example, a single running speed has been considered
so far. Obviously as this running speed rises and its associated oil film co
efficients change, the stability margin reduces. According to our simple model the
orbits should increase with running speed and at the 'resonant oil whirl' threshold,
should grow without limit. Under these circumstances, however, it is obvious that
constraints associated with bearing clearances etc. will take over so that the
increase in orbit size may not be as dramatic as Figure 7 predicts. In any case
the present theory deals with small vibrations and above a certain level the force/
spatial relationships within the coupling may become much more complex than those
assumed here.

The nearer the running speed approaches the 'true' oil film instability
threshold, the easier it is to supply the small amount of friction damping necessary
to promote limit cycles. Figure 8 shows a shaft weighing 700 kgs supported on 2,
100 mm dia hydrodynamic bearings. Figure 9 shows friction being introduced by 3
small steel 'fingers' supported on a freely rotating layshaft. The pressure between
the fingers and the main shaft was introduced by elastic bands. This very small
amount of damping produced from the dynamic slope of the shaft was quite sufficient
to lower the instability threshold and produce stable limit cycles. The stabilizing
effect of large misalignments was also demonstrated by varying the angle of the
layshaft.

CONCLUDING REMARKS

The author has dealt with the subject of limit cycles in a shaft/hydrodynamic
bearing system with rotating damping present in the drive coupling. The same
conclusions, i.e. that limit cycles exist when saturable damping mechanisms are
present, would apply whenever friction was caused by shaft flexure. The point
brought out in the present paper is that the destabilizing effect of Coulomb damping
can be 'biased out' by friction arising in a fixed plane and resulting from the
rotation. In many practical situations such a biasing would be impossible in that
very heavy fretting wear would be brought about by the gross misalignments necessary
to achieve stability. Nevertheless, the mechanism is of some interest and may not
only apply to couplings, but to other sources of rubbing in rotating elements.

REFERENCES

I. Crandall, S.H.: Physical Explanations of the Destabilizing Effect of Damping in

Rotating Parts. NASA Conference publication 2133, May 12-14, 1980, Pages 369-382.

51
Fi gure 1. - Diagrammatic arrangement of shaft and couplings.

Figure 2. - Coordinate system.

52
 /
Ol"'l
System Damping %
/
!- = C!).lll
c!H1.3&t- 02.35 Jf 01 ... 3
__ ...... -- ..... __ .t'
'.---
Ie -_ ........... -

A 80 r ad!sec
M .. 17000 kg
C .. 300 1Jm
r .. 300 DID
L .. 2.0 m

01 05

Figure 3. - Relationship between system damping and rotating viscous damping.

Net Work
Per Cycle! F2
M).'2

2.0

Figure 4. - Net work N orbit size.

53
 o ---------------r--------.------r-------'
...... 10/ 20 r,30
-I \ I

-2

t\ '
I
-3
\

\ I
\

-5 \
ox ..:L
-F- F
Curve
' MV MV

\J
1 10 0
\ 2 0 5
-7
3 20 0

Figure 5. - Effect of misalignment on work function.

54
 oy

Stable

ox

Critical Misalignment

Figure 6. - Critical misalignment boundary compared with

orbit of fully aligned system.

Orbit Ratio
30
Ret 50 Hz

Resonant Whirl

20

10

Running Speed Hz
O -------'r---'--'
50 '-0 70 80

Figure 7. - Growth of orbit size with running speed.

55
Figure 8. -Model rotor supported on 100mm dia. hydrodynamic bearings.

56
Figure 9. - Method of introducing friction damping into model rotor.

57
 INSTABILITIES OF GFARED COUPLIN3S - 1HBJRY AND PRACTICE

R. Gordon Kirk, Roy E. Mbndy, and Richard C. MUrphy

Ingersoll-Rand COmpany
Phillipsburg, New Jersey 08865

SUMMARY

The use of couplings for high speed turbocompressors or pumps is essential to

transmit power from the driver. Typical couplings are either of the lubricated
gear or dry diaphragm type design. Gear couplings have been the standard design
for many years and recent advances in power and speed requirements have pushed the
standard design criteria to the limit. Recent test stand and field data on contin
uous lube gear type couplings have forced a closer examination of design tolerances
and concepts to avoid operational instabilities. Two types of mechanical instabil
ities are reviewed in this paper: (1) entrapped fluid, and (2) gear mesh instability
resulting in spacer throw-out onset.

Test stand results of these types of instabilities and other directly related
problems are presented together with criteria for proper coupling design to avoid
these conditions. An additional test case discussed shows the importance of proper
material selection and processing and what can happen to an otherwise good design.

INTRODUCTION

In many instances of pre-order coupling selection, little consideration is

given to either the dynamics of the coupling spacer as an individual body or its
half weight effects on the shaft ends of the coupled units. Some efforts may be
made to minimize weight but typically standard couplings are selected from a horse
power-speed aspect. This method is adequate in most cases because the spacer be
haves as a rigid body with minimal influence on the connected shaft ends. It is
this minimal influence on adjoining shaft ends and lack of response cross-coupling
across the spacer which enables a single body analysis to accurately represent the
individual unit's characteristics in the field.

The general technique for studying spacer tube dynamics is to consider it as

a uniform beam with simple supports at each gear mesh as a worst case in the un
torqued condition. As the torque level being transmitted through the spacer in
creases, the theory allows the coulombic gear mesh forces to restrain the spacer
tube ends from rotating and effectively raise the spacer's critical speed. For
some equipment strings the gear coupling's spacer tends toward this behavior.
However, on units with flexible shaft ends, light journal loads, and large shaft
end separation, the spacer cannot necessarily be considered to adhere to this
proposed theory.

The amount of torque which a gear coupling is transmitting at a particular

speed not only affects the amount of moment restraint which the adjoining shaft
ends may induce upon the spacer but also greatly influences a mesh instability
phenomena known as spacer throwout. Spacer throwout occurs when the mass unbalance

58
forces on the spacer become sufficiently great as to cause the spacer to move off
of the gear mesh pitch line creating a step function increase in the unbalance
forces being exerted on the adjoining shaft ends.

In attempts to minimize the potential for throwout problems, the tendency is

to specify a coupling with the lightest spacer possible which is capable of trans
mitting the required torque. Employing this philosophy may result in a spacer tube
calculated critical falling near design speed. If the spacer tube critical is at
least 120% of design speed, then satisfactory operation should result according to
Staedeli and Vance ( ref. 1 and 2). On large center hung machines with short, stiff
shaft ends, this criteria may be acceptable. However, on machines where the shaft
ends are flexible and the coupling half weight is a substantial percentage of the
total journal loading, then the 120% criteria can result in unacceptable system,
lateral dynamics.

In order to alleviate this potentially problematic coupling - shaft end con

figuration, the tendency is to increase both inside and outside spacer tube diameters
while decreasing concentricity tolerances and gear mesh clearances. The result is a
special design of coupling which may be significantly more expensive than the orig
inal, standard coupling would be. Although this coupling design would eliminate the
amplification effects associated with operating near the spacer tube critical and
reduce the maximum displacement possible due to throwout, the increase in spacer
tube weight may nearly offset the gains of the coupling design change. Therefore,
if the coupling is not properly specified, one set of dynamics problems may be
simply traded for another set of problems. One major problem which may exist in
this increased diameter spacer design is that of entrapped coupling lubricant within
the spacer tube. Although this type of phenomena is documented in the literature
( ref. 3,4,5) it can be difficult to diagnose during field operation due to limited
vibration data acquisition capability. Another more important fact is that the
vibration characteristics may not match the reported theories in all regards as this
paper will illustrate. Therefore, efforts to preclude oil from entering the spacer
should be included as part of the coupling design.

The remainder of this paper documents several case histories of turbomachinery

vibration problems wherein the source of the problem was concluded to be inherent in
the coupling design selected for each particular application. Further discussion on
the theories and methods of calculations for coupling selection and analysis follows
the review of the test results. Design standards are given and recommendations are
made to assure the best possible dynamic response characteristics for high speed
couplings.

REVIEW OF TEST STAND DATA RELATING TO GEAR COUPLINGS

The following case histories exemplify the basic problems described above.
Each case history spotlights a particular problem encountered and what actions were
required to correct the problem. Since all corrective actions were made on the
coupling itself and none were made on either the test driver or driven unit, all
problems were considered to be initiated by the coup1ing1s dynamic characteristics
and its interaction with the adjoining units.

59
 Case 1 - Spacer Critical Speed and Balance

The first coupling to be considered is shown in Figure 1. The shaft end sep
aration is 1727.2 mm (68 inches) and the coupling wei9ht is 79.1 kg (174 pounds).
The spacer is 149.2 mm (5-7/8 inches) 0.0., 136.5 mm (5-3/8 inches) 1.0. This cou
pling is rated for 11.6 MW (15,500 horsepower) at 95 Hz (5700 RPM) with a maximum
continuous operating speed (MCOS) of 100 Hz (6000 RPM). The calculated spacer
critical speed is 131 Hz (7860 RPM) or 1.31 times MeOS. In the field this coupling
transmits power from an overhung design expander to a center slung design axial flow
compressor. The compressor was tested using this coupling and at 100 Hz (6000 RPM)
the compressor1s bearing vibration adjacent to the coupling was approximately 0.025
mm (1 mil), Figure 2. This same coupling was then used to drive the expander for
its mechanical run test.

The set-up for the expander test is shown schematically in Figure 3 where the
variable speed motor drives the expander through a 5:1 speed increasing gearbox.
The initial results of this test are depicted in Figure 4 which represents the ini
tial run up to speed. As can be seen, vibration levels are acceptable at 83.3 Hz
(5000 RPM), approximately 0.010 mm (0.4 mil), but the amplitude rises sharply above
83.3 Hz (5000 RPM) approaching 0.051 mm (2.0 mils) at 91.7 Hz (5500 RPM). The am
plitude was nearly all at synchronous frequency. This rapid increase in synchronous
response amplitude is usually indicative of a critical speed. A review of the rotor
dynamics analysis of the expander, including the coupling half weight, did not indi
cate a critical speed in this speed range. Upon running an analysis of the test
system, the indication was that a system critical existed which was primarily con
trolled by the coupling. It was necessary to field balance this coupling placing
balance weights at the spacer midspan to enable a satisfactory vibration level to be
achieved. The field balance required numerous trials due to the sensitive nature of
the coupling spacer critical. Figure 5 shows the final balance condition with an
acceptable level of 0.038 mm (1.25 mils) being achieved at 100 Hz (6000 RPM). The
existence of the critical speed is indicated by the sUbstantial shift in phase angle
above 91.7 Hz (5500 RPM).

Case 2 - Gear Tooth Damping

Several proposals were considered to resolve the vibration problems of Case 1.

One proposal was to coat the coupling gear teeth with a product which the coupling
manufacturer recommends for the running in of new couplings. This coating is a
polyphenylene sulfide resin placed on the flanks of the teeth to help the teeth
share the load. After a reasonably acceptable level of balance was achieved by
field balancing, this coating was applied to the coupling teeth. Figure 6 depicts
the results. With the coating on the coupling, the vibration as measured by the
expander1s coupling end vibration probe reached .076 mm (3.0 mils) at 91.7 Hz (5500
RPM) as seen on Curve 1. The coating was removed by a light sandblasting; the
balance of the coupling was considered to be unchanged. Curve 2, showing 0.025 mm
(1.0 mil) at 91.7 Hz (5500 RPM), reflects the results upon rerunning the coupling
on the test stand. Since the only change between these two runs was the removal of
the coating, it was concluded that the difference in response was due to an increase
in the coulombic damping in the gear meshes which resulted from the removal of the
slippery coating that had been applied to the teeth flanks.

60
 Case 3 - Entrapped Oil

Since test stand field balancing of the coupling was not considered as an
acceptable permanent fix, an investigation of prior experience with similar con
figured machinery was undertaken. One result of this investigation was the discov
ery that on other jobs the ratio of calculated lateral critical speed of the cou
pling floating member to maximum continuous operating speed was substantially greater
than on this application. This ratio was found to vary from about 1.9 to 6.3. It
was concluded from this data that for a permanent fix, a coupling was required
wherein the spacer critical to MCOS ratio would have a minimum value of 2.0.

The coupling design resulting from this study is shown in Figure 7. The shaft
end separation is still 1727.2 mm (68 inches), but the weight has been increased to
97.7 kg (215 pounds). The main spacer has an 0.0. of 235 mm (9.250 inches) and an
1.0. of 223.9 mm (8.813 inches). The calculated critical speed is 207.6 Hz (12,457
RPM) or 2.08 times an MCOS of 100 Hz (6000 RPM).

In order to confirm the acceptability of this coupling, it was run on the next
similar unit to go on the test stand. Figures 8-a through 8-d depict the initial
results. An extremely high level of vibration would suddenly occur with a very
small increase in speed. As can be seen from a chronological review of the figures,
the speed at which this phenomenon would occur was lower with each successive start
and the source of the high amplitude was a subsynchronous component varying from
.94/rev. on the first start to .83/rev. on the fourth start. Although the amplitude
at synchronous frequency was never higher than 0.015 mm (0.6 mil), the subsynchronous
component reached levels approximating 0.241 mm (9.5 mils) at 84.8 Hz (5090 RPM) on
the first start.

The first step in disassembly of the unit was the removal of the coupling.
Upon breaking the flanges, a substantial quantity of oil was observed to have been
entrapped in the spacer tube. In order to permit the oil to escape from the spacer,
twelve 6.35 mm (0.250 inch) holes were drilled in the spacer; four at each end and
four in the middle. The spacer assembly was check balanced and found to be accept
able. The unit was reassembled. Figures 9 and 10 show the results of the retest
run. Figure 9 shows that the subsynchronous component, due to the entrapped oil,
has been completely eliminated and Figure 10 indicates that the critical speed,
previously seen at about 96.7 Hz (5800 RPM) has been eliminated. An additional
modification has since been incorporated into this coupling design to preclude oil
from entering the spacer from the gear mesh area. This modification consists of
interference fitting dam plates in the ends of the spacer. The oil drain holes are
retained.

Case 4 - Manufacturing Tolerances

Of necessity, couplings similar to that of Figure 7 consist of a number of

pieces which must be assembled. Where possible, rabbet fits are used to maintain
consistency in reassembly. However, fits may have clearance and may not be con
centric with each other which can result in either constant or variable unbalance
being manufactured into the coupling. This same problem can also exist for the
fits and pilot diameter of the gear meshes.

The coupling of Figure 7 has a floating member weight of 84.1 kg (185 pounds)
which results in an unbalance of 106.7 gm-cm (1.48 ounce inches) for each 0.025 mm

61
(1 mil) of floating member eccentricity. Figures 11 and 12 graphically depict what
can happen as a result of manufacturing tolerances and eccentricities. Figure 11
indicates an overall amplitude of 0.081 mm (3.2 mils) at a speed of 96.7 Hz (5800
RPM). For this test run the coupling was assembled using the manufacturer1s match
marks. The assembled coupling was indicated with mechanical dial indicators and
depending upon axial location, ran out as much as 0.127 - 0.152 mm (5-6 mils). When
ignoring the manufacturer1s match marks, and rotating components relative to each
other, the run out was reduced to under 0.051 mm (2 mils). The test results for
this configuration are shown in Figure 12 where the overall vibration level at 96.7
Hz (5800 RPM) is approximately 0.037 mm (1.45 mils).

Case 5 - Throwout of Mesh

As discussed in the introduction, gear type couplings require a minimim torque

transmission in order to center the floating member. In the absence of this minimum
torque, the residual unbalance in the floating member may cause it to move eccentric
until the tooth tip clearance is reduced to zero in the heavy spot1s radial direction.
This situation creates an additional unbalance in the coupling.

This phenomena is graphically displayed in Figure 13a, where a speed reducing

gear box is being driven by a gas turbine with the load compressor uncoupled. As
long as the accelerating torque is applied, the vibration amplitude gradually in
creases to a level of 0.013 mm (0.5 mil) at 209.2 Hz (12550 RPM). Since the accel
erating torque decreases as the unit approaches idle speed, the vibration quickly
increases to 0.018 mm (0.7 mil) at 211.3 Hz (12680 RPM). When torque is reapplied
through the coupling during acceleration to MCOS, the vibration amplitude immediately
drops to .009 mm (0.36 mil) and gradually increases again as the driving torque de
creases while the unit comes up to speed.

Figure 13b shows a separate test run for the subject equipment string. Both
acce1 and decel signatures clearly show the response step increase associated with
coupling spacer throwout. The speed-amplitude drift signature is included as addi
tional information to show the absence of a response critical of any kind in this
speed range.

Case 6 - Coupling Failure

Figure 14 depicts a peak hold plot of overall vibration during the startup
of a unit similar to the one just described above. As can be seen, the amplitude
starts rising rapidly above 120 Hz (7200 RPM) and peaks at about 0.330 mm (13 mils)
at 160 Hz (9600 RPM). The unit was tripped and after coast down, the coupling guard
was disassembled. Figure 15 shows the condition of the coupling spacer. Three large
cracks were evident. The darkened area above the center balancing band was caused
by the coupling rubbing the guard. The driven gear box showed no distress and the
coupling was not transmitting load as the load compressor was not coupled for this
start nor had it been coupled. Metallurgical investigation later determined that
the cracks had been formed during the heat treatment cycle of the spacer.

62
 THEORY AND GUIDELINES TO PROPER COUPLING DESIGN FOR
ROTOR DYNAMICS CONSIDERATIONS

Entrapped Fluids

The analysis for the dynamics of a rotor partially filled with liquid was
published by Wolfe (3,4) and Ehrich (5) in the mid-sixties in the U.S. Wolfe
references earlier work from Europe dating to the late fifties. Other authors have
more recently written on the subject of entrapped fluids (6,7) which refine the
fluid dynamics and mathematics of the work by Ehrich. The published theory of en
trapped fluids has been proven by experimental tests on vertical rotors wherein the
system undergoes a zone of instability after passing a "reduced critical speedll
This reduced critical is not a function of the fill ratio but is determined by cal
culating the system critical with an equivalent rotor mass. The equivalent rotor
mass is found by adding the mass of the entrapped fluid required to fill the rotor
full of liquid to the dry rotor mass. This theory indicates that higher and higher
speeds may be reached as the rotor is filled with liquid before the instability on
set occurs. The test stand data presented by Ehrich indicated the presence of a
0.87 x synchronous speed component that presented a speed zone of large response
that the jet engine was able to pass through. The results of his analysis seemed to
satisfy the test stand results of the jet engine for the one run he reported.

The results of the more extensive test runs with increasing volumes presented
in this paper indicate that a mechanism of instability exists that produces a lowered
threshold of instability as the volume of entrapped fluid increases nd further that
the whirl rate varies from almost synchronous for small volumes to 0.84 x synchronous
as the volume increases. The difficulty in applying the simple theory to a real
machinery situation is the identification of the critical of concern and the effec
tive mass of the rotor. Additionally, the theory indicates a nonsynchronous compo
nent at 0.50 x synchronous for small volumes of entrapped fluid with whirl ratios
increasing as the rotor is filled with more fluid. Wolfe noted that when his exper
imental rotor was unstable it whirled for the most part at frequencies below the
reduced critical speed, the exact ratios not reported .

. No attempt is made in this paper to apply the published criteria for insta
bility onset since the nature of the observed instability differs from the theory
as published to date. The practical solution is to prevent the admission of oil
into a long spacer and further, to provide ample discharge outlets along the spacer
length. The oil flow in gear couplings should not be oversimplified since the
current design problem was considered by the manufacturer with the conclusion that
it was impossible to get oil into the spacer. This indeed is logical if one does
not consider axial pumping action or gear mesh oil ring instabilities that could
increase the oil ring thickness. A minimum dam height at the spacer ends of double
the normal gear mesh oil ring thickness is considered essential and preferably a
solid blockage plate should be used on all coupling spacers that can entrap oil.
The severity of the problem is amplified when lightly loaded coupling end bearings
are present such as in the case of overhung expanders, turbines, compressors, or
lightly loaded gear boxes.

When consideration is given to the potential imbalance from spacer 1.0. non
uniformatives that can produce a nonuniform oil film in the spacer, the admission
of oil into a spacer is quickly concluded to be highly undesirable.

63
 Example 1

Consider the imbalance of 0.254 mm (10 mil) thickness of oil, 50.8 mm (2.0
inches) wide, along a 1524 mm (60.0 in.) spacer tube with an 1.0. of 223.5 mm (8.8
in.), and an oil having a weight density of y = 0.0307 lb/in3

U = y x V x 0/2 = .0307 lb/in3 x (2.0 x 60.0 x .010) x (8.8) x 16

U = 2.59 oz-in or 186.7 gm-cm

(A typical coupling component balance is made to a level of 0.1 oz-in).

Coupling Spacer Critical Speed and Amplification

Manufacturers of high speed couplings typically calculate a spacer lateral

critical by considering it as a simply supported beam as the worst case (see Eq. 1).
The general opinion is that the critical is typically higher due to the frictional
constraint and damping arising from the torque condition for a gear coupling. This
type of constraint is not present in a dry diaphragm type coupling.

For large center hung turbomachinery this calculation is typically adequate

with the bearing reactions effectively giving a restraint to validate the approximate
calculation. For lightly loaded coupling end bearing machinery, however, the coup
ling represents a large portion of the modal mass of the coupling end critical
(typically the turbomachines 2nd mode) and hence the system sensitivity to imbalance
on the coupling is important not only at the spacer critical but also at the rotor
system criticals. The standard practice of using coupling half weights located at
their respective c.g. 's is considered totally acceptable for machinery having the
spacer critical ratio of approximately 2.0 or larger.

The analysis of turbomachinery having the coupling spacer critical speeds less
than 2.0 times the maximum continuous operating speed should include a response anal
ysis of the coupling joining the two adjacent rotor bearing systems. This type anal
ysis is commonly referred to as a train or partial train analysis. The detailed
train analysis results in the identification of the actual system critical speeds
and response sensitivities. The justification for this is supported by the result
ing amplification or deviation from a speed-square response as the spacer critical
speed ratio Ncr/N becomes less than 2.0. This is shown in Table I where it is noted

deviation is indicated for Ncr/N = 1.25 .

that a 33% deviation from a speed-square curve exists when Ncr/N = 2.0. A 176%
This means that any residual imbalance
force has been amplified by 176%. For Ncr/N = 1.1, an amplification of 425% would
occur for the residual imbalance forces to the adjoining rotor bearing systems.

The proposed design procedure for preliminary selection of couplings is as

follows:

1. Size coupling for torque capacity requirements.

(; ]
2. Check placement of spacer Ncr by the following equation (American Units)
.7
Ncr = 4 3
L
106 D2 + -J
we
(RPM) (1)

1 + L{02-d2)

64
 where

D =
spacer nominal outer diameter, in.

d =
spacer nominal inner diameter, in.

We = effective modal mass of flanges or torque meters referred to the

spacer midspan, lbm.

L =
mesh to mesh distance, in.

Spacer is assumed to be of standard steel having a Young's Modulus, E =

28.7
X 106 lb/in2 and weight density, y 0.283 lb/in3 =

3. If the spacer is in the satisfactory region given in Figure 16 (Zones I and II,
and the upper part of Zone III) for each of the adjoining machines journal re
action ratio, 1/2 Wcplg/Wjournal, the design specifications need not require a
train analysis.

4. If the design falls in the marginal region of Figure 16, a train or partial
train analysis should be performed for test stand and field conditions if the
spacer cannot be redesigned to move into the satisfactory region.

5. If the design falls in the unacceptable region the coupling spacer should be
redesigned to move the calculated critical as far as possible toward the satis
factory zone, analysis being made as noted above.

Coupling Throwout

The previous discussion on amplification is important to be able to fully

appreciate the potential problems relating to the mesh throwout instability.
According to the results given in reference (1), a gear coupling in perfect balance
will center on the mesh pitch line when the torque transmitted is either equal to o r
larger than
T M x E xw2 x D/4
=
(2)

where

M =
mass of floating member

E =
spacer eccentricity to be centered

D =
mesh pitch diameter

w =
rotor speed, rad/sec.

This torque is typically very small for standard tooth tip clearances.

Exa m p l e 2 (American Units)

Assume the following conditions:

E =
.001 in. =
.0254 mm

N =
6000 RPM; w =
628.3 rad/sec.

65
 W =
160 lb. = 72.6 Kg

o =
9 .0 in. =
228.6 mm

Applying Eq. (2) and converting to horsepower gives

2 (W E w2 x 1/12) x N
H. P. = =

33,000 33,000

H.P. =
17.52 to center mesh at 6000 RPM.

Consider now the influence of the spacer imbalance on the centering action of the
mesh. For a coefficient of friction of 0.15, it may be shown that the centering
will not be maintained for an imbalance larger than

1 .07 X 1010 x H.P.

UT.O. =
( American Units) (3)
0 x N3 X Q

where

H.P. =
transmitted horsepower

0 =
mesh pitch diameter, in.

N =
rotor speed, RPM

Q =
spacer amplification factor

U T.O. =
oz-in. imbalance to cause throwout

Example 3

H.P. =
500 =
372.5 kw

0 =
9 in. = 228.6 mm

N =
6000 RPM

Q = 5.0
1 .07 X 10 1 0 x 500 0.55 oz-in (39.6 g-cm )
UT.O. = =

9 x (6000)3 X 5

This imbalance represents an eccentricity of 0.43 mil TIR of a 160 lb. spacer.
Hence, the likelihood of centering under unloaded or lightly loaded test stand
conditions is very small for this example coupling.

A formula given by Staedeli (1) referred to as the Conti-Barbaran formula,

which was developed from test stand experience, gives an estimate of goodness:

P X 109 (4)
E = ( Metric Units )
1.36 x M x 02 X N3
where

P =
power, KW
M =
spacer mass, Kg
o = pitch dia., meters
N =
speed, RPM

66
 For E ? 10, stable and trouble free

E < 5, problems with mesh centering

Converting to units of H.P., lbm, in.,

1.8 7 X 1012 x H.P. (American Units) (5)

E =

W X D2 X N3

Example 4

Consider the calculation of Eq.(5) for the conditions of Example 3:

1.8 7 X 1012 x 500 .334 --- Problems

E =
=

160 X 9.02 x 60003

For H.P. = 10,000

E = 6.68 - - - Problems not likely

The above formula is similar to Equation 3 for the calculation of UT.O. The
advantage of Equation 3 is that it gives a result that has physical meaning and gives
the designer an imbalance level to achieve for best results.

The concern for mesh centering is two-fold in that gear wear would increase
with the mesh off the pitch line and secondly, imbalance of the spacer is a function
of the eccentric position of the floating member. The nominal tooth clearance at
zero speed is increased at running speed and temperature to give typically 0.051 -
0.076 mm (.002 - .003 inch) diametral clearance. For the 72.7 Kg (160 lb.) spacer
of the previous example, this would represent an additional imbalance level of 160
x 16 x .003/2 278.3 gm-cm (3.84 oz-in).
=

Typical balance standards would call for a balance of

U = 4W/N = 7.21 gm-cm (0.1 oz-in)

For 40 WIN; U = 1.0 oz-in (72.1 gm-cm)

Hence, the throwout imbalance is over 38 times the desired balance level and 3.8
times the practical balance level. This type of imbalance further amplified by
proximity to a spacer critical is not desirable for turbomachinery.

Guidelines for Manufacturing Tolerances and Balance Requirements

The design and manufacture of long couplings required for machinery such as
reported in this paper must be treated with due concern for the potential problems
previously discussed and demonstrated by test stand experience. The design of such
couplings typically necessitate a multicomponent design spacer with bolted flanges
and close tolerance fits. It is therefore of utmost importance for couplings falling
into Zone III of Figure 16, that the concentricities of the individual components be
held below certain minimum standards. Component balance of the hubs, sleeves, and

67
the floating member is required. The final coupling assembly, as mounted on the
turbomachinery , must have a total indicator reading on each of the ground balancing
journals on the floating member of less than .051 mm (.002 in). This requirement
necessitates that the coupling manufacturer achieve runout levels of less than
.038 mm (.0015 in) for each balance journal relative to its mounting bore or rabbet
fit. This criteria is relaxed for couplings in Zone I and Zone II as indicated on
Figure 16.

When the floating member balance journal runout levels cannot be reduced to
the required levels by selective component assembly , it becomes necessary to require
an arbor balance of the total coupling assembly. The concentricities of the balance
arbor must be controlled to preclude significant arbor induced imbalance which can
result in a built in imbalance when a correction is made and the coupling is removed
-

from the arbor.

For couplings having a spacer critical speed ratio larger than 2.0, 2 plane
correction on the floating member is adequate. For couplings having the spacer
critical lower than 2.0, correction on the floating member should be made in three
or more planes along the spacer lengt h, to reduce unwanted modal amplification of
the residual imbalance .

CONCLUDING REMARKS

This paper has presented test stand data that points clearly to the impor
tance for rigorous design standards on couplings for turbomachinery, with added
precautions for machines that may have lightly loaded coupling end bearings.
Turbomachinery likely to experience this condition includes power turbines and
compressors of an overhung design. The major conclusions that may be drawn from
this paper are:

1. Coupling designs that are below the marginal zone of Figure 16 should not
be accepted for Turbomachinery application.

2. Coupling spacers should be desig ned to have a critical speed ratio of 2.0 or
larger. Smaller ratios may be acceptable if adequate analysis is conducted
to assure an insensitive design as outlined in this paper.

3. Design features of gear couplings should preclude the admission of oil into
the spacer tube.

4. The manufacturing tolerances for couplings should meet specific minimum

.

standards as outlined in this paper.

5. Gear couplings shoul d be designed to have a minimum increase in mesh tip

clearance at design speed and temperature.

6. Turbomachinery conditions having low horsepower requirements and utiliz ing

g ear coupling s are particularly susceptible to larger than normal imbalance
levels on the coupling due to spacer throwout.

7. Long spacer couplings having a c ritical speed ratio less than 2.0 should
have at least three planes for balance correction.

68
8. The balance procedures outlined in this paper could apply to all couplings
but are considered essential for those applications in Zone III of Figure 16.

9. Attention to proper material processing and heat treatment is essential in

addition to the above consideratins.

REFERENCES

1. Staedeli, 0.: Toothed Couplings. MAAG Gear Wheel Company, Ltd.

8023 Zurich, Switzerland.

2. Vance, J.: Influence of Coupling Performance on the Dynamics of High

Speed Power Transmission Shafts. ASME Paper 72-PTG-36, 1972.

3. Wolf, J. A., Jr.: Whirl Dynamics of a Rotor Partially Filled with Liquid.
ScD. Thesis, Massachusetts Institute of Technology, Cambridge, Massachusetts,
Sept. 1966.

4. Wolf, J.A., Jr.: Whirl Dynamics of a Rotor Partially Filled with Liquid.
Journal of Applied Mechanics, Trans. ASME, Dec. 1968, pp. 676-682.

5. Ehrich, F. F.: The Influence of Trapped Fluids on High Speed Rotor

Vibrations. J. of E. Ind., Trans. ASME, Nov. 1967, pp. 806-812.

6. Saito, S. and Someya, T.: Self-Excited Vibration of a Rotating Hollow Shaft

Partially Filled with Liquid. Journal of Mechanical Design, Trans. ASME,
ASME Preprint 79-DET-62.

7. Lichtenberg, G.: Vibrations of an Elastically Mounted Spinning Rotor

Partially Filled with L iquid. Journal of Mechanical Design, Trans. ASME,
ASME Preprint 81-DET-54.

69
 T ABLE I. - UNDAMPED MODAL A MPLIFICATION DUE TO APPROACH TO A CRITICAL SPEED

Dimensionless Percent Deviation

Speed Ratio Speed Speed-Square Undamped Response from
Ncr Speed-Square Curve
-

.! .-!... - Q
rrQ
-

N s/ 2
cr 1_Q

CO 0 0 0 0.0
4 .25 .0625 .0667 6.4
3. .33 .1089 .122 13.2
2.5 .4 .16 .19 19
2.3 .434 .189 .233 23.2
2.0 .50 .25 .333 33 (33.04)*
1.81 .55 .302 .434 43
1.67 .60 .36 .5625 56 (55.5)*
1.42 .70 .49 .96 96
1.25 .80 .64 1.77 176
loll .90 .81 4.26 425
1.0 1.0 1.0 CO en (900)*

*Denotes value
for typical
damping value
of Q - 10

MARINE TYPE REDUCED MOMENT TYPE

SPACER END SPACER END

Figure 1. _ Expander original coupling design employing both marine and reduced
moment spacer design features.

70
 .
i l. l l !
b'.'6. I TII U J i .1 I II I!I I Ii jll,l l,!-/lllJ 1',. tl Ij!uiUlmlwm1tm- m
II
III rmmnml i II i ri iii Ii iii i I iI
1
:!
., 'II II' i i !i Ii : i ! III I I
:+1
-tt ,
I',
O
i !I 111w11 xI ' 'll ' Ii- !' \.!-,.' ,
I ;++li-
j
l +HH- H1+l+I

i.....

f+l+tt
t t + t ttH+
I

.
tt
tI ! r. '
r

I
' "

t ' I
]': lI,n it 1 :m1 HJI
tHH-tHH-
SYNCHR NOUS
II ! iii II

1 2 3 4 5 6
ROTOR SPEED X 10-3 (RPM)

Figure 2. - Acceptable axial compressor vi,bration ( speed vs. amplitude ) signature.

Four-hour mechanical test run.

TEST SET UP

EXPANDER
GEAR
BOX
5:1
SPEED
RATIO
SUBJECT COUPLING
1750 H.P. @ N 6000 RPM
1750 RPM
MCOS =

VAR. SPEED
MOTOR

Figure 3. - General configuration of expander test stand assembly.

71
 -
o -
2 3 4 5 6
-3
ROTOR SPEED X 10 (RPM)

Figure 4. - Expander vibration signature for initial acceleration on test stand.

til
..:r
H

11' i[+ : ,m m 'Ti' li l i!, ! ['!-; l .J;J ! , ..cl,' i [I'I, il i flhh;, :ti., "j,,, ; ;
jl!!
I I:r i : IIii I trdr IIfltT t, Hi ,1 , ,:[ [ ti l', I IF' 'iJj' [,:; I-:"-
l:, iili' i ; II I Ili !ilj -rhl nij l! . ;:
i' , "

' :1 I
._,

I!I I I : Ilii
'

o !
il II i :

2 3 4 5 6
-3
ROTOR SPEED X 10 (RPM)

Figure 5. - Expander vibration signature for final b alance condition of

coupling on test stand during mechanical run.

72
 --- -+ -t-- ---t-I +-+---+'-'-'-+'-
I
- +-'-'--+l/III---+'-'--+-'---'I
-+-'- -F-
2
--+---,- . !
--I ----ri- i -+-:-:-'-i---'-+-'-'-+----+c-'-:J:-c:..+:-:..:.:+:-'-'-'-I
- I--+--

: ___ L c j-,---- '----.j'_cc ,_+---'-+--'-

-- -'-+ -'-- "--lJl-fi-'--+-'-'---
-i
-: '1-
1 1"--+: ---'-+---'-i .LJ::d-
- -11--+-: ,--+--,-+i ---,-+--+i--lf--J ---,++'''-f.1
ffi
--c I-:+ i _+
- ! .... l)
[:l 1--- -- ---- _
- --- -----

2 3 4 5 6
-3
ROTOR SPEED X 10 (RPM)

Figure 6. - Test stand vibration signature comparison for balanced spacer with

and without polyphen ylene tooth coating.

Figure 7. - Expander redesigned coupling with in creased spacer tube diameters.

73
 ___J_
I---,-_r.;.--_
.,

.
.
. __ . ___
.
. ----I--L-+--.
3 tI--'--'.j.I...
::..:c
r,'
Case History

g0. , Expander Mechanical Test

C-.. ..- ... First Test Run =
-- -- ---- Coupling End Horizontal Probe
... .

-' -J- i _._ b:O'013 j---.---t- ...

!1 I
__

! .
.

1 ! i ' i

( a) First acceleration with entrapped oil.

. ::t.

"c:!: .. _ . . _._ -

(b ) Second acceleration with entrapped oil.

Figure 8. - Test stand vibration spectrum signature.

74
 ==F::::;: ---
r;sG$-':i =V:-T-
:r-c-,
-;c
, r,--;- -'-:Tt:r= Case History 3

4704 RPM __ ,____+:-_'

1----- --'- -

,h-'-p

(c) Third acceleration with entrapped oil.

---"
'--f------

(d) Fourth acceleration with entrapped oil.

Figure 8. - Concluded.

75
 , I" It:!.:.. "'

0.013 ... ====

5925 RPM - MCOS

5475 RPM

---- . :.

5205 RPM

4500 RPM

3975 RPM

_ _ = ___ =:c_

3000 RPM
-F=--

=
--=
--.;=- .
+
2475 RPM ... Case History 3
-":..:i: =--==t:-:::}-=c:- " E
.:

-
Expander Mechanical Test
-'" Final Test Run

;
_=
2025- RPM
-

-
--
=-=
- -, Oil Drain Holes in Coupling
Coupling End Horizontal Probe


_=_ E=--"::- -- __ ::T:::_-: __ __
- =
-_:':;=::i>-===-;':'-"-c:--=_ -
--- ='-:::'::--.f-::-- - :=-::

Figure 9. - Test stand vibration spectrum signature after oil drain holes were
drilled in spacer.

1 2 3 4 6
-3
ROTOR SPEED X 10 (RPM)

Figure 10. - Test stand vibration signature for retest run with oil drain holes in
coupling spacer.

76
 II
til
H
H
I!
:;:

4 5 6
-3
ROTOR SPEED X 10 (RPM)

Figure 11. - Test stand vibration signature with coupling assembled u sing
manufacturer's match marks (5-6 mils TIR).

'Ii! I11I I lilt Ii 11111 ! 111111 ill II

til
H
H
x:

o 2 3 4 5 6
-3
ROTOR SPEED X 10 (RPM)

Figure 12. - Test stand vibration signature with coupling assembled

to minimize coupling spacer runout imbalance.

77
 ROTOR SPEED X 10-3 (RPM)

(a) Exemplifying spacer throwout.

Ui
...l
....
:f! 1.0
p..
I


0

....
...l
'"

r.:I
Ul

'"
Ul
r.:I

: !: .::! : i ;! ! I ! i I i
I
I !
:
i .: : . . . : . . . . . . ..
0.0
6 9 12
ROTOR SPEED X 10-3 (RPM)

(b) For acceleration and deceleration showing spacer throwout.

Figure 13. - Gas turbine test stand vibration signatures.

78
 4 tT:

3 ..- ....

..;::::.. :::;:: ::-:-

0 9 12
0 3 6
-3
ROTOR SPEED X 10 (RPM)

Figure 14. - Gas turbine test stand vibration signature for the condition of a
cracked coupling spacer.

(a) (b)

Figure 15. - Failed coupling spacer.

79
 .!i NOTES ON OPERATION

:!:l 1- EXTENSIVE
0 PROBLEMS
0 2 - SENSITIVE TO TEST
0
l3 CONDITIONS - ZERO
oil TIP CLEARANCE ON

GEAR MESH
3 - MANUFACTURING
"Wl
0
TOLERANCES HAD TO

..: BE IMPROVED
....
J -12 ... 4 - NO PROBLEMS ON TEST
5 - NO PROBLEMS
6
., ... SENSITIVE. GOOD
U ....
-

.... TOLERANCES REQUIRED

iii 7-16 - NO PROBLEMS
III :z:
I 0
... - - N
U 9 8

-
7

Cl
'

P-
III

UNACCEPTABLE
--- GEAR TYPE _
- - - - DIAPHRAGM 0

0
0 .2 .4 .6 .8 1.0
WEIGHT RATIO (1/2 Wcp1g/Wjnl)

Figure 16. - Chart for coupling selection showing degree of acceptibility with
overhung turbomachinery. Experience overplotted. Zones indicate runout
level of balance bands relative to the mounting bore or rabbet fit.

80
 SHAFr VIBRATIONS IN 'IUROOVl!\QU NERY EXCITED BY rnAQ{S

Burkhard Grabowski
Institut fur Mechanik
Universitat Hannover
Hannover, Federal Republic of Germany

SUMMARY

During the p ast years the dynamic behavior of rotors with cracks has been
investigated mainly theoretically. This paper deals with the comparison of analytical
and experimental results of the dynamics of a rotor with an artificial crack. The
experimental results verify the crack model used in the analysis. They show the
general possibility to determine a crack by extended vibration control.

INTRODUCTION

Frequently cracks in turbine rotors have been found. Nevertheless, until now
it is not really"known, how cracks can be recognized early enough so that large
ensuing damage can be prevented.

The latest greater damage in Germany occured in the nuclear power plant in
Wlirgassen (ref. 1). For one year the plant had to be put out of operation because
two new rotors had to be produced. Both of the LP-rotors had a crack in about the
middle of the shaft. A higher level of vibration amplitudes at running speed was
noticed and during rundown the resonance amplitudes were very large, but these
vibrations were not assumed to be caused by a crack. The crack has been discovered
accidentally. This demonstrates the unawareness of the influence of a crack on the
vibrational behavior of a shaft. Certainly there is to take into account that cracks
in practice are comparatively rare. On the other hand Stodola (ref. 2) in 1922 has
already shown on principle the effect of a crack and these investigations were
continued by other authors (refs. 3, 4, 5).

As far as known the first measurement results of a rotating shaft with a crack
were published by Mayes and Davies (ref. 6) in September 1980 and at last by Inagaki
et.al. in September, 1981 (ref. 7). Like in Wlirgassen an abnormal increase in shaft
vibration was noticed in many other cracked rotors. But in some cases the warning
signs were too small and some rotors were broken. Some turbine plants even virtually
exploded and fragments of the shafts flew away upto a few hundred meters. Henry and
Okah-Avae (ref. 8) also present cases in which deep cracks have been found without
any influence on vibration amplitudes.

The cause for shaft vibrations due to a crack is the asymmetric cross section
at the crack position in connexion with the self-weight of the shaft. Depending on
the depth of the crack, on its position and on the damping of the system, the crack
can excite small or large vibration amplitudes or the system can even become unstable.

81
 CRACK MODEL

For a location above the horizontal'

diameter, the crack is subjected to only
compression and the entire cross-section
is s upporting. When further rotated, a
part of the crack area opens. A model
according to figure 1 was selected to
present this gradual change. At the
angle of = 900 the complete opening
of the crack is assumed. Thus in the
range between = 900 and = 2700
the simple model for the complete gaping
crack is taken for valid. Due to the
opening and closing of the crack area we
call it a breathing crack in opposite to
an always gaping crack, e.g. a cut.

The main axes of the second moments

of area at the crack position are not
Figure 1. - Crack model cross-sections
fixed to the body anymore. However,
showing rotation-angle-dependent stressed
this fact will have no influence on
regions ( shaded areas ) .
the numerical calculation, since space
fixed coordinates will be used anyway
because of the non-conservative sleeve
bearings. The second moments of area
along the vertical and horizontal axes
and the deviation moment are needed.

When the crack opens the change in

stiffness along the axis of the rotor
does not jump, but the change is conti
nuous in the neighbouring range.
Buerhop (ref. 9) examined the effects =90'
of such cross-sections in more detail
and has come to the conclusion that
the 450 approximation (wedge angle Figure 2. - Simulation of decreased cross-
a = 900) which has been applied up to section open crack location.
now, is sufficient. In order to simplify
the mathematical model, the wedge-shaped cut-out in figure 2 is approximately re
placed by a square cut with L = T .

A clear picture of the stiffness depending on the angle of rotation can be

attained when observing the static deflection of the cracked shaft due to its self
weight during one revolution.

Where a constant gaping crack is concerned, the shaft would move on a circular
orbit twice per revolution. Where a breathing crack is concerned, the shaft moves
only once on an onion-shaped orbit. The calculated curves of deflection in figure 3
are in good agreement with measured results of Ziebarth et. ale (ref. 10). In order
to compare the shapes of the two curves the level of the measured and the calculated
amplitudes have been equalized.

82
 y [""I
In addition to the development of the
theoretical crack model the stiffness of a -.8
cracked shaft has been.investigated. In a
shaft of 46 mm diameter and 300 mm length E
( fig. 4) the crack was simulated by a thin !. -.9
milled transverse cut of 0,5 mm width. a
.!:!
This shaft was loaded in such a way that
-1.0
near the crack a constant bending moment
existed.
-1.1
Parallel to these measurements the
stiffness of the shaft with a gaping crack
-1.2
has been calculated. At the crack position
a square cut with L = T was assumed, but
only for crack depths up to 50 percent. -1.3
For deeper cracks L has to be reduced
analogically. .0 90.0 180.0 210.0 360.0
Angle of rotation [degree J
X [""I
In figure 5 the results of measurement
and calculation are compared. The total com
.2 - Cal culation(cracl< model)
pliance of the shaft depending on the ---- Measurement (ref .10)
crack depth is plotted. The difference E
E
between measurement and ca,lculation amounts to =- .1
!i!
less than 10 percent, except for crack depths c
o
N
of 20 and 70 percent. In consideration of ..
.O----------+----- ______ -.
the simple crack model even with these d iffe
rences the agreement is remarkably good.
-.1

-.2

.0 lID. 0 180.0 210.0 3&0.0

Angle of rotation [degree)

Figure 3. Example for selfweight'

deflection for a breathing crack of 50
percent depth.

F
I Measurement plane
Material: 42 CrMo4

----2------

------ 300--------

Figure 4. - Shaft with milled cut.

83
 200

ETz
==-=
100
(5
cS
..
u
c;
"
Ci.
E 50
0
u

..
IO_---- - Calculation (crack model)
Q Measurement

50-IO-2O--3O--'O'--50-'6O--'7O-80
Crack depth [o'oJ

Figure 5. - Compliance at t of shaft depending on crack depth.

o

ANALYSIS

The mathematical model for a real rotor with cracked cross-section is a system
of differential equations with time dependent coefficients. An exact solution is
not possible. The vibrational behavior can, however, be calculated by means of
numerical integration. For a large system, this kind of procedure with digital cal
culation takes a lot of time. Therefore, the behavior of the rotor can be described
by its first few eigenfunctions, where the eigenfunctions Yo(x) and Zo(x) Gf the
uncoupled conservative system for the vertical and the horizontal planes are a good
approximation.

The range of the local change of stiffness due to a crack covers the length L.
A mean _ stiffness between open and closed crack will be taken approximately for
the calculation of the eigenfunctions in this region. This is necessary for the
consideration of the major curvature at the crack position, compared with the un
cracked shaft.

The deflections n the vertical direction y(x,t) and in the horizontal di

rection z(x,t) will be composed by the first few eigenfunctions weighted with the
generalized time dependent coordinates q (t), as follows

y(x,t) (I )

z(x,t) ( 2)

The transformation, which includes the complete bearing stiffness and damping coeffi
cients, yields a system of K + L coupled equations of motion for the generalized
coordinates q(t) which is

84


Mq + Cq + K (t) q = F (t) + g (3)

M, C and K can be considered as mass matrix, damping matrix and stiffness matrix.
On the right-hand side the function F (t) includes the out-of-balance distribution,
while g contains the self-weight load.

The time dependent elements in the stiffness matrix K (t) can be separated
and we obtain

Mq + Cp + [K + K (t) ]q F (t) + g (4)

m

A Runge-Kutta-Fehlberg procedure of 5th order with constant step-size (ref. II)

which proved to be very exact for the necessary computing time, will be used by the
numerical integration for the calculation of the instationary vibrational behavior.
The necessary initial values will be taken from the stationary solution of equation
(4) without the term K (t) . The time variable system has a stationary periodic
solution.

After the retransformation according to equations (I) and (2) the deflection
of the entire rotor which depends on the angle of rotation is obtained. The first few
harmonics for certain positions on the rotor can be determined by means of Fourier
analysis.

A detailed presentation of the mathematical method is given in references II

and 13. Additional theoretical results are in references 13 and 14.

EXPERIMENTAL AND THEORETICAL INVESTIGATIONS

Experimental Rotor and Crack Propagation

To confirm the theoretical results of the vibrational behavior of cracked

shafts experimental investigations were carried out. Figure 6 shows the experimental
rotor, which is supported by journal bearings in an experimental Helium-compressor
housing. At first it was intended to produce the crack by anpscillating load.
However, other experiments with this kind of load have shown, that a crack surface
due to such a treatment is different from a crack surface which is produced by an
alternating load. Therefore, we decided to produce the crack during the rotation of
the shaft in the rig itself by application by an external force.

Due to the high bearing load as a result of the external force F (fig. 6)
the journal bearing at the crack side of the rotor was replaced by a ball bearing.
The load was afforded by a ball bearing, too. To reduce the necessary amount of
the external force the diameter of the shaft at the crack position was reduced to
a diameter of 45 mm with a cut radius of 5 mm . For crack initiation a 4 mm deep cut
was sawn with a thin wire of 70 m diameter (fig. 7).

For crack propagation an external force of 5000 N and 3000 N, depending on the
crack depth, was necessary during about 3 h. The crack propagation was controlled by
observing the twice per revolution resonance amplitude near th crack poston
using a 2-channel Fast-Fourier-Analyser. The rotating speed was equal to correspon-

85
 pickup
S7.S8 r-t;t'50
.
S6
jll3
-------- ,.

o
--- ...
N

1---- 500 --------2.5

I------
---- D --
----

Figure 6. - Experimental rotor.

ding resonance frequency. Unfortunately,

two cracks came into existence. Probably
on the opposite side of the cut a small
groove was the cause for the second crack.
Therefore, the comparison of the theore
tical and experimental results is uncertain.

For the theoretical calculations we

have assumed a linear shape of the crack
ground (fig. 7) and a crack depth of 45 %
of the diameter. We intend to carry out
additional experiments with this rotor,
e.g. with di fferent unbalances. Therefore,
we did not break the shaft at the crack
position.
104---- 45 ----.I

Figure 7. - Cross-section at crack

position.

Theoretical Model of the Rotor

The vibrational behavior of the rotor has been caluclated with the FEM and the
transfer matrix method. To make use of the crack model with L = T , the geometry
in the neighbourhoo.d of the crack had to be modified. We have assumed one element
of a diameter of 64 rom and a length of 32 mm (fig. 8). This gives the same flexural
shape as the real geometry.

The change of stiffness due to the crack is the same as in the case of the
original rotor. But d ifficulties result from the uncertain knowledge of the stiff
ness and damping coefficients of the j oumal bearings.

86
 o
...
N
---------

.,.

1----- 500 ----------265

1------1030
1--- 135' ------------------------
------____ --1

Figure 8. - Rotor model for calculation.

The vibrations excited by a crack depend severely on the curvature of the

eigenmodes at the crack position and on the weight influence (see ref. 1 3).
If
R,
f g (x) Y
ok
(x) dx 0 (5)
o

gravity constant,

shaft weight/unit length,

eigenmode,

the crack does not excite any vibration. This is approximately the case for the
second eigenmodes in the horizontal and the vertical plane (figs. 9 to 12) and
accordingly for the third and fourth complex eigenmode of the complete system
(fig. 1 3 ), too.

So, already the eigenmodes give an idea about the speed ranges in which
crack-excited vibrations can be expected.

THEORETICAL AND EXPERIMENTAL RESULTS

Eigenfrequencies and Eigenmodes

The results of the numerical investigation of the vibrational behavior of the

cracked rotor are obtained by using the uncoupled eigenmodes of the conservative
system (figs. 10 and 12). The modal transformation includes the complete bearing
stiffness and damping coefficients.

Figures 8 and 9 and figures 10 and 11 show, that the difference between the
calculated eigenfrequencies of the uncracked and the cracked rotor amounts to
approximately 3 %. The measured frequencies in figure 14 (uncracked) and figure 16

87
(cracked) differ approximately 12 %. This indicates that the assumed mean stiffness
reduction in the calculation of the eigenfrequencies is too small. In the case of
the uncracked rotor the measured and calculated eigenfrequencies are in good agree
ment.For comparison only the first few eigenmodes of the complete coupled nonconser
vative rotor (fig. 13) are calculated with FEM.

Rotor with Cut

An interesting experimental result is depicted in the figures 14 and 15. The

cut with a depth of approximately 10 % of the diameter has no influence on the once
per revolution vibration amplitudes. This corresponds to the theory for a gaping
crack. (The small difference may be the result of a change in the distortion due
to the storage during some days.) The amplitudes at low speeds seem to be due to the
runout. The rotor is neither balanced nor equipped with an additional unbalance.
The twice per revolution vibration amplitudes may also contain runout, but less
than 1 m. The resonance amplitude increases up to 16 m. One or two additional
resonance frequencies can be observed at higher speeds. The small amplitudes
of these resonance vibrations can be explained by the weight influence (see eq. (5 .
The rotor with cut has not been theoretically investigated.

Cracked Rotor

As mentioned, the missing knowledge of the exact bearing stiffness and damping
coefficients is a problem when calculating the vibrational behavior of rotating
shafts. The theoretical model of this rotor does not include the measured rigid
body eigenmode at 2400 rev/min (figs. 16 and 18). The resonance frequency at
4000 rev/min is not theoretically determinable, too. But for crack detection this
effects are of secondary order. The comparison of the measured (figs. 17 and 19)
and calculated (figs. 16 and 18) once per revolution vibration amplitudes shows a
very good agreement at both mesuring planes S 56 and S 78. We indeed expected a
good agreement, but this exact agreement must be an accidental one.

On the other hand, between the measured and calculated twice per revolution
resonance amplitudes exist a factor of 2. This is to be explaned with the two cracks
in the shaft. The stiffness changes two times per revolution and therefore we have
a greater excitation of twice per revolution vibrations than the crack model delivers.

In figures 20 and 21 the complete frequency spectrum of the rotor with cut is
compared with the spectrum of the cracked rotor. The scales are the same. This re
presentation gives a good survey of the change of the vibrational behavior due to
the crack.

At the rotational speed of 7000 rev/min a resonance vibration appeared with a

frequency of approximately 3500 rev/min for a very short time. Until now we have
no explanation for this effect.

88
 1.0

.5

O ---'r----,---,r--
. +-I
I
I
I
I
I
I
I

5 nc=14940 rev/min I
. I
I
-

I
I
I
I
I
I
B

-1.0 A
..o-o------2 5OT: - ------5- - T: - 0------- 75- 0:- 00- -----
.r
000 2- 5Or: 0-0---
-1 - :rO-0------1
00 00 0
length [mmJ

Figure 9. - Horizontal eigenmodes of undamped system {uncracked rotor}.

'.0

.5
nc = 3773 rev/min

.0 +----7-.------------------L----------------- __
I
I
I
I
I
I
I
I

-
.
5 I
I
I
n c = 14560 rev/min I
I
I I
I I
I I
I I
i I
I I
-1.D A A
2 50. 00

2 50.00
I I
750.00 1
000.00 1
.00
length [ mm ]

Figure 10. - Horizontal eigenmodes of undamped system {cracked rotor}.

89
 1.0

nc = 4048 rev/min
.5

. o +----T----

-.5

- 1.0ri------4Lri ------ - - , r
i - ------ - ' i- - -------- i --------
.00 2 50.00 500.O 0 750.0 0 1000.00 1250.00
length [mm]

Figure 11. - Vertical eigenmodes of undamped system (uncracked rotor).

1.0

.5 n = 3917 rev/min
c

.0+-----4n----TTT--

- 5
. nc = 15310 rev/min

-1.0 ri---- ----4-- i ------i ----ri ------

i --------r_
.00 250.00 500.00 750.00 1000.00 1250.00
le ngth [mm ]

Figure 12. - Vertical eigenmodes of undamped system (cracked rotor).

90
y (1111) Y(11111

10.0
2.0

fI.O
L. 13f1S"" 1.0 L. 13I1S""

2(11111 2(1111)
fl. 0 10.0 1.0 2.0

nC= 4144 rev/min

nc= 4004 rev/min

Y(I1I1)
1 Y (1111)

2.0
2.0

1.0
1.0 L. 13I1S"" L. 'US""

/
2(1111] 2(1111)
1.0 2.0 1.0 2.0

n c= 13609 r ev/mi n nc=14847 rev/min

system
Figure 13. First four complex eigenmodes of complete damped
-

(uncracked rotor).

91
 O
. Or.O 8 O
O O
O 2OOOT
O 4 - OT
O 6 O
0 .
. . 00 . 0
Speed [rev/min]

Figure 14. - Once-per-revolution vibration am plitudes of rotor without and with

cut measured at position 578.

20.0 -
---
v

E
::1.
15.0
"
,I
x
a " I
E I I
0::: I I
I I
10.0 I I
I I
I
I \
I \
I \with cut
I V
" \
5.0 I without cut
I
I
Jv
..
.I
--

.0 2000.0 4000.0

Spee d [rev/min]
Figure 15. - Twice-per-revolution vibration am plitudes of rotor without and with
cut measured at position 578.

92
 50.0 ------
v

40.0
E
::t
x
o
E 30.0
e:::

20.0

10.0

.0 2000.0 4000.0 6000.0 8000.0

Speed [rev/min]
Figure 16. - Measured vibration amplitudes of cracked rotor at position S56.

50.0 -----'
v

40.0 Rmox
E
::t
x
0
E 30.0
a::: w

20.0

10.0

.0 2000.0 4000.0 6000.0

Speed [rev/min1
Figure 17. - Calculated vibration amplitudes of cracked rotor at position S56

( crack depth 45 % of diameter ) .

93
 100.0
v

80.0 Rmax

E
::1.
'--'

x 60.0
0
E
e:::

40.0

.0 2000.0 4000.0 6000.0 8000.0

Speed [rev/mi n J
Figure 18. - Measured vibration amplitudes of cracked rotor at position 578.

100.0
V

80.0 Rmax

E
::1.
60.0

x
0
E
e:::

40.0

20.0

.0 2000.0 4000.0 6000.0 8000.0

Speed [rev/min]
Figure 19. - Calculated vibration amplitudes of cracked rotor at position 578
(crack d epth 45 % of diameter).

94
 200.0
v
E
:1-
150.0
x
0
E
a::
100.0

50.0

Figure 20. - Frequency spectrum of rotor with cut.

E 200.0
v

x
o
E 150.0
a::

100.0

50.0

.0

Figure 21. - Frequency spectrum of cracked rotor.

95
 CONCLUSION

The comparison of calculation and measurement shows in principle that it is

possible to predetermine the vibration amplitudes excited by a crack. But at the
same time this investigation shows the problems in modelling the system. The de
veloped crack model seems to be useful.

For crack supervision the phase should be taken into account, too, because
the amplitude of the sum of the original vibration and the vibration due to crack
can become smaller with crack propagation.

In the future the influence of unbalance is to be investigated. Our calcula

tions have shown, that this influence frequently is small for greater turbine rotors
because the statical deflection is greater than the vibration amplitudes. But in
rotors constructed like this experimental rotor an influence can be expected. Here
the vibration amplitudes at crack position have the same magnitude like the statical
deflection.

REFERENCES

1. Haas, H.: GroBschaden durch Turbinen- oder Generatorlaufer, entstanden im Be

reich bis zur Schleuderdrehzahl. Der Maschinenschaden, 50. Jahrgang, Heft 6,
1977, pp. 195-204.

2. Stodola, A.: Dampf- und Gasturbinen. Springer, Berlin, 5. Auflage, 1922,

pp. 931.

3. Kellenberger, W.: Biegeschwingungen einer unrunden, rotierenden Welle n ho

rizontaler Lage. Ingernieur-Archiv, XXVI, 195B, pp. 302-31B.

4. Tondl, A.: The Effect of Unequal Moments of Intertia of the Sha ft Section on
the Motion and Stability of a Rotor. Some Problems of Rotor Dynamics, 1st ed.,
Chapman & Hall, London.

5. Bishop, R.E.D. and Parkinson, A.G.: Second Order Vibration of Flexible Shafts.
Phil. Trans. of Royal Society of London, vol. 259, no. 1095, 1965.

6. Mayes, I.W. and Davies, W.G.R.: A Method of Calculating the Vibrational Beha
viour of Coupled Rotating Shafts Containing a Transverse Crack. Second Int.
Conference Vibrations in Rotating Machinery, University of Cambridge, 2-4
Sept., 19BO.

7. Inagaki, T.; Kanki, H.; and Shiraki, K.: Transverse Vibrations of a General
Cracked-Rotor Bearing System. ASME Paper No. BI-Det-45, Sept. 19B1.

B. Henry, T.A. and Okah-Avae, B.E.: Vibrations in Cracked Shafts. Conf. on Vibra
tions in Rotating Machinery, University of Cambridge, Sept. 1976, pp. 15-19.

9. Buerhop, H.: Zur Berechnung der Biegesteifigkeit abgesetzter Stabe und Wellen
unter Anwendung von finiten Elementen. VDI-Fortschrittsberichts, Reihe I,
Nr. 36, 1975.

96
10. Ziebarth, H.; Schwertfeder, H. und MUhle, E.-E.: Auswirkung von Querrissen auf
das Schwingungsverhalten von Rotoren. VDI-Berichte 320, 1978, pp. 37-43.

II. Fehlberg, E.: Neue genaue Runge-Kutta-Formeln fUr Differentialgleichungen

zweiter Ordnung bzw. n-ter Ordnung. ZAMM 40, 1960, pp. 449-455.

12. Grabowski, B.: Zur modalen Simulation des instationaren Schwingungsverhaltens

von Turbolaufern. VDI-Fortschrittberichte, Reihe II, Nr. 25, 1976.

13. Grabowski, B.: The Vibrational Behavior of a Turbine Rotor Containing a

Transverse Crack. Trans. ASME, J. Mech. Design., vol. 102, Jan. 1980,
pp. 140-146.

14. Grabowski, B.: Das Schwingungsverhalten eines angerissenen Turbinenlaufers.

VDI-Bericht Nr. 320, 1978, pp. 31-36.

97
 DETERMINATION OF BALL BEARINJ DYNAMIC STIFFNESS*

R. F. Beatty and B. F. Rowan

Rocketdyne Division
Rockwell International
Canoga Park, California 91304
ABSTRACT

The dynamic radial stiffness characteristics of rolling element bearings are

currently determined by analytical methods that have not been experimentally verified.
These bearing data are vital to rotating machinery design integrity because accurate
critical speeds and rotor stability predictions are highly dependent on the bearing
stiffness. A tester was designed capable of controlling the bearing axial preload,
speed, and rotor unbalance. The rotor and support structures were constructed to
permit critical speeds that are predominantly determined by a 57 rom test bearing. A
curve of calculated critical speed versus stiffness was used to determine the actual
bearing stiffness from the empirical data. The results of extensive testing are used
to verify analytical predictions, increase confidence in existing bearing computer
programs, and to serve as a data base for efforts to correct these programs.

INTRODUCTION

The radial stiffness of a ball bearing is one of the key elements in the rotor
dynamic analysis of rotating machinery and is strongly dependent on axial preload,
the constraints of the outer race and shaft speed. Attempts to determine bearing
stiffness in the past by measuring applied loads and deflections have been compli
cated by the large required loads and very small deflections. Also, shaft rotation
is difficult to include in such conventional tests. The method used in this experi
ment avoids these problems by utilizing a tester designed such that the lowest rotor
critical speed is primarily a function of the test bearing stiffness. The overall
purpose of this test program was to advance the technology of measuring ball bearing
dynamic stiffness in support of theoretical rotordynamic analysis and rotating machin
ery design.

DESCRIPTION OF TESTER

The tester (Fig. 1) consists of a simple rigid rotor with control over the var
iables that affect the bearing stiffness. The significant features of the test
rotor are as follows:

1. Axial preload can be accurately controlled by a hydraulic piston. This pre

"
load can be easily varied without disassembly of the tester.

2. Setscrew holes are provided in the large rotor disk for in situ balance cor
rections and access is provided in the supporting structure for easy
corrections.

3. The rotor mass can be changed by substituting a heavy or lightweight spacer

disk to shift the critical speed so that the stiffness can be found at vari
ous speeds. The rotor weight differential is 9.3 kg (20.5 pounds ) between
configurations.

4. The mass distribution of the tester rotor is proportioned so that the cri
__________ tl:ca=l speed is effectively independent of the slave bearing stiffness.

*Work partially sponsored by NASA Marshall Space Flight Center, Contract NAS8-27980.

98
 5. Locations are provided for two pairs of orthogonal displacement transducers
to monitor rotor motion.

6. The test rotor is driven by a variable speed motor.

7. Different test bearing sizes can be accommodated by substituting bearing

sleeves. The selection of a 57 mm test bearing was influenced by the radial
stiffness effect of this bearing on the rotordynamic characteristics of the
Space Shuttle Main Engine (SSME ) High-Pressure Oxidizer Turbopump ( HPOTP ) .
Details of this turbine end bearing on the dynamic performance of this
machine are documented in the literature by Rowan (ref. 1).

Figure 2 shows a schematic arrangement of the test setup, and Fig. 3 is a photo
graph of the actual installation. The rotor and supporting structures were designed
so that their dynamic properties could be accurately predicted by analysis. Pertin
ent design data of the tester is presented in Table I.

ANALYTICAL APPROACH

A finite element model of the rotating assembly was generated, and a parametric
computer analysis was performed to calculate the rotor critical speeds as a function
of the test bearing stiffness to ground. The resulting critical speed plot is shown
in Fig. 4, and the close spacing of curves A, B, and C indicate that the critical
speed is nearly independent of the slave bearing stiffness. The critical speed can
be easily detected by test, and by referring to Fig. 4, the test bearing stiffness
can be determined. The slope of the curve in Fig. 4 indicates that the critical
speed is highly dependent on the test bearing stiffness and should ensure accurate
results. The lowest rotor critical speed mode shape is shown in Fig. 5.

By changing the mass of the rotor, the critical speed can be shifted so that a
curve of test bearing stiffness versus speed can be plotted. Additional tests can
be run with different axial preloads and / or unbalance to investigate these parameters.

TESTING

The general testing procedure consisted of determining the rotor residual un

balance, setting the hydraulic cylinder to maintain a given axial preload, and run
ning a ,slow speed ramp through the critical speed while monitoring the rotor motion.
The radial motion was monitored by two pairs of orthogonal displacement transducers
that produced orbit plots on oscilloscopes. Photographic documentation of the
orbits were taken every 1000 rpm starting at 15,000 rpm while speed was controlled
manually to produce a series of step dwells to define the critical speed.

The rotor was installed with the heavyweight spacer, and the initial run indi
cated that no balance correction was needed because the residual was sufficient to
produce a resonance, without excessively loading the bearings. All tests conducted
are listed in Table II and the bearing stiffness was found by using the Fig. 4 curve,
which relates the critical speed to test bearing stiffness. Table II also contains
the analytically predicted stiffnesses in accordance with the procedures of ref. 2
for comparison.

Typical radial displacements from Test No. 3 are shown in Fig. 6 for 445 Newtons
(100 pounds ) axial preload and an estimated 6.4 gm-in. unbalance. This test indi
cated a critical speed at 20,000 rpm and a resulting test bearing stiffness of 9.1 x

107 N/m (0.52 x 100 lb / in. ). Bearing outer race clearance in the fixture is evident

99
by the abrupt post-critical drop in displacement above 22,000 rpm. This agrees with
the theoretical behavior of a rotor with deadband bearing installations.

It was found during testing that the facility gearbox drain was not large
enough to allow free oil flow above 23,000 rpm. Since increased preload or use of
the lightweight spacer would tend to raise the current critical speed in the 19,000
to 20,000 rpm range, testing was limited to that shown in Table II and further test
ing suspended.

CONCLUSIONS

The tester and very practical testing technique functioned as intended with
the exception of the gearbox lubrication, which is not related to the tester de
sign. The data indicate that the dynamic stiffness of this 57mm ball bearings is
on the order of 8.75 x 107 N/m (0.5 x 106 lb/in.) for 445 to 890 Newtons (100 to
200 pounds) axial preload in the vicinity of 20,000 rpm. The value may differ in
the usual spring preloaded installation since the preload piston in this test may
resist bearing tilting more than a conventional preload spring and this tilt re
sistance may be responsible for some of the difference.

The analytically calculated value with the race free to tilt is approximately
20 percent lower than the measured value for the larger rotating radial load situa
tion. A closer lower stiffness correlation is obtained for the lower dynamic load
ing cases. Stiffness increases for larger loading are typical of angular contact
ball bearings and a nonlinear system. The data do indicate some sensitivity to
the nonlinearity of the clearance between the outer race and the fixture.

Computer program bearing stiffness such as calculated per ref. 2 are adequate
approximations for rotordynamic analyses if &dequate design margin is incorporated.
It is recommended as a guideline that rotor systems be designed so that the oper
ating speed is not within 20 percent of any calculated critical speed.

FURTHER TESTING

To increase scope, the tester is designed so that tests with various clearances
between the support and outer race can be conducted. This clearance is generally re
ferred to as the bearing deadband. The effect of outer race constraint can be in
vestigated by machining the diameter of the test bearing support so that an increased
clearance is produced and installing a preload spring between the hydraulic piston
and outer race. The rotor would then be ramped through its critical speed with simi
lar axial preload and unbalance conditions for comparison with the original clearance.
This nonlinear effect of the deadband produces an apparent shift in the critical speed
and influences the effective bearing stiffness. As of this time, these tests are
still in the planning stages, and no schedule is available. Budget and the rotor
dynamic community interest will be the determining factors.

REFERENCES

1. Rowan, B. F.: "Rotordynamics Analysis of the Space Shuttle Main Engine High
Pressure Oxidizer Pump," presented at the 21st Structures, Structural Dynamics,
and Materials Conference, Seattle, Washington, May 1980.

2. Jones, A. B.: "A General Theory for Elastically Constrained Ball and Radial
Roller Bearing Under Arbitrary Load and Speed Conditions," Journal of Basic
Engineering, June 1960.

100
 TABLE I. - TESTER DESIGN DATA

ROTOR MATERIAL INCO 718 STEEL

TEST BEARING 57mm

SLAVE BEARING 45mm

BEARING SPACING 34.3 em (13.5 INCHES)

DUMMY MASS OUTER DIAMETER 20.32 em (8.0 INCHES)

HOLLOW SHAFT OUTER DIAMETER 10.16 em (4.0 INCHES)

ROTOR WEIGHT:

HEAVY DISK 33.3 kg (73.5 POUNDS)

LIGHT DISK 24.0 kg (53.0 POUNDS)

MAXIMUM SPEED 35,000 RPM

TABLE II. - 57mm BEARING DYNAMIC STIFFNESS

MEASURED
BEARING ANALYTICAL BEARING
CRITICAL STIFFNESS' STIFFNESS"
TEST SPEED, N/m X 10-8 WITH TILT AT 20,000 RPM %
NO. TEST DESCRIPTION RPM (LB/IN. X 10-6) N/m X 10-8 (LB/IN. X 10-6) ERROR

1 445 N (100 LB) PRELOAD 18,000 0.70 0.61 -12.5

ESTIMATED UNBALANCE = 2.8 GMIN. (0.40) (0. 35)

2 890 N (200 LB) PRELOAD 19,000 TO 0.857 0.945 +10.2

ESTIMATED UNBALANCE = 2.8 GM-IN. 20,000 (0.49) (0.54)

3 445 N (100 LB) PRELOAD 20,000 0.91 0.70 -23.0

ESTIMATED UNBALANCE =
6.4 GM-IN. (0.52) (0.40)

'SLAVE BEARING STIFFNESS CAN PRODUCE 5% ERROR. AVERAGE CURVE B IS USED (FIG. 4)
" REF. 2, CALCULATED BY FRANCIS LEE, ROCKETDYNE.

101
DISPLACEMENT TRANSDUCERS NO.1 AND NO.3 DISPLACEMENT TRANSDUCERS NO.2 AND NO.4
90 APART
90 APART

Figure 1. - Bearing tester.

(\
HYDRAULIC PISTON PRELOAD CYLINDER

TEST BEARING

. DISPLACEMENT TRANSDUCERS

SLAVE BEARING

o QUILL SHAFT

F igure 2. - Test setup schematic.

102
 Figure 3. - Test setup.

50
CURVE SLAVE BEARING STIFFNESS

40 X 10-6 LB/IN X lO-S N/M

x

A 0.50 0.S7
a: B 0.75 1.31
0' 30 C 1.00 1.75 C
w
w B
OBSERVED CRITICAL
!JI A
..J SPEED FROM TEST 3

u
E 20 -- ------_ \ _--- --

a:
u
o
w

..J
::>
U
..J

u
10
3 4 5 6 7 8
5
10 LB/IN
2 9 106 LB/IN 2

I I I I I I
7
10 N/M 1.75 3.5 7 .0 lOS N/M 1.75 3.5

TEST BEARING STIFFNESS

Figure 4. -
Heavyweight spacer critical speed plot.

103
 1.0 r<::-'I:::::r:----'

o
o FINITE ELEMENT MODEL JOINT LOCATION

-1.0 -....-....-...........
. , ...-...-...-......
""",- ...
o 25.4 em 50.8 em
(10.0 IN.) (20.0 IN.)

AXIAL, COORDINATE

Figure 5. - Test rotor lowest mode shape relative displacement versus

axial coordinate.

o DISPLACEMENT TRANSDUCER 1 PRELOAD = 445N (100 POUNDS)

.t.DISPLACEMENT TRANSDUCER 3 UNBALANCE = 6.4 GM-IN.
ODISPLACEMENT TRANSDUCER 2 VERTICAL FULL SCALE= 2.54 x 1 em
10
DISPLACEMENT TRANSDUCER 4 (10 X 10 IN.)

TEST BEARING SLAVE END

END OF ROTOR OF ROTOR

...
Z
w
:!
w-
(,)
e(e(
....IW
8511..I
-

C
....Ie(
e(w
-II..
c
e(
cr:

24

SPEED, RPM X 10-3 SPEED, RPM X 10-3

Fi gure 6. - Test no. 3 displacements vs. speed.

104
 ROfOR VIBRATION CAUSED BY EXTERNAL EXCITATION AND RUB

(Bruni Matsushita, Michiyuki Takagi, Ratsuaki Kikuchi, and Makio Raga

Mechanical Engineering Research Laboratory
Hitachi, Ltd.
502, Randatsu-Machi Tsuchiura-shi,
Ibaraki-ken 300 Japan

SUMMARY

For turbomachinery with low natural frequencies, considerations have been

recently required for rotor vibrations caused by external forces except unbalance
one, such as foundation motion, seismic wave, rub and so forth. Such a forced
vibration is investigated analytically and experimentally in the present paper.
Vibrations in a rotor-bearing system under a harmonic excitation are analyzed by
the modal techique in the case of a linear system including gyroscopic effect. For a
nonlinear system a new and powerful quasi-modal technique is developed and applied to
the vibration caused by rub.

From the above analyses the following results are derived.

1. In a high speed rotor, two resonances of a forward whirl motion and backward one
are induced under a directional external excitation, because of large gyroscopic
effect. Although rub of rotor usually causes instability, it is possible to
stabilize the rotor vibration by select ing a rubbing stopper suitable to each
resonance severity.
2. In a low speed rotor such as a vertical pump with small gyroscopic effect, a
response is a directional vibration within limits of no rubbing. However, when a
rotor begins to rub, a directional respose changes to a stable forward whirl
motion due to the nonlinearity in the water bearing.
,

These analytical results characterized by gyroscopic effect and bearing

nonlinearity are confirmed by excitation tests of a high speed spin rotor and a p ump
model rotor, respectively.

1. INTRODUCTION

Many studies deal with vibration responses for some kinds of excitation in
structure dynamics. However, a response of rotor under external excitation shows
different features of motion from that of structure, i.e. forward whirl motion and
backward whirl motion at each resonance. Some papers (ref. 1, 2) describe such a
response of a rigid rotor.
In this paper, a general description of a forced vibration in linear rotor
bearing system is given and a rub vibration induced by contact between a rotor and a
stator is also included in the discussion on nonlinear rotor dynamics.
Two kinds of rotors are selected for the study of rub vibration. The one is a
rotor with high speed rotation and large gyroscopic effect. The stopper effect for
the suppression of a rotor v i br at i on due to rub is discussed. The other is a pump
model rotor with low speed rotation and water lubricated bearings. Nonlinearity of a
water bearing and rub vibration of a journal are considered.

105
 SYMBOLS
it It
B , B.
1. , modal quantum in state equation
Cg gyroscopic matrix
C modal gyro
l
Cf, Cb complex type of damping matrix
= x,y) damping matrix
F external force
i imaginary unit !=1
Id transverse moment of inertia of a disk
Ip polar moment of inertia of a disk
Kf, K b complex type of stiffness matrix
= x,y) stiffness matrix
K stiffness of stopper with a clearance
it
K modal stiffness
m ; mass of a disk
M , Mit., mass matrix modal mass
S, Si normal coordinate
t time
Qz = Q x + i Q y nonlinear restoring force
z = X + iY rotor absolute displacement

o = Xo + i o foundation absolute displacement

z = Z - &0 rotor relative displacement
8" clearance
S = - 0(11).1 modal damping ratio CICc
A , A./( complex eigenvalue
friction factor
f';
v excitation frequency
, i ; eigen vector
W I wf tV/? natural frequency in rotation
tUn natural frequency in no rotation
Sl. rotational speed

2. E QUATION OF MOTION AND MODE SEPARATION

2.1 Equation of Motion in Rotor-Casing System

The equation of motion on a coordinate system fixed in the space is derived for
entire system including a rotor, bearings and casing, as shown in Fig.1. The system
is discretized by a beam model along longitudinal axis with E -direction.
Representing the absolute displacements at each location of a rotor with X and Y and
that of a casing with x and y , the equation of motion is given as follows

M -9.C,Y
X. + kx)( (X-Xo)+kx'j(Y-:!o)+Cxx(X-Xo)+CXj (-(-J:) = Qx
(1)
M Y S2.X
+ + K'jx. (X -:to) I<J'J ( '( - Jo) + CJX ( X-io) T C'j'j (i-jo)
+ = Qy

where Qx and Qy are nonlinear restoring forces.

A complex numbered type of displacement is introduced here for the sake of
convenience of description.

Z = X + iY (2)
.-
Since the acceleration Zo (t) of the foundation of the casing is usually measured, the
equation (1) is rearranged to obtain the acceleration excitation of foundation.
Applying relative displacement z=Z-z o, where the rotor motion is measured with
respect to the casing motion, the following equation of motion is obtained from Eq.(1).

106
This formula is fundamental for the study of vibration under seismic excitation in
rotor dynamics.
In the formula, the second term with the coefficient of rotational speed shows
gyroscopic effect which exists only in a rotor system. The matrices M and Cg consist
of inertias of each disk on a shaft in the following forms.

M oL,id.!{Jnd-1 ( - - ... PI" > Iet .... )

(4 )
o PLidrna--J ( - ... , 0 ) 1f ... )
'

= [do, 80, =0],

.. , ,.

Therefore, if the casing moves in a parallel way with acceleration ()

excitation force, stated on the righthand side of Eq.(3), reduces to only inertia
force M Zoof mass as usual. On the other hand, if the casing is rocking in a non
parallel way with i=[ it" Q ], an additional force -Qc.?8o adds up the excitation
force. That is a gyroscopic moment which is p ropor tional to a polar moment of
iner tia of disks and rotational speed.

2.2 Response at Each Mode

Table 1 is utilized to derive the modal response at each mode based upon
orthogonality of eigensolution. There are two columns I and II in Tab.1. One of
them concerns with a rotor system suspended vertically and the other one with a
horizontally suspended rotor system. The former one indicates a free motion with a
circular whirl and the latter one indicates a motion with an elliptical whirl. This
classification is due to symmetry or asymmetry in bearing dynamic properties.
For each case, it is assumed that eigensolution of free equation of motion
indicated in the line No.3 of table 1 is given by the formula as shown in the line
No.4. The corresponding eigenvalue problem is obtained, and orthogonality condition
is guaranteed in the forms given by the line No.8.
Therefore, by using these eigensolutions o!' tree yibratiQn, it is pos s ible to
separate a general response into each mode response on normal coordinate system.
From a physical system with symmetrical bearings of I in Tab.1 to a normal coordinate
system, the modal transformation is defined by the relationship as follows.

(5)

The state equation with normal coordinate resulted by post-transformation becomes

Si .As, Ai 1>: p
- :::: _
(6)
(3
<..
where
!3/ -),.f c/>/ /'1 Ti
;: + cp/ K CPt
The normal transformation for a system with asymmetrical bearing of II in tab.l is
also derived by the same manner.

3. HARMONIC FOUNDATION EXCITATION

3.1 Harmonic Excitation

The external force in the equation of motion Eq.(3) becomes

where the harmonic acceleration in a certain direction e.g., fixed at X-axis

107
 (8)

is substituted into Eq.(7). Then, the external force is defined as following

exponential expression.

F= - (M + ) _ e L'J->t _ (M _ . Cj ) e-ivt (9)

Movements of the foundation in parallel and rocking motions are generally included in
the amplitude a of an acceleration. As usual, in the case of no rocking motion the
external force is simplified into the following formula.

M (l)t M a. (9' )
e-(pC
F = e
, 2 2
- -
As the one directional force acts upon the rotor, it is equivalent to two whirling
forces expressed by exponential form, with forward component i and backward component
- i)) .
Now, suppose that an eigenvalue A, .. -O<+i.w (tV >0) with a rotor in rotation is
obtained, a resonance occurs when excitation frequency Y comes close to a natural
frequencyw i.e., ))W . On the other hand, the conjugate value ):.=-cX.-iW does not

appears at the condition of - -M

become an eigenvalue in rotor system, because of gyroscopic effect. So no resonance
Therefore, it is clear that only one of the two
whirl forces contributes to a resonance.
Each whirl force, divided from one directional harmonic force, has half of the
magnitude of one directional force. Then the rotor response on a resonance, while in
rotation, is reduced to about half of the amplitude, compared with the one with no
rotation.

3.2 Features of Response

Some features appearing in the response are summarized for a vertically suspended

vector is obtained by substituting Eq. (9)' into Eq. (6).

rotor here. A modal response corresponding to an eigenvalue and also to an eigen

A cl Lyt ].v a - vt
- e - e
Cv -A.-
----
2.
+ -----
-(,'v ]V 2 J (10)

[1] Forward Whirl Resonance; As the condition yWf with the relationship
between eigenvalue). =-.,(+Lwj- and exciting frequency.v , a forward whirl resonance
becomes remarkably large as described in the first part of Eq.(10).
2 ei (JC -2'70)
S
({) -L cj:>tM
2 kif AS
(11)
I + W//uJ;t
The phase relationship between rotor position and exponential exciting force is
shown in Fig.3, for three instances of pre, on and post resonance when the excitation
frequency is swept. The phase lag of rotor behind the force is 270 deg. when just on
a resonance in the X-direction. And it is less than 270 deg. at pre-resonance and
more than 270 deg. at post-resonance.

and excitation frequency V approaches the range of -1).", - t'Jb , the rotor response
[2] Backward whirl Resonance; When the relationship between eigenvalue).=-o(-iwb

becomes a backward resonance, as described in the second part of Eq.(10).

-L cfotM ;;2 e-i(vt-2700)

sLt) (12)
2. K;If I + w{/w.,,2.. 0<-)'
The phase lag of rotor behind the force is 270 deg. when just on a resonance. And it
is changing from less than 270 deg. to more than 270 deg. through the resonance as
shown in Fig.4.

108
[3J Whirl Motion; The rotor has two eigenvalues which correspond to a mode
shape. These values are different from conjugate relation and are independent from
each other. Here the forward natural frequency is higher than the backward one.
Therefore, when rotating the rotor at a high speed, the response whirl motion is
changing from left to right as shown in Fig.5, corresponding with increase of
excitation frequency. At the beginning the backward resonance occurs with a circular
orbit, and later on through ell iptical orbits the forward resonance appears again
with a circular whirl orbit. On the boundary line, between forward direction and
backward direction, an elliptical whirl orbit becomes a straight line thus whirl
direction can be interchanged.

[4] Low Gyroscopic Effect; In the case of low gyroscopic effect such as low
rotational speed or small disks, forward eigenvalue and backward one corresponding to
an eigen mode are not independent of each other, and they approach to a conjugate
relation. The typical state such as low gyroscopic effect leads to dynamics of
nonrotational rotor. Eigenvalues maintain conjugate form, for example,
X-
K =' )v fi-/(rl = .A-
Resonance response is described by the following form

it- evt
-- -- e +

iyA-
such that rotor, vibrating only in X-direction, coincides with the direction of the
acting force without a whirl motion.

3.3 Harmonic Respose Curve

Figure 8 shows the response curve under harmonic excitation. In the case of
=0 i.e., no rotation, a resonance peak appears. On the other hand, at high speed
rotation two resonance peaks appear with backward whirl and forward whirl. The
separation of a peak into two peaks is due to gyroscopic effect or rotor rotation.
This figure is described by nondimensional frequency and amplitude, which are normal
ized by the values of the resonance frequency and amplitude at no rotation,
respectively. The peak at the forward resonance is less than the half of the peak
at the resonance in no rotation, because of tJj Iw.." i in Eq. (11). On the other hand,
the peak at the backward is more than the half of that, because OfWbW! 1 in Eq.
(12). Therefore it is clear that these two peak amplitudes reduce to about half of
the amplitude at no rotation and the backward amplitude is slightly higher than
the forward one.

4. SEISMIC EXCITATION

Instead of equation of motion in physical coordinate system, the state equation

in normal coordinate system is used for analysis of time history response caused by
seismic excitation. In Eq.(6), an arbitrary external force

(14)

is defined by insertion of the acceleration lo measured in the field. Substitution of

Eq.(14) into Eq.(6) and integration of the resulting equation gives time history
response of rotor in normal coordinate system. In order to return the normal response
to the response in physical system the following relationship is used:

( 15)

109
 5. S EI SMIC EXCITATION WITH NONLINEAR SYSTEM

Here we discuss a rotor vibration induced by seismic excitation which becomes

larger than the clearance between a rotor and a stopper. In such situation the rotor
begins to contact the stopper fixed on the casing.
In this rub vibration, the response on normal coordinate system is given by
Eq.(6) and the external force is defined as follows.

F = - M i 0 - (52 0 20 + Q i! (16)

The force Qz is a nonlinear restoring force generated by the contact of a rotor with
a stopper. The two types of configuration of a cylindrical stopper are mentioned
here.
(1) Type I contact between an outer surface of a rotor and an inner surface of a
stopper i. e., a rotor in a stopper
(2) Type I! contact between an inner surface of a rotor and an outer surface of a
stopper i.e., a stopper in a rotor
The types I and II here are called an outer contact and an inner contact with respect
to the rotor.
Representing the stiffness of, a stopper the clearance between a rotor and a
stopper and the friction factor with K, E and r , respectively, the nonlinear restoring
force is defined as follows.

for outer contact type (17)

Q =Fr iF& +

- for inner contact type

where
Fr = -I< ( 121 - J" ) :a/I E I
and
F (}:::: r F l'" sign ( angular velocity of contact face on rotor)
In the simulation of a nonlinear vibration caused by rub, the second term in
Eq.(17) indicates an unstable friction force which induces a whirl motion. Therefore,
to suppress the rotor vibration by a cylindrical stopper, the outer type of a stopper
is more effective in a rotor with higher forward resonance severity. On the other
hand, the inner type of a stopper is more effective with higher backward resonance
severity.
For the response history analysis in such a nonlinear system a quasi-modal
technique is effective, and a hybrid integration method based upon it is developed,
as described in A ppendix.

6. EXCITATION TEST OF HIGH SPEED ROTOR

6.1 Harmonic Excitation

The features of a rotor response excited by a harmonic wave, stated in the
previous chapters, are reconfirmed by an experiment in which the rotor shown in Fig.7
is used. The rotor rotates in very high speed, and it has large gyroscopic effect.
The rotor is suspended by bearings at both sides. The vibration mode with a low
natural frequency is the conical mode in which the right side of the rotor vibrates
with the nodal point in the left side.
The vibration response waveform, the response curve and the Lissajous figure
of
rotor motion are presented in Fig.8, including a harmonic waveform with
the excitation
frequencyV. When the rotor does not rotate, the rotor vibration response is one
, directional and the resonance frequency agr e with the natural frequency
tVn in a
structure i.e., V =W = 4.95 Hz.
n

110
 Under a rotation of the rotor, two resonance peaks in the rotor response appear
in the backward and forward resonances at the excitation frequency V = O.6 6cVn and
y 1.21 Wn, respectively.
= The peak amplitude severity Q at the backward resonance
becomes about half of the peak amplitude severity Qn = 156 f m/gal at the resonance
of the rotor with no rotation. The other one at the forward resonance is about one
fifth of Qn. The calculated values of the resonance frequency and peak amplitude
severity, as shown in Fig.5, well coincide w ith these experimental values.

6.2 Rub Vibration

An aseismic stopper with a fine gap is installed in the rotor system with large
gyroscopic effect, as schematically shown in Fig.7. This stopper is not a usual
circular type, but one directional type with a clearance of x-direction as shown in
(i) of Fig.6.
Figure 9 shows vibration history response and its Lissajous figure of the rotor
subjected to a seismic wave. The magnitude of excitation acceleration increases in
turn by (A), (B) and (C). In the case of (A), excitation acceleration is so small
that the rotor moves in the clearance with no rub and nearly circular whirl orbit.
In the case of (B) and (C) rubs occur. C omparing the case (C) with the case (B) a
larger acceleration causes a stronger rub between the rotor and the stopper, close to
an impact.
In the rotor system with large gyroscopic effect, the vibration of a rotor is
always generated in a circular whirl motion by an external force in one direction.
Then, vibration is satisfactorily suppressed by even one directional stopper at an
arbitrary phase location, as demonstrated in this numerical simulation. One
directional stopper is rather preferable than a circular stopper, because the duration
of rub in the former is less than one in the latter and the corresponding friction
force which induces unstable vibration becomes less.
The effect of the stopper upon vibration suppression is indicated in Fig.l0.
The horizontal axis shows the maximum value of the acceleration in the seismic wave,
normalized by the value in which the rotor begins to rub with the stopper. The
perpendicular axis gives the value of the displacement in the response, normalized
by the clearance $. The stable suppression of the rotor vibration is seen in the
range of about five times of the acceleration at the beginning of rub in Fig.' 10.
This numerical result proves a good agreement with the corresponding experimental
result.

7. EXCITATION TEST OF PUMP MODEL ROTOR

7.1 Outline of Experiment

Figure 11 shows the outline of the experimental equipment which is simplified
into a vertically suspended pump model rotor for atomic power plant. The experimertal
rotor has two disks and two bearings. A self-centering ball bearing is used at
the upper part and water is filled in the cylindrical water bearing located at the
lower part. Two weights are added to the lower part of the rotor. An aseismic
support is located at the lowest part of the casing in order to prevent large
vibration of the casing.
The pump model rotor is vertically located on a shaking table which can stand
the load of 294 kN (30 tons). The excitation wave is sinusoidal one in this test.
In this measurement b;stem, the usual way is employed, except for a load cell with
strain gages to measure the dynamic force in the lower water bearing.

7.2 Natural Frequency

The experimental rotor is discretized by beam element and a calculation model is
obtained as shown in Fig.11. Natural frequencies and corresponding mode shapes are
represented in Tab.2 and Fig.12,respectively.

III
 From the calculationresults considering virtual mass effect (Ref.3) of water
and assuming water bearing stiffness of 9.8 MN/m (103 kgf/mm) natural frequencies are
13.6 Hz and 18.0 Hz. With the stiffness of 39.2 kN/m (4.0 k gf/mm) , they are 14.9
HZi 6.4 Hz and 27.6 Hz. The former stiffness is picked up to simulate the contact
of the journal with the stator of bearing under the condition of large journal
vibration, and the latter one is equivalent to the stiffness of water bearing under
the condition of small journal vibration. In the latter, the rotor vibrates
remarkably in the first and second mode and the casing in the first mode.

The comparison of calculation values of natural frequency with experimental

ones is presented in Tab.2. The first natural frequency of 6.4 Hz in calculation
agrees well with the experimental value of 7.0Hz. Concerning the casing mode, a
good agreement is also obtained by comparison between the calculation value of
13.6-14.9 Hz and the experimental value of 14.0 Hz.
With respect to the second natural frequency of rotor mode, the experimental
value of 18 .... 20 Hz is negligibly d ifferent from the calculation value of 27.6 Hz
under the assumption of no contact within small journal vibration. However, in the
case of large stiffness i.e., assuming a contact in the bearing, this experimental
value becomes close to the value obtained by the calculation. From these compar-'
isons, it is estimated ttat the restoring rorce with a hard type of nonlinearity
due to a contact is induced in a water lubricated bearing, and the large journal
vibration increases the bearing stiffness.

The first mode resonance does not appear remarkably in the excitation test
because of very high modal damping ratio.
As stated above, the good agreement of natural frequencies between the
calculation and the experiment suggests the reasonable modeling of the pump rotor
system, including dynamics in the water bearing.

7.3 Rotation Test

A rotation test is done before an excitation test. In this test the casing is
tightly fixed by the aseismic support with no clearance. The vibration of the rotor
itself is seen in the amplitude response curve in Fig.13. The first unbalance
resonance appears at the rotational speed of 7 8 rps.
The second resonance varies greatly with the unbalance. In the case of a small
unbalance, the second resonance peaks occur at the rotational speed of 11 '" 19 rps,
It changes to 20 rps when a large unbalance is applied to the rotor. It is noticed
that the peak amplitude becomes very sharp in this case.
The large unbalance induces a large amplitude of vibration such that the rotor
moves in a large whirl orbit in the clearance of the bearing, and the rub often
occurs. Consequently the increase of the stiffness and the decrease of the damping
effect at the bearing are induced by a large whir l motion and they lead to a sharper
amplitude and a higher resonance frequency. In fact the scratches of rub can
be seen when opening the bearing.
The comparison between amplitude responses in the rotation test and in the
harmonic excitation test with no rotation is described in Fig.13 and 14. Both
response curves are very different from each other. In the excitation test, ofrly one
high peak appears at the excitation frequency of about 18 Hz. Its response curve is
similar to the unbalance response curve with large unbalance.
Therefore, it can be said that this excitation acceleration of 0.1 G seems
to be small in value, but it has a large influence on the rotor vibration. From the
fact, an actual pump rotor with low load capacity in the water bearing is very
sensitive to an external excitation. This fact indicates the necessity of aseimic
design for a pump rotor suspended vertically.

112
 7.4 Harmonic Excitation Test
Influence of some parameters on a rotor response is examined by sweeping of a
harmonic excitation frequency.
(1) Influence of Clearance in Water Bearing
The displacement of the rotor vibration is shown in Fig.14 by changing the
bearing clearance of 0.127 mm and 0.381 mm. Here, the casing is fixed by the
aseismic stopper with no clearance. In case of 0.127 mm (normal specification),
the resonance occurs at the excitation frequency of 18.5 Hz and its peak
amplitude reaches up to the value of 760fm. With the clearance of 0.381 mm,
the corresponding peak amplitude is 580m. As shown in the response curve with
the clearance of 0.381 mm, larger clearance results in high damping effect which
reduces the peak amplitude, and it produces a curve which shows true nonlinear
phenomena due to a hard spring.

(2) Influence of Rotor Rotation with Fixed Casing

A harmonic excitation test is done at each rotational speed with the casing
being fixed. The rotor response and dynamic bearing load, including the
responses in X and Y direction, are summarized in Figs.15.
When the rotor is rotating and is subjected to harmonic excitation,
Y-direction vibration appears in addition to X-directional vibration. This
result indicates that a rotor response is a whirl motion, in spite of the
extremely small gyroscopic effect in the pump rotor against the fact as mentioned
in paragraph 3.2. It is seen that the hydraulic force in a fine clearance and
the rub force due to larger journal vibration cause a whirling agitation for a
rotor.
From these graphs, it is clear that an increase in rotational speed
decreases the response peak amplitude and the dynamic bearing load in X-direction.
However, the peak amplitude and the bearing load increase in Y-direction. When
a rotor is in rotation, the sum of vibration responses in both directions
becomes in balance with the power of external force. Then, it is clear that a
rotor with no rotation is most sensitive to an external force and the rotation
reduces such sensitivity.
Although two peaks are seen in a response curve just like influence
of gyroscopic effect, it is not clear why the separation of a peak appears in
spite of the rotor with an extremely small gyroscopic effect.

7.5 Whirl Orbit due to Rub

The response waveforms and whirl orbits of the journal motion at a resonance in
the sweeping test are given in Fig.16. The one directional orbit at the no rotation
changes into a circular whirl orbit with rotation due to rub. Therefore, the waveform
of the bearing load at the rotation contains the ordinary harmonic component plus
the spiky peak component due to rub.

7. 6 Simulation of Rub Vibration

A dynamic model of the water bearing is assumed by a nonlinear relationship
between the journal of displacement and the restoring force, as described in Fig. 17.
The non linear stiffness K is extremely greater than the linear stiffness Ko,
like a hard type of spring.
The numerical result given in Fig.18 simulates a resonance in the harmonic
sweeping test. The journal vibration is suppressed by the bearing clearance, but the
Y-directional vibration occurs simultaneously near the resonance. The waveform of

the bearing load is nearly harmonic. High spiky peak is superposed on the harmonic
one at the instance of rub, like projections on the waveform of bearing load.

113
 The whirl orbits obtained in the calculation of history response are shown in
Fig. 19. At no rotation of 0 rps, one directional vibration is generated, coinciding
with the excitation direction. A whirl motion is induced by rub at the rotation of
24 rps.
Comparing Fig.19 with Fig.16, the calculated whirl orbit at no rotation agrees
with the experimental one. With respect to results at 24 rps, they are different
from each other. The experimental whirl orbit is nearly circular and the
calculation one closes to be rectangular. The disagreement may be attributed
to the fact that a whirling force, generated in a fine gap when the journal rubs
with the stator in the water bearing, and an unbalapce force which actually exists
are neglected in the calculation.
The comparison of the bearing load between the experiment and calculation is
given in Fig.20. The horizontal axis shows the value assumed by a nonlinear stiffness
K. Generally speak ing, good agreement is seen with respect to the harmonic component
and the spiky peak component in the bearing load waveform.

8. CONCLUSIONS

Conclusions summarized in this study are as follows.

(1) The modal technique is acceptable for the vibration analysis in a linear system
with rotor, casing and bearings under an external excitation, and the division
of the vibration response at harmonic excitation into each component of
eigen modes is achieved. On the other hand, for the nonlinear vibration
analysis, a new technique called quasi-modal is developed, and a hybrid
integration method based on it is presented. The effectiveness of the new
method is proved by numerical simulations of vibration due to such rub and
nonlinearity in a water bearing.
(2 ) The peak amplitudes at a backward resonance and a forward resonance under a
harmonic excitation when rotating a rotor in high speed are reduced to about
half of the peak amplitude at no rotation. It is also proved that the
backward resonance severity is slightly greater than the forward one. These
general features of rotor response at a harmonic excitation are made clear in
relation to gyroscopic effect.
The, selection of a sto pper for the s uppression of rotor vibration i.e., inner
or outer type, depends upon which response severity is greater between a
backward resonance and a forward one. It is verified by calculation and
experiment that One directional stopper is rather stable and effective
for the
suppressi<:>J:l_() f rotor vibration in the system with large gyroscopic
effect, .
,
compared with a cylindrical type of stopper
as usually used.
(4) Very high response sensl tivi ty of vertically suspended rotor
such as a pump to
external excitation and rub becomes experimentally clear,
incl uding the
dependence on bearing clearance and rotational speed.
(5 ) In spite of very small gyroscopic effect in a pump rotor,
the rotol'r'gsponse
indicates a whirl motion due to rub and nonlinearity in a water
bearing. The
response of bearing dynamic load contains a ordinary harmonic
component and a
spiky component in waveforms. It is demonstrated that these experimental values
of bearing load can be estimated by the calculation assuming
a dynamic model of
a water bearing with clearance.

REFERENCES

(1) Kawamoto, I and Urushidani, H ; Trans. vol.45-400 (1979-12), P 1451

JSME,
(2) Kanai, S. and Tamura, A. ; Preprint, No 784-5 (1978-3), P 77
JSME
(3) Fritz, R.J. ; Trans. ASME Sere B, 94-2 (1972-2), P 167

114
 TABLE 1. - DAMPED EIGENVALUE ANALYSIS OF ROTOR-BEARING SYSTEM

No. ITEM I II

Circular Whirl Orbit with Symmetric Support Eliptical Whirl Orbit with Asymmetric Support
1 SYSTEM
Vertically Suspended Rotor Horizontally Suspended Rotor

DYNAMIC PROPERTIES k=k.. kxy=-k,. --

2
OFBEARtNG cxx=Cyy cxy=-Cyx

MZ +iJlCo1 +(Cd -iCc)z +(Kd -iKe)z =F M:t +i!1Cgt + C,z + C.z+ Krz+ KbZ =F

3 EQUATION OF MOTION K,=Kd- iKe =

[ k ;kyy j + i [ k,.;k" j

K.=
[ k ;kw j +i [ k,.;k"]

where
C,-C d - iCc =[ 2 w] + i [
c c c,.;C,, ]
C. =
[ c"2cyy j + i [ c,.;C" j
.\ EIGEN SOLUTION z = <I> e Al z - <l>t e A l + ifisel, 1<1>,1>1.I

r'.j AA]
= -M
[ Kd-iKe j 0 0 0 0
]
5
EIGENVALUE PROBLEM B
o
(/I= [ :j B= [ B2
] B.= [ -M
B. K,
] B2 = [0 0 K. (/1= ,
1>.
AB(/I=A(/I i C.+G.d-iCe Kd-iKe 1
A= [ /l j A .= [ i : C' ,] _ C. K.
cf. ABt.=A'.
Kd-,Ke 0
A= [ A2A. A2- [
K. 0

6 MATRIX B,A symmetric --

7 EIGENVALUE A=a+iq Damped Frequency=q Modal Damping = -a/ (a2+q2

=O {;-0
8
ORTHOGONALITY
I
(/)'B(/)
{ (/ItA (/II
for i *j
IJI"IB (/Ij (/Ij
{;
-0 for i*'j
CONDITION 1 *,0
{ : for i=j 0 IJI"IA 0 for i=j

B'(s-[A)s) - [A)Mod(I)'F B'(s-[l)s) =[l)(ModN:J-F + Mod(</I /,)11

9 FORCED VIBRATION where Z-XISI for i-1-2n where Z=X('iSI+ii.;"Ki) fori,-2n

=Mod( ).S + Mod().

-Mod(),s ,

TABLE 2. - COMPARISON OF NATURAL FREQUENCIES

Vibration Mode

CA5INb ROTOR

l-ST 1-5T 2-ND

Item Comment
Natural Frequency, Hz

EXPERIMENT

Impulse response 14.0 Hi t casing

u n balance r espons e 14.0 7 17-19

Harmonic excitation 14.0 No peak 18-20 Rotation N = 20 rps

CALCULATION

Water bearing (linear) 14.9 6.4 27.6 Bearing stiffness

= 4.0 kgffmm
= 39 kNfm (no rub)

water bearing (nonlinear) 13.6 18.0 Bearin stiffness

= 10 Kgf fmm
= 9.8 Mfm (with rub)

115
 Q

x Xo
D
CASING

Figure 1. - Rotor-bearing-casing system.

Figure 2. - Coordinate system.

Zo

L+---t--t-'l- 0-
X

1 PRE-RESONANCE

2 ON-RESONANCE

BACKWARD 3 POST-RESONANCE FORWARD

Figure 3. - Phase relationship between rotor and excitation acceleration.

2
BACKWARD FORWARD

Figure 4. - Whirl orbit.

116
 O ------
w ------
y' ......
----

(f) ", Yo
- 1 80 t---___::... _-_ (i) ROTOR
I
a..
- 360 !::::====!:=====S===:d

1.0

BACKWARD Fr

Q=O

w X OUTER CONTACT FORM

o
:::> Yo
r 0.5
1--+--- FORWARD (ii) ROTOR
..J
a..

Xo
Fr

O =-------J
O. 5 1. 5
INNER CONTACT FORM
EXCITING FREQUENCY

Fi gure 5. - Response curve in Fi gure 6. - Stopper with clearance.

harmonic excitation.

ROTOR A

CASING

Figure 7. - A high speed rotor.

117
 o rpm

NO ROTATION
1.41wn ...... X-DIRECTION

RATED SPEED
___ X-DIRECTION
0----9 Y-DIRECTION

1.0

U)

t=l

.....

H
..... H
00 0.5

U)
z
o

U)

o
0.5 1.0 1.5

EXCITATION FREQUENCY

Figure 8. -
Response of high speed rotor in harmonic excitation.
 .y

SEISMIC WAVE

0.3&
( A) 0 A1l/:l.1lVmIltYMH\lH;AI-k'roY\.nHllltfl\!

1
( B) 0 1-A1+A'AV\tII-\IttWII-IlH\!ffiQI\\lHIi

( c )

VIBRATION
LISSAJOUS

Figure 9. - History response analysis (one directional stopper).

1.0 L- 1--2:tO::::t=:=]
__

<-.)
-.::c
EXPER IMENT o
a:
(/) CAlOJLAnON -

CSO.5--+--------- - r- - - T---
m ali ze d
>< CLEARANCE 1.0 mm
ACCELERATION .0.043 G

1. 2 3 5
MAX. ACCELERATION
Figure 10. - Suppression of rotor vibration by stopper.

119
 p:j
MOTOR
o 14.9 Hz
E-I
o
::g

l-st
p:; CASING MODE
o
E-I
o
p:; 6.4 Hz 1B.0 Hz

BEARING JJ:; C;l

LOAD C;l 0 i2;
:z; E-I H
CELL H 0 U)
U) JJ:;
8x,8y c:t:
c:t: 0
0

'1
\ BEARING

A SEISMIC STOPPER l-st 2-nd

ROTOR MODE
Figure 11. - A pump model rotor
and its beam model.
Figure 12. - Free eigen modes.

X -DIRECTION
s Boo -------.
::1..
X-DIRECTION

600 ------JrIr---, +>

600

UNBALANCE I (])
+> S

400
- SMALL I (])
()

(])
---- LARGE , til 400
r-l
til
()
I P-<
r-l I CJJ
-.-I
P-<
I I=l
' CJJ
peak 2-nd peak 200
- .-I I H
I=l J? o )(
+> C\I
H(l:: o
o p:j
+> o o
o 5 10 15 20 25 5 10 15 20
p:j 25
EXCITATION FREQUENCY Hz EXCITATION FREQUENCY Hz

Figure 13. - Influence of unbalance. Figure 14. - Influence of clearance '

in water bearing.

120
 800 __------------------------ .
s 800
- NO ROTATION
:i.. - NO ROTATION

..... . - 14 rps

:I
_.--- 14 rps
p 600
+>
rps
QJ
S

"I
Acceleration QJ Acceleration
c:J
O.lG ro 400 O.lG
r-I
P-I
Ul
X-DIRECTION 'rl Y-DIRECTION
(:1

+>

o ;...
N
\

oD.:::
10 15 20 25 Il<
0 5
'5 10 15 20 25
z 60
co ---- NO ROTATION z 60 --------,
co - NO ROTATION
ccJd-. --- 14 rps
ro ccJ 0\ _.- 14 rps
- 24 rps
H
o
40
_

40
-"- 24 rps
Acceleration
H
no O.lG Acceleration
p
'rl O.lG

ro 20 X-DIRECTION 20
Y-DIRECTION
QJ )(
P=1CO
o
5 10 15 20 25
o
10 15 20 25
5
:EXCITATION FREQUENCY Hz EXCITATION FREQUENCY Hz

Figure 15. - Influence of rotational speed.

WHIRL

ORBIT

DYNAMIC MODEL 0.1 ")

OF WATER BEARING .c mm

P'1
.2.0 Kgf
0 x 9. 8N
p:::;
0
""

0
12<
H
p:::;
1.0
0
I:-<
U)
P'1
p:::;

o rps
24 rps DISPLACEMENT 0.06 mm

Figure 16. - Waveforms and whirl orbits. Figure 17. - Restoring force
in water bearing.

121
 WHIRL ORBIT CALCULATION
Harmonic o rps 24 rps

Wave

Bearing
Load

Jouna1
Figure 19. - Whirl orbit in resonances.
Vibration

TIME

Figure 18. - Response waveforms near

a resonance.

.
toO t)Q --------

100 --------.

:>::;2;
co

.::10\
< ..
:3 80
o o rps EXP.
;2;
H
---
It<
<
60 o rps
rz.
o
---'::'---":1
A----- --------
24 rps EXP. I!z1
P
H

---- >- ---- EXPERIMENT
EXPERIMENT
24rps j 24rps NoRlinear
l
Nonlinear 20 -- - - Orps
---- Orps p.. CAL.
CAL.

STIFFNESS Kgf/mm STIFFNESS Kgf/mm

x 9.8 KN/m
x 9.8 KN /m

Figure 20. - Comparison of bearing dynamic load.

122
 APPENDIX

Time History Response Analysis of Nonlinear Rotor-Bearing System

by Quasi-Modal Technique

1. INTRODUCTION

In structure dynamics with many degrees of freedom of motion, time h istory

response analysis involves two integration methods. One of them is direct inte
gration and the other is modal. Generally speaking, the former is accurate, but
not convenient for large structure systems because too much computational time is
required. The latter is effective due to the great reduction of freedom of motion by
the normal transformation, but its accuracy depends how many eigen vectors are
introduced in the modal matrix. Each integration method inevitably contains its
own advantages and disadvantages.
These methods are also applied for non stationary rotor vibration analysis in
rotor dynamics as well as in analyzing the gyroscopic effect and large damping force
in bearings. The authors have already presented the special form of the orthogonality
condition in rotor dynamics and the modal integration technique for analyzing non
stationary vibration of a rotor in a system with linearity or somewhat weak non
linearity.
In many cases, the modal technique is more convenient and powerful than the
direct one. However, even the modal technique is not perfect and is inconvenient for
time history response analysis in a rotor-bearing system with strong nonlinear
boundary conditions.
However, in our experience it is very difficult to numerically simulate such a
vibration mode -changing due to rubbing- by the modal integration technique,which
guarantees only a linear system, and not a nonlinear one. The modal technique is
limited to linear analysis and its- forcible application to a nonlinear system is
essentially unreasonable.
In this study a new method,which is exclusively applied for nonlinear analysis
is developed on the basis of modal synthesis and the substructure method. A quasi
modal transformation is defined considering gyroscopic force and damping force
for a general rotor system. The integration method is then introduced by the quasi
modal technique.

NOMENCLATURE

matrices after the quasi modal transformation

damping matrix
equivalent damping coefficient for the inner system
gyroscopic matrix
equivalent gyroscopic effect for the inner system
external force
external force acting upon the inner system
force on the quasi modal coordinate system
imaginary unit
stiffness matrix
equivalent stiffness for the inner system
mass matrix

123
 Q(z2,z2,t) : resistance force acting upon the boundary point
t
S=[S1,S2v,s2dJ : state vector on the quasiodal coordinate system
s1 weighting value corresponding to mode
s2d= z2 weighting value corresponding to c5 mode,i.e.rotor dis
placement of the boundary point
weighting value corresponding to mode,i.e.rotor velocity
of the boundary point
t time
Z=[Z 1,z2Jt rotor displacement vector in a complex form (z=x+iy)
z1 rotor displacement vector of the inner system
z2 rotor displacement of the boundary point
c5 deflection mode generated by the forced displacement
complex eigenvalue
deflection mode generated by the forced velocity
matrix of the quasiodal transformation
eigen mode of the inner system
eigen mode of the entire system including the inner system
and the boundary point
rotational speed
Dirac's function

2. QUASI-MODAL TRANSFORMATION

Equation of motion

This study deals with the rotor vibration in an entire system, includi n g
bearings and disks,as shown in
Fig.1 . The rotor displacement is measured o n the
coodinate system of O- which is fi xed in the space. The complex form of the rotor
displacement is denoted here by z=x+iy for the displacements in the X-direction and
Y-direction of x and y, respectively. The equation of motion of a rotor is written
as:

(1)

This equation represents the elliptical whirl motion of a rotor supported by general
bearings with asymmetrical dynamic properties. In the case of the circular whirl
motion due to symmetrical dynamic properties,
where

kxx=kyy kxy=-kyx Cxx=Cyy Cxy=-Cyx

the equation of motion becomes little simpler, and can be written as:

(2)

To simplify the description, the vibration of a vertically suspended rotor with

symmetrical bearing dynamics is studied first and a new quasi-modal technique is
presented. This technique is also ex tended to general rotor vibration including
asymmetrical bearing dynamics, but the explanation is omitted in this paper.

Discrimination of boundary point and inner system

The cantilever type of rotor, as s hown in Fig.2, is selected in order to

clearly explain the concept of this new technique. In this rotor system, the left
side is a fixed point which is clearly a boundary location. A bearing with non-

124
linearity is located on the right side of the rotor. It is assumed that the bearing
dynamic properties are not constant.
This rotor system is a typical example of a system having one changing boundary
condition. The right side is thus designated as a nonlinear boundary point, denoted
by z2. The rest of the rotor system is designated as a linear inner system, and
its displacement vector is denoted by z1.
A resistance force of Q2 acts upon the boundary point, and an external force
of F1 is distributed on the inner system of the rotor.
The equation of motion (2) is rewritten, inserting displacements z1 and z2,
as

(3)

Where, subscripts 1 and 2 mean the inner system and the boundary point, respectively.

Transformation modes

An arbitr a ry vibration mode shape in the system with a changing boundary

condition can be arranged by the synthesis of three kinds of special mode shapes, as
shown in Fig.3. The quasi-modal transformation matrix thus consists of the three
modes: , 0 and f, The first two modes, and 0 , are used in the conventional
methods, called the substructure technique or modal synthesis. The third mode,
f, ,is introduced in this study for the first time and its necessity is emphasized.

<p mode

This is the complex eigen mode of the restricted system on the boundary point,
as shown Fig.3(a). It is obtained by damped eigenvalue solving with respect to the
M1-Cg 1-K11-C1 system.
The equation of motion of the inner system becomes

(4)

and the free vibration solution is assumed to be given by the following formula with
the mode,

z= e At

The orthogonality condition with respect to these eigen vectors is derived as:

(6)

= Ojk

o mode

This is the deflection mode when the rotating shaft is subject to f orced
displacement on the boundary point ,as shown in Fig.3(b). Under a load Q2*, the

125
relationship between the shaft stiffness and the displacement is expressed as:

(7)

where Z 1 = 0 is the deflection mode

and Z2=1 is the forced displacement on the boundary point. This load is equivalent
to the shaft stiffness, when the inner system is viewed from the nonlinear boundary
point, and is determined by the following relationship,

(8)

.; mode

This is the deflection mode when the rotating shaft is subject to f orced
velocity on the nonlinear boundary point, as shown in Fig.3(c). This mode shape has
not been used heretofore.
Under a load Q2 *, which generates the forced velocity, the relationship in
cluding the gyroscopic force and the damping force up to the first order differ
-
ential equation of Eq.(3) becomes

(9)

Assumin g the deflection of a nonlinear boundary point with

(10)

the deflection mode shapes of the inner system with the form
.

Z1 = 0 ,Z 1 = 0 t+ .; ( 11)

are determined by the following relationship

.; =i ';g+ ';c (12)

where,

1
.; g= -K11- Cg1 0= K11-1 Cg1K11-1 K 12

.; c= -K11-
1 C
1 o= K11-1 C1 K11-1 K12

Furthermore, this load Q2* is equal to the sum of equivalent gyroscopic effect
cg* and equivalent damping coefficient c*, when the inner system is viewed from the
boundary point. The latter values are determined by the following formulas :

cg*= Cg2+ 0t Cg1 0 (13)

c* =C2+ otC1 0 (14)

Where,

Q l =k*t+i Cg*+c*

126
 It is obvious in Eq.(12) that the s mode is generated by the gyroscopic force,
reflecting the spin effect of the rotor, and by strong damping forces in the bearings.
Thus these three modes must be considered in analyzing rotor dynamics. For ordi
nary structure dynamics, however, the third mode is not needed for the vibration
analysis of nonrotating structures with negligible damping forces. This is why the
conventional modal synthesis techniques applied to structure dynamics employ only
the eigen mode of a restricted system and the deflection mode 0 at a forced dis
placement. The new technique presented here for analyzing rotor dynamics is thus
based upon a more general concept, and includes the usual substructure and modal
synthesis techniques as well as the s mode.

Transformation of coordinate system

The quasi-normal transformation matrix is defined by analogy with the modal

transformation, as follows :

s
(15)

The equation of motion of Eq.(2) is rewritten using a state equation

(16)

Putting the transformation relationship of Eq. (15) into the state equation of Eq.
(16), and premultiplying the transposed matrix of the transformation matrix , a
simple differential equation with respect to the quasiodal coordinate of s vector
is obtained as

Bs=As-Fs (17)

where,

o o

* * *
B= B21B22 0 A= 0 i ncg +c k F s= otF1+Q2

* *
o 0 k 0 k 0 0

The structures of the transformed matrices B and A of Eq. (17) are shown in
Fig.4 . The matrices B11and A11 contain zero elements in the off-diagonal part,
because the eigen vectors of the linear inner system are orthogonal with each other,
and these matrices become diagonal. The rest of B matrix,B 12=B21t is equal to
the nonzero elements because there is no orthogonality in the relationship between
eigen vector and deflection modes 0 and s . All off -diagonal elements of the
coefficient matrices B and A are not zero,but many parts of them are. Consequently,
this transformation can not give a theoretically perfect diagonalization of the
transformed matrices, it does facilitate almost perfect diagonalization with non-zero
elements at the edge of matrix B. This transformation technique is much closer to the
modal one ; hence we call it the quasiodal technique.
It should be noted that the almost perfect diagonalization with the non-zero

127
edge can not be accomplished without the mode. That is why we introduce the mode
for this new quasiodal technique.
In state equation (17) obtained by a quasiodal transformation, the state
vector s contains ph ysical c oo r dinates of z2=S2 v and z2=s2 d , corre sp o n ding to
the velocity and displacement on the boundary point, respectively. And the resistance
force Q2 on the boundary point is also included in the force vector F s without
being reformed. These quantities related to the boundary point are maintained by the
same description in the physical coordinate system. Therefore,the superposing oper
ation is guaranteed after the quasi-normal transformation, as in the same manner as
the FEM operation.
The degree of state equation (17) is equal to the degree of the modal transfor
mation for the linear inner system plus twice the number of boundary points. Conse
quently the scale of the problem is greatly reduced,as compared with Eq. (3) or
Eq.(17).

3. TIME HISTORY RESPONSE ANALYSIS

Hybrid Integration

The integration of the state equation (17) on the quasiodal coordinate system
gives the time history response under an external excitation F 1(t).
The vector s1 of the quasiodal vector s indicates the weighted values cor
responding to the eigen modes of the inner coordinate system. Thus, it can be said
that the inner system is processed by the well known modal integration method.
On the other hand, the remaining elements of s2 v and s2d in the quasiodal
vector s equal the physical coordinates with the relationship

(18)

Thus, the boundary points are handle by physical quantities even in post-transfor
mation, and are easily processed by the direct integration method.
Therefore, integration with respect to the quasiodal coordinate s is equiva
lent to hybrid integration where modal and direction methods are mixed, as shown
in Fig. 5. The two conventinal methods have their own advantages and disadvantages,
but the advantages of both methods are guaranteed in this new method. Thus the
new technique is highly applicable to nonlinear vibration response analysis.

r--..
W'--\::;;n--rl---.--,-1! U-,-J-..-,J --

Figure 1. - Rotor-bearing system.

128
 -
: Bearing
_J

Rotor
I
Z\ Zz

Inner coordinate Boundary coordinate

Figure 2. - Boundary point and inner system.

(a)

Constrained system
free vibration
Ijodes <Pj
mass gyro stiffness damping force

2, , A, .. "'",A. iJ A 8
r
=l "
.'It '"
2, 0 1 I 0
z, ... ", f (I z, f " '"
z, 0 I z, z,

z, =]; , SII + 2:, + S Z2

" "
( c) .'i:!/. ... ,{
Unit velocity
Mode I; Quasi normal system

Figure 3. - Transformation modes.

Figure 4. - Quasi normal coordinate system.

I Direct Intergration I
z(t)=J.i{ t) d t
Accurate
Large computational -
time and cos t
Inconvenient
IHybrid Intergration I
Non-linearity is OK
+

+
)- [:?sJj[:J J[:}[;]
=

Zd= Z
[jiQdal Interg r

s(t) =J s (t)ell
-
Q2 : non-linear

z=[",)s(l) Convenient Econom i cal

Convenie,-;t Economical Non-lineari ty is OK

Non-lineari ty is NG. Oifficul t Programming

Figure 5. - Hybrid intergration.

129
 LINEAR FORCE AND lVD\1ENT EQUATIONS FOR AN ANNUlAR Sl.V[X)ffi

SHAFI' SFAL PEEmJRBED OOIH ANJUlARLY AND lATERALLY*

J. Fenwick
Rocketdyne Division
Rockwell International
Canoga Park, California 91304

R. DiJulio, M. C. Ek, R. Ehrgott,

H. Green, and S. Shaolian
California Stat Uhiversity at NOrthridge
NOrthridge, California 91330

SUMMARY

Coefficients are derived for equations expressing the lateral force and pitching
moments associated with both planar translation and angular perturbations from a nom
inally centered rotating shaft with respect to a statiqnary seal.

The coefficients for the lowest order and first derivative terms emerge as being
significant and are of approxima'tely the same order of magnitude as the fundamental
coefficients derived by means of Black's equations. Second derivative, shear pertur
bation, and entrance coefficient Tariation effects are adjudged to be small.

The outcome of the investigation delineated in this report defines the coeffi
cients of the equations:

L = AE + BE + CE + Da + Ea + Fa
(1)
F = aE + bE + CE + da + ea + fa

The assumptions utilized, mathematical means employed, and conclusions derived

thereby will be the objective of this report.

The effects of the additional terms upon a typical rotordynamic system are
presented.

INTRODUCTION

The factors leading to increasing pressures, speeds, and temperatures in jet

engines, high performance pumps, and turbomachinery for rocket engine applications
have been widely noted in the literature, e.g., Rothe (ref. 1).

The increased incidence of stability problems as a function of increased power

density, particularly the phenomenon of "subsynchronous whirl ' has been noted by many
authors.

Among the design features strong ly contributingto this phenomenon in a wide var
iety of high-powered turbomachinery are annular shaft seals. These seals are in gen
eral utilized to separate regions of high and low gas or liquid pressure. In addition
to their strong effects as pseudo bearings and destabilizing devices, they offer, when

*A11 work performed under NASA Marshall Space Flight Center Contract NAS8-27980

130
properly configured, in many designs the only real possibility for the introduction of
damping in rotating systems. Various authors have delineated the extent of both the
problems and remedies surrounding these possibilities; for example, Alford (ref. 2),
Ek (ref. 3), Childs (ref. 4), and Gunter (ref. 5).

A very significant part of the understanding of the dynamic behavior of annular

smooth shaft seals has come from the pioneering work of Henry Black of Heriott Watt
University, Edinburgh, United Kingdom, who in a series of papers (refs. 6-12) defined
the effect of shaft displacements with both long and short seal assumptions up to and
including the second derivative of the displacement. A number of workers have ex
tended his work significantly, these being Jensen (refs. 7 and 8), Hirs (ref. 13),
Childs (ref. 4), and Alliare (refs. 5 and 14). The recent trend is to shift to a mod
ification of the bulk flow theory to define more closely a set of equations approxi
mating experimental results IHirs (ref. 13)].

On many significant recent problems, e.g., the problem on the high-speed rota
ting machinery for the Space Shuttle Main Engine (SSME) IEk (ref. 3)], Black's equa
tions were the only analytical representa-
tion available for significant computer
modeling. Present efforts to extend this
work include continuing work by Allaire
(ref. 1 4) at the University of Virginia,
Childs (ref. 4) at Texas A&M, and Fleming
(ref. 12) of NASA Lewis, as well as exper
imental work with liquid oxygen and hydro
gen at the Rocketdyne Division of Rockwell
International.

This study represents a departure

Figure 1. Orientation of references
-

from the general approach being carried

axes for seal equations.
on by other investigators in that it ex
plores the effect of pitching moments and angulation in the seal, as shown schemati-
cally in Fig. 1.

NOMENCLATURE

The nomenclature in general is that employed by Black (ref. 6) and specific

nomenclature is introduced as comments in the body of the nomenclature table, in the
figures, or presented below. Additional functions are delineated in the text as
required.

A-F } =
coefficients in force equations
F
n
= force on seal, n is defined in the
a-f text
b =
gap thickness, y(x,t), inches
H =
PRTIa/6A(1++2a), general coeffi
D =
d/dt, differential operation cient for moment and force equa
with regard to time tions = 0.5
E shaft elastic modulus =
PRTIaL/AY

F =
force on seal, total due to
perturbations

I
=
shaft area moment of inertia, m 4 e =
aF/dE, N/m3 (lb/in.3)
(in.4 )
V = kinematic viscosity, m2 /s
K
=
effective bearing spring rate, (in. 2 /sec)
N/m (lb/in.)

131
L length of seal, m (in.) friction coefficient, p

5l, component spacing on rotor, m (in.)

area b
N speed of rotation, rpm r
n wetted perimeter,Z
p total pressure difference across
Z
seal, Pa (lb/in. )

P pressure within passage, p(x,t),

Z
Pa (lb/in. )
loss factor per Yamada (ref. 15)
R =
shaft radius, m (in.)
11jk functions of a, coefficients for
R axial Reynolds No. ZVyo/v the perturbation terms
a
R rotating Reynolds No. Rwyo /V absolute viscosity, Pa-s (lbf
r 11
T passage time, L/V, seconds sec/in.Z )

t time, variable, seconds entry loss coefficient =

0.5
nominally
u axial fluid velocity, u(x,t), m/s
(in./sec) p mass density, i' kg-s Z/m4 (lb
g
sec Z/in.4)
V =
mean fluid velocity, m/s (in./sec)
a AL/yo
W =
weight of rotor, kg (lb)
T =
pitching moment taken about
x axial distance, independent
variable, m (in.) x =
, N-m (lb-in.)

y,z radial displacement of rotor mass T pitching moment at x 0, N-m

o
a angular displacement (lb-in.)

y 1.5 + Za, = 0.5 dF/da, N/rad (lb/rad)

3 3 dT/de:, N (in.-lb/in.)
<5 weight density, kg/m (lb/m )

E lateral displacement, yet), m (in.) dT/da, N-m (in.-lb/rad)

w circular frequency, rad/sec

w design speed, rad/sec

ASSUMPTIONS

In the following equations, assumptions are made as follows:

1. The system (shaft within seal) is assumed to be centered.

Z. Perturbations about the nominal centered position are small.

3. Whenever two second order terms are multiplied, they are assumed to be higher
order, i.e., the product approaches zero.

4. Short seal theory is assumed, Le., perturbations have negligible effect in

the tangential direction.

5. The period of time required for a particle of fluid to travel from the inlet
to the discharge is small compared to the frequency of the system.

6. There is no change of properties in the working fluid as it passes through

the process.

7. The entrance coefficient into the seal is assumed to be 0.5.

l3Z
 8. The effect of fluid shear variation on the shaft moment is neglected in the
fundamental equations.

9. Total linearity is assumed, i.e., any of the effects can be added individ
ually with no interdependence between individual effects.

10. Tangential fluid velocity equal to one-half the shaft tangential velocity is
developed immediately at the seal entrance.

11. Perturbations are planar; however, Yamada's (ref. 15) work in defining A
for rotating shaft effects applies.

APPROACH

The process of deriving the fundamental equations departs from that used by Black
(ref. 6) in three important respects: (1) clearance is expressed as a function of
both x and a, (2) the equations for continuity and velocity are completely rewritten
as a consequence of (1), and (3) the order of integration is reversed. In addition,
the first derivations lead to a result in which all perturbations, moments, and
forces are taken at x = O. This is done because of great simplification of the
boundary conditions. These are then translated to perturbations about x = L/2, and in
the case of the moments, T
o
is translated to T
L/2
I
Explanations of each step and
necessary symbol derivations occur as needed in the text.

DERIVATION OF FUNDAMENTAL EQUATIONS

I
We first consider two plates of length
L and unit width, nominally y in. apart, I----x b P2

P, .po ---+-U DRAIN

with the top plate fixed, the bottom plate Y
perturbed: perturbations are positive up
ward and counterclockwise, as are forces
SOURCE lO-- -- O'+ WJ777hI

a
and moments. Flow is positive to the right NK
-------- ------..-x
(Fig. 2). No fluid moves tangentially or I
normal to the surfaces of the plates. ::+ + T L ----- ..-il
The steady-state pressure distribu Figure 2. - Derivation of fundamental
tion through the fl process is reflected equations.
by the following relationships:
2
p(1.S + 2cr)'jL (2)
2

1. 5 2
P - p = - pv (3)
1 0 2

The fundamental equation for head loss through the flow process is

(4)

From the Navier-Stokes equations for unsteady flow,

l (5)
p dX

133
If a control volume of length dx, and bounded by the two plates, is now considered,
the plates can be thought of as perturbed streamlines 0 The general continuity equa
tion for an unsteady stream tube of fixed length ds is

a(pA)
-+

A)
at +L
as (p o (6)
0

A A
-+ -+
Since p = constant, b, u ub, s = x,

ab a(ub)
+ o (7)
at ax
or

(8)

Now, approximately

a E: x
dU
ax =

Y
u + Y: + y: a
OJ
o
or
(9)

au a
---u
E: + xa
dX y
o

and U(t) v +v(t),

2
I x
(10)
u - EX + 2 a
(0 0) +V +v +
Y
=

Higher order terms (HOT) are droppedo Now, integrate and set the following boundary
conditions:

2
at x = 0, p =
P
o
=
P
I
- 1;5 p( V + 2Va)

and
at x = L, P = P
2

1
- - p (V +
2VA
v ) x +
x
2
E +
( Vx
+
AVx
2
) E:
p y 2y Y 2
o o o Y
o

3 2 3
x Vx AVX
(11)
+ --
6y
a + --- +
00

2
( --- ) a
0

Y
o o 3y
o

2 2 2 2
V x
+
1 AV x ) AV
+ c(t)
a +
+
( y 2 2 Y
x

o Y o
o

Therefore
PI
1. 5 2
c(t) - -+ P (v + 2Vv)
-
p 2
o

134
But
2
PI
- P2
= P = p(I.5
+
20)
o - [V +
(1. 5
+
20) V] +
A [E +
f 2 (1 +
0) E
+
20
: ]
(12)
+
[a +
2(3 +
o)a
+
3(
2
+ 30)a ]
+
If now, p = P p(t), then Eq. (11) becomes
steady

x2 Vx + Avx2
_
p
= (v + 2 OA v ) x
+
2y
E
+
( Yo Yo
) E
o

+ A 2
V x
(13)
-- 2 e:
Yo

2 2 2
+
( V x +
y
1.
2
AV 2 x ) a
+
1. 5 Vv
o Yo

DERIVATION AND SUMMARY OF FORCE AND PITCHING

MOMENT EQUATIONS

The symbolism and form of the equations for shaft is:

}
a

+ + + +
L = Ae: BE CE + Da Ea Fa
+ + + + +
(14)
F = ae: bE CE da ea fa

The fundamental assumption of linear independence and superposition is made, i.e.;

+ +
Fl ae: bE CE
+ +
F2 da ea. fa

+
L
l
Ae: BE + CE
(15)
+
L2 = Da + Ea. Fa

+
F Fl + F2 , L = L L2
=

In the above, all displacements, moments, and forces are taken about x = L/2. How
ever, the first solutions will be taken about x = 0, and the results transferred to

x = L/2. The subscript convention at x = 0 will be

L (16)
o

135
For the cases about x 0, certain simplifying parameters will be defined:
= the re
sulting form of the equations will be:

+ b + c t: + T +
2
F a t: H ( t: T E) (17)
lO 10 ll 12
=

0 0 0

+ Ta +
2
+
F d ex. + e ex. f a. H ( ex. T ) (18)
20 0 0 0 20 2l 22
a.

+ + 11
+ 2
T A t: B C E H ( t: + T T E) (19)
lO 0 0 0 I 30 3l 32

+ Ea + Fa +
2
T Dex. H ( ex. + Ta T a) (20)
20 I 40 4l 42
where T L/V, and H, H ' and j,k will be defined as part of the derivation.
I
=

FORCE AND PITCHING MOMENT EQUATIONS FOR A SHAFT

The definition of pressure perturbations seen in Eq. (13) is for two plates of
unit width. Utilizing these, the following force and pitching equations can be writ
ten by integrating around a perturbed shaft as in Fig . 1. For example,

F
_0_ f L
6p dx _ (v +
2VAV
)
2
+
6
3
E +
( VL
2
+
AVL
3
) E
Rnp
=

o
{ y
o
2 y
0
2y
0
3y
0

2 2 2 3
+
_ V _L_
( 2y
o
+ _A_v_ -=-_
2y
o
) ex. + 1. 5 Vv L I (21)

2 [ 2
F
o
Rnp -{ [ iT + (20
+
3)v 1 V L
+ --
6
Yo
2 t: +
L

V
..
V
L
(3 + 20)t:
.
+ 30t:
]
2 2 [ 2
+ +O )ex. ]
+
2 a (8 + 20)a + 12(1 (22)

and similarly,
4
0
}
T
o
R'lTP
_ f VL2
3
( .!:c v. + 2OV
V
+
(1. 5) 2v ) 2
+ LV a-
8A
[ L

V
2

2
.
t: .
+ 8L
V
( 1+.2:.
3 4
) .
t:

+
2 2
L V 0
30A
[ L
2

2
..
ex. +
30L
V
( 1 +
4 15
) ex.. + 30 ( 1 + 30 ex.
3 8
) J} (23)
v

After eliminating v, v between these equations (23-24) and (12-13), the forces and
torques can be defined. This is done piece-wise, then finally superposed.

136
FORCE DUE TO LATERAL DISPLACEMENTS

This will be expressed in the form

(24)

From Eq. (12), the equation expressing v,v as a function of , , E, is, with

( a, Ct, a 0)

o " [v + I (1. 5 + 2a)v) + A [E' + I 2(1 + a)E + 2ac) (25)

This can be expressed as

2 2
o ITD + y]v + [T D + 2(1 + 0)TD + 20] (26)
L n

Now, from Eq. (19), with a, a, a 0

}
=

F 2
L 2
3 ]
l O ( VL V [T D (3 20)TD
- ) - [TD + (20 + 3)]v + + + + (27)
-

R'ITP =

t2 6y
o

Eliminating v, v between these two equations (26-27) with TD y, eliminating

(TD) of third and higher power,

(28)
where

01TRP
H y (1.5 + 20)
6"Ay
=

90
Y
2
(9 + 120 + 40 ) Y - 90
2
Y
3 2
(80 + 180 + 180)
3
y

This agrees with Black ( ref. 6) except that in ' Black has 190; 180 is correct.
12

FORCE DUE TO AN ANGULAR PERTURBATION

This will be expressed in the form

(29)

137
Proceeding as before,

PR7[(1
H Y 1.5 + 20'
6AY
=

2
1120 =

.
31(20' + 60' + 3)
Y

1 2 2
11 -Z [y(20' + 11.5 0' + 12) - (60' + 180' + 9)]
21
=

1 2 2 2
11 = {y (4.25 + 20') - y(20' + 11. 5 0' + 12) + (60' + 180' + 9)J
22
y

PITCHING MOMENT DUE TO A TRANS1ATION

The pitching moment due to a translation will be expressed in the form

(30)

The equation expressing the boundary condition is (26)

v 0' 2 2
o = [TD+y] v +
L 2A IT D + 2(1 +O')TD + 20')E (31)

From Eq. (21), a, a, a = 0,

t V13 2 (TD + 20'+ 1.52)v

Rnp
- --

+ v2 I [ T2D2 + 8 ( + % ) (32)

Eliminating v,v, as before,

(33)
where
Rn0'1P
, y 1. 5 + 20'
AY

1
-

Y
=

138
The above pitching moment is about x = O.

TORQUE DUE TO AN ANGU1AR PERTURBATION

After proceeding as before, Eq. (20) is expressed in the form

4
(3 )

where
PR1W1
H
YA
2
1 2
P4 = 4 y [22.50 + 50.670 + 22.5]
0 5

4
l
=
4 2
r y(80
2
+ 39.750 + 33.75) - (22.50
2
+ 50.670 + 22.5)]
5y

2
- 1
[80 +
4 2
l . 25)y - YC80
2
+ 39.750 + 33.75)
W42 - 4 y3
5

2
+ (22.50 + 50.670 + 22.5)]

TRANS1ATION OF EQUATIONS TO x = 1 /2

The translation of the solutions for perturbations about x 1/2 can now be =

written. These involve the ex terms only. An x rotation at 1/2, can be con =

sidered as the sum of an ex perturbation at x 0 plus a displacement of -ex , so =

that at L/2 there is a pure ex perturbation. 2

1 1 1 ..
F F F d ex + ea + f a a -ex b -a - c - ex
2 20 lO(-ex) 0 =

0 0 0 2 0 2 0 2

(do - a )ex
o
( o - b0 2:1 ) . (fo co -12)"
+ e ex ex (35)
and

+ + C
1 " )
L
20 =

( A0 ) (E0 - B0 !:)a
Do - 2
(F0 - 0
"2 ex (36)

These are both for ex rotated at x = 1/2, but the moment is still at x O. The next =

step is to translate the resultant force and moment to x 1/2. The fundamental re =

lationships are (F F ),
o
L (37)
o

To summarize, these become:

Forces:
a EO: + b E: + c E since a a, b b' c c
0 0 0 0 0
= =

(38)

139
 Moments
- F
L (A a )E + (B b )E + (Co - c
L)..E
T T
l IO l 2 o o o2 0 0 2

L ( L L2)
+
T2 T20 - F 2 2 Do - A 2 -
=
2 ao -
4
a 0
-
- C
0

( L L 2).
+ F - B
o 2 e +2 b -a
0 4 0 0

+ (F - C
L2) a
L - f + C 4--
..
(39)
o 2 2 o0 0

where
2
a 0
H
]110 b 0
]112 TH
]111 T C
0
H

2
d 0
H
H
]120 , e H
0
]122 T ]121 T , f
0

and C H ]1 H T2 T
A 0
H
I ]130
, B 0 I 32 I ]131 0

E H T F H T2
D H ]1
I 41 , I ]1 42
0I ]140 0 0

The symbols are previously defined. And finally,

d

1
T AE + BE + CE + Da + Ea + =

(40)
F aE + bE: + CE + da + ea fa +

where
A A a L2 B B b L2 0 0
C C C L2 0 0
-
0 0
-

L L 2
D D A 2 - d + - a
02 4
0 0 0

2
L
E E
0
B L2
0
- + b
2 e
4 0 0

L L L 2
F F -C 2 - f
- + C
=

0 0 2 4 0 0

and where
a=a b=b C = C
0 0
0

L
d d 0
a L2 e = e b L2
0
-
0 0
- f f 0
C
0
-
2

140
 EVALUATION OF SEAL COEFFICIENTS

The geometry and parameter values shown in Table I were used to evaluate a
sample rotor. For comparative purposes, the properties of two fluids, water and
steam under supercritical conditions were used.

TABLE I. - GEOMETRY AND PARAMETERS

SEAL

R - RADIUS 7.62 em (3 INCHES)

L - LENGTH 2.54 em (1 INCH)
Yo - CLEARANCE 0.254 mm (0.01 INCH)
N - SHAFT SPEED 30,000 RPM
p .- PRESSURE DROP 344.6 N/em2 (500 PSI)
JJ - VISCOSITY (STEAM) 2.067 X 10-9 NS/em2 (3 X 10-9 LBSEC/IN 2)
fJ - DENSITY (STEAM) 13.84 Kg/m3 (0.0005 LB/IN.3)
JJ - VISCOSITY (WATER) 1.013 X 10-5 NS/em2 (1.47 X 10-5 LB.SEC/IN.2)
6 - DENSITY (WATER) 996.5 Kg/m3 (0.036 LB/IN. 3)

ROTOR

E - ELASTIC MODULUS 2.067 X 107 N/em2 (3 X 107 LB/IN.2)

I - AREA MOMENT OF INERTIA 24.97 em4 (0.6 IN.4)
K - BEARING SPRING RATE 1.75 X 106 N/em (106 LB/IN.)
J1 - COMPONENT SPACING 15.24 em (6 IN.)
W - ROTOR WEIGHT 36.28 Kg (80 LB)
G - .,e3K/EI

As indicated in ref. 6, if all dynamic terms are retained in the coefficient

formulation, the results are in the form of a frequency-dependent cubic divided by
a first-order term. A Taylor's expansion was used for the denominator to obtain a
simple polynomial form for the coefficients. Questions remain as to how many terms
should be retained and what range of frequencies are allowed before the approxima
tion degrades.

To answer the question of higher order terms, the equations for the seals
were established in the cubic form and evaluated for the conditions of Table I.
The resulting transfer functions are shown in Table II. From these examples, it
is obvious that the only affects that may legitimately be considered are those of
stiffness and damping. Only those two affects were considered in the evaluation
that follows.

Seal coefficients were evaluated over a wide range of speed by assuming that
pressure drop across the seal varied as the square of speed. The density for water
was assumed to be constant but the density of steam was assumed to be proportional
to pressure drop. The spring rates were found to vary with speed squared for both
water and steam. The damping coefficients, however, varied with speed for the case
of water and speed squared for steam. The restoring force and torque as a function
of shaft translation and angulation are therefore described by

141
 TABLE II. - SUMMARY OF SEAL COEFF ICIENTS AT 30,000 rpm

WATER

e (N/cm) = 7.84 X 104

( 1 )(
1
+ 5 8.3
1
S
+7347)( )/( )
1
+12 00
1
+7 30
"" 7.84 X 10
4
+
150.2 S

</J (N) = 6.806 X 105

( ; )(
1
+ 3 53
1
+ ( ) ( t);( )
1.948
10 40 + 10 40 1
S
+ 79 30
"" 6.806 X 105
+
264.1 S

x (N) = -3.323 X 104

( )(
1
S
+ 7949
1
S
)( )
, 1
S
+1057 '/, +7930
"" -3.323 X 4
10 -31.43 S

5
"" -1.2308 X 10 19.76 S
+

STEAM

e (N/cm) 9.327 X 104 "" 9.327 X 104 18.29 S

+
=

</J (N) = 1.0163 X 106

( : )( ( ) ( n;( )
1
+ 2 295
1 1.872
+ 84 80 + 84 80
1
+42 63 ""1
.0163 X 10
6
+
25.71 S

x (N) = -3.949 X 10
4
( )( )/( )
1
S
+5880
1
S
+45365
1
S
+42063
"" -3.949 X 10
4
- 8.524 S

1/1 N-cm) = 3.536 X 105

( )( )( ; )/( )
1
+41 00
1 -
80 63
1
+23 705
1
+42 63
-3.536 X 10
5
+
2.688 S

where S = jw

] ]
Where operating point variations are described by

e e
dt dt

d d
E E
dt dt
steam water

ROTOR MODEL

A simple rotor model was necessary that would exhibit both translation and
angulation at the seals. A flexible massless shaft was chosen with a mass load at
the center. Seals were placed at each end of the shaft and an ideal bearing was
located halfway between each seal and the central mass. Properties for the rotor
model are contained in Table I.

The restoring force due to the shaft, bearings, and seals due to the lateral de
flection (Y or Z) at the mass is given by

Fly N/:5 (41)

where

[ 1 + ( 1 + 34 G ) e 3 G
+'2 K ( + x) + 2 2G + 12 K
5 (G )2 (e x)
N -2K
K K
-
}'

142
 D (1 Q) (+
3
+
3
+ L
36
C ) c8 (2 1:.4 ) (
K
+ + C
tK
+ X)

+ (
2
+
12
) C
2
+ (1 ) ()2
3
+
18 tK (8ljJ - X)
.tK

Inserting values for the structural properties and for the frequency dependent journal
coefficients, we obtain a dynamic force-deflection relation at the mass.

F
Y
(42)

Since the journal equations were developed for a condition where the flow is
ostensibly axial, we assume that they are fixed to Couette coordinates. These
equations must be transformed to rotor coordinates to determine the whirl orbit.
Now assume, steady circular motion of the shaft in the y-'z plane. Transforming the
force-deflection equations to coordinates rotating with the shaft, we obtain

2
w
4
(43)

The equations for the rotor with the central mass unbalanced a distance I:,. from
the axis of rotation in the y direction are

w
2W- y F
g Y
(44)
w
2W- z F
g z

Solutions for shaft motion due to the unbalanced mass at a particular speed (w) are
obtained by first scaling the seal coefficients that were obtained at design speed
(w). Then solutions of equations (41-44) produce the normalized orbit radius, or
amplification ratio

EVALUATION OF ROTOR RESPONSE

Normalized rotor deflection vs shaft speed is shown in Fig. 3 and 4 for the
steam and water journals, respectively. Rotor response considering only the trans
lation coefficient (a) is shown for reference. The rotor response with the complete

143
 200r----,
10
TRANSLATION
COEFFICIENT ONLY
/ TRANSLATION
COEFFICIENT ONLY

FULL COEFFICIENTS
IN BOARD FLOW
o o
a: i=

a:
z z
o o
100
5
c.J
u:
::::i

LA.

... ::::i
:E ...
:E

O ..------------------------
goo;:----:;;;250::::
FREOUENCY,HZ
200 250
FREQUENCY, HZ
300

Figure 3. - Rotor response with Figure 4. - Rotor response with

steam as fluid. water as fluid.

set of forces are shown for cases where the flow is outboard through the seals and
where the flow is reversed. Reversed flow (inboard flow direction) was simulated
by changing the sign of the off diagonal coefficients ( and ).

To evaluate sensitivity of the various coefficients, cases with the off

diagonal terms increased by 10 percent were run as well as a case where the dia
gonal coefficient () was increased by 10 percent. A summary of these cases in
terms of critical speed and effective damping (/ ) are contained in Table III.
cr

TABLE III. - SUMMARY OF CRITICAL SPEED CALCULATIONS

STEAM WATER
SEAL STIFFNESS
MATRIX SPEED (RPM) DAMPING SPEED (RPM) DAMPING

13266 0.0% 13266 0.0%

[ ]
13585 0.28% 13780 5.12%
[ ]
13807 0.46% 15540 7.58%
[ ]
13323 0.60% 15510 8.47%
[- -: ]
13584 -1.2% 14880 -11.2%
[ ]
[.1X
]
1 13830 0.77% 15900 6.36%

13806 0.32% 15410 7.87%

[ tl til]

144
 DISCUSSION OF THE RESULTS

When angulation affects are included, as well as the translation, the critical
speed is shifted slightly and the whirl radius is significantly reduced.

Off diagonal seal coefficients are very significant; without them, the effect
of angulation is destablilzing. This indicates that the centers of pressure for
rotation and for translation do not coincide. Reversing the journal flow (to the
inboard direction) actually increased effective damping for the cases investigated.

The second order "mass" affects of the journals were negligible and for whirl
the mass affects may be deleted.

For the two fluids that were considered in the journal, spring rates were very
similar but damping coefficients were an order of magnitude higher with water. This
is evidenced by an order of magnitude increase in damping at critical speed for thal
journal. For both fluids the spring rate varies with seal pressure drop and there
fore speed squared. The damping coefficient varies with speed for an incompressible
fluid and the square of speed for a compressible fluid.

CONCLUDING REMARKS

From a study of the results, the following conclusions have been reached:

1. Journal forces associated with the second time derivative of shaft motion
should be ignored. The coefficients are very small and are significant
only in a frequency range where the approximations required to obtain them
are questionable.

2. Forces associated with the first-time derivative are very important.

3. Pitching moments and angulation effects are as significant as translation.

4. The net effect of moment and angulation is a slight shift in critical speed
and a significant decrease in peak amplitude.

5. While the partial of moment with respect to angulation appears to be de

stabilizing, off diagonil coefficients result in net stabilization. This
result is due to a lack of coincidence between the centers of pressure
for translation and rotation.

6. Seals, where the flow is outboard, showed lower damping than where the seal
flow was inboard. Higher critical speeds, where a different mode shape is
involved, may not exhibit the same trend.

7. Damping forces at speeds below the design value are reduced more for a
compressible fluid than for an incompressible one.

8. Moments due to variation in fluid shear along the shaft surface are very
small.

9. Coefficient accuracy is not known in the absence of experimental data.

Specifically this approach assumes full turbulent flow with a rotation at
w/2 throughout the seal. Black (ref. 11) and Hirs (ref. 13) comment on
this. For short seals, a significant error may occur because of the slow
development of tangential velocity.

145
 REFERENCES

1. Rothe, K.: "Turbopump Configuration Selection for the Space Shuttle Main Engine,"
ASME Paper No. 74 FE-23, Joint Fluid Engineering Conference, Montreal, Canada,
May 1974.

2. Alford, J. S.: "Protecting Turbomachinery From Self-Excited Rotor Whirl," Journal

of Engineering for Power, Trans. ASME, Series A, October 1965, pp. 333-344.

3. Ek, M. C.: "Solving Subsynchronous Whirl in the High-Pressure Hydrogen Turbo

machinery of the SSME," Journal of Spacecraft and Rockets, Vol. 17, No. 3, May
June 1980, p. 208.

4. Childs, D. W.: "The Space Shuttle Main Engine High-Pressure Fuel Turbopump Rotor
dynamic Instability Problem," Gas Turbine Conference, Philadelphia, Pa, 27-31
March 1977, ASME Paper No. 77-GT-49.

5. Allaire, P. E., Lee, C. C., and Gunter, D. J.: "Dynamics of Short Eccentric
Plain Seals With High Axial Reynolds Number," Journal of Spacecraft and Rockets,
Vol. 15, No. 6, 1978, pp. 341-347.

6. Black, H. F.: "Effects of Hydraulic Forces in Annular Pressure Seals on the

Vibrations of Centrifugal Pump Rotors," Journal of Mechanical Engineering
Science, Vol. 11, No. 2, 1969, pp. 206-213.

7. Black, H. F. and Jenssen, D. N.: "Effects of High Pressure Ring Seals on Pump
Rotor Vibrations," Fluids Engineering Division, ASME, 7l-WA/ FE-38, 1971.

8. Black, H. F. and Jenssen, D. N.: "Dynamic Hybrid Bearing Characteristics of

Annular Controlled Leakage Seals," Institution of Mechanical Engineers, Vol.
184, Pt. 3N, 1970, pp. 92-100.

9. Black, H. F. and Cochrane, E. A.: "Leakage and Hybrid Bearing Properties of Ser
rated Seals in Centrifugal Pumps," Paper G5, 6th International Conference on
Fluid Sealing, Munich, 1973, pp. 61-70.

10. Black, H. F.: Lateral Stability and Vibrations of High-Speed Centrifugal Pump
Rotors," Dynamics of Rotors, International Union of Theoretical and Applied
Mechanics Symposium, Lyngby / Denmark, 12-16 August 1974.

11. Black, H. F., Allaire, P. E., and Barrett, L. E.: "Inlet Flow Swirl in Short
Turbulent Annular Seal Dynamics," 9th International Conference on Fluid Seal
ing, BHRA Fluid Engineering, Leeuwenhorst, Netherlands, April 1981.

12. Fleming, D.: "Effective Spring Rates of Tapered Seals," NASA Lewis Turbo
machinery Work Shop, March 1981.

13. Hirs, G. G.: "A Bulk Flow Theory for Turbulence in Lubricant Films," Journal
of Lubrication Technology, Trans. ASME, Series F, Vol. 95, No. 2, April 1973,
pp. 137-146.

14. Allaire, P. E., Gunter, E. J., Lee, C. C., and Barrett, L. E.: "Load Capacity
and Hybrid Coefficients for Turbulent Interstage Seals," Report No. UVA/ 528;
49 / ME76/ l03, September 1976.

15. Yamada, Y., "Resistance of Flow Through an Annulus With an Inner Cylinder
Rotating," Bulletin JSME, Vol. 5, No. 18, 1962, pp. 302-310.

146
 MEASUREVIENf OF INfERSTAGE FLUID-ANNULUS DYNAlVllCAL PROPERI'IES

M. L. Adams
Uhiversity of Akron
Akron, Ohio 44325

E. Makay
Energy Research & Cbnsultants Cbrporation
Mbrrisville, Pennsylvania 19067

I. A. Diaz-Tous
EPRI
Palo Alto, california 94303

ABSTRACT

The work described in this paper is part of an Electric Power Research

Institute sponsored effort to improve rotor vibrational performance on power plant
feed water pumps. A major objective of this effort is to reduce vibration levels by
devising interstage sealing configurations with optimized damping capacity, realizing
that the typical multistage centrifugal pump has several more interstage fluid
annuli than it has journal bearings . Also, the fluid annuli are distributed between
the journal bearings where vibration levels are highest and can therefore be
"exercised" more as dampers than can the bearings. Described in this paper is a
test apparatus which has been built to experimentally determine fluid annulus
dynamical coefficients for various configurations of inter-stage sealing geometry.

INTRODUCTION

As originally cited in an Electric Power Research Institute survey on feedwater

pump outages, Reference (1), excessive vibration is responsible for many power plant
forced outages. The major cause of this excessive vibration is now widely recognized
as the fluid dynamical forces generated within high-head centrifugal pump flow
passages. These dynamical forces are a natural by-product of the high rate of energy
transfer to the fluid within a relatively small space and the fact that this transfer
of energy cannot, of course, take place at 100 percent efficiency. The farther away
from the best efficiency flow a feed pump is operated, the stronger these dynamical
forces become, particularly under the low-flow conditions required at part-load
operation. Hydraulic excitation forces will remain an inherent feature of feed pumps.
Further research on pump hydraulics may possibly reduce their intensity, but their
elimination as an important practical consideration would appear to be unlikely. The
most promising approac is to optimize system damping, the classical approach when
the dynamical forces are not adequately controllable.

One of the approaches presently being pursued is to devise high-damping inter

stage fluid-annulus co.nfigurations, Reference (2). The typical multi-stage centri
fugal pump has several more inter-stage fluid annuli than it has journal bearings
(see Figure 1). Also, the fluid annuli are distributed between the journal bear
ings where vibration levels are highest and can therefore be exercised more as
dampers than can the bearings.

147
 As shown in Reference (3), wear-ring geometry is already known to be a poten
tially major influence on critical speeds. Currently used geometries are shown in
Figure 2, with some type of serrations often preferred to accommodate rubs.
However, some pump manufacturers have employed smooth or shallow groove geometries
to utilize the resulting radial stiffening effect which can raise the first crit
ical speed considerably above the operating speed. However, as also shown in
Reference (3) (see Figure 3), this stif.fening effect (called "Lomakin" effect)
deteriorates with wear, which can cause high vibration levels after several hours
of normal operation. In spite of the attention this potential stiffening effect
has received, practically no attention has been given to the potential damping
capacity of interstage fluid annuli. Raising or lowering critical speeds can not
circumvent the undesireable effects of large hydraulic excitation forces but proper
ly adjusted damping can.

ROTOR VIBRATION DAMPING EVALUATION

A linearized vibration mathematical model is generally the appropriate start

ing pOint to study and understand rotor vibration characteristics. In the presence
of vibration, an interactive dynamic radial force occurs where there is a close
running clearance filled with a liquid or gas (e.g., journal bearings, seals,
wear-ring clearances). Such interactive forces are commonly characterized in a
linear model as shown in the following matrix equation.

Here, [K], [B] and [0] are the stiffness, damping and virtual mass matrices, re
spectively, of the entrapped flufd within the close-running radial clearance.
(x,y) is the instantaneous rotor-to-stator radial displacement vector with respect
to static equilibrium and (fx,f ) the instantaneous radial dynamic force vector..
y
Presently, there is little reliable information on the dynamic matrix coefficients
for feed pump fluid annuli.

To evaluate and compare damping capacity of various fluid-annulus geometries,

it is a considerable simplification if harmonic motions are used. First of all,
this provides a convenient way to absorb the [0] matrix into the [K] matrix as
commonly don.

[K] = [K] - Q2[D] (2)

Here, Q is the frequency of the orbital vibration (see Figure 4). As described in
Reference (4), the [B] and [K] matrices are non-symmetric for journal bearings and
other fluid annuli contained within a rotating and non-rotating boundary. To
separate out conservative and non-conse,rvative (damping) effects, the [B] and [K]
matrices are decomposed into symmetric and skew-symmetric parts.

[B.] -1 [B.. B ..] , symmetric (positive damping)

2
= +
lJ lJ Jl

[B] -1 [B.. - B .. ] , skew-symmetric (contributes no damping)

2
=

lJ lJ Jl

148
 1
[K.J
lJ
=
-
2
[K . . + K . .
lJ Jl
J , symmetric (contributes no damping)

1
[KJ
lJ
=
-
2
[KlJ.
. - K
Jl
. . J , skew-symmetric (negative damping) (3)

The instantaneous non-conservative interactive force vector on the rotor can there
fore be expressed as follows.

(4 )

The net energy imparted to the rotor (at a fluid annulus) per cycle of orbital
motion can therefore be expressed by evaluating the integral of the non-conservative
force vector, {P}, with the differential radial displacement over one period of
harmonic motion, expressed as follows.

E
cyc
=
f {P} { dX}

sS
_ _ 1T
s
[ rG (Bxx X
2
+ BSyy y2) _ 2K
xy
sin (e -
x
e)J
y
(5 )

Here, x and y are the principal coordinates of and X and Y the corresponding [BsJ,
single-peak amplitudes, and , e ) the respective phase angles.
(e
x y

For any co-rotational orbit sin (ex - ey) > O. Furthermore, if as with journal
bearings, x 'B and K are all Byy
positive, one sees
the presence of both positive
and negative damplng effets on forward whirls. It is clear from equation (5) why
rotor-bearing instability always occurs as a co-rotational or forward whirling
vibration.

A compact way to evaluate the net damping capacity of a fluid annulus (or
journal bearing) is to determine Ecyc as a function of vibration-to-speed frequency
ratio. For example, trends of journal bearing damping (Ecyc) are shown in Figure
5, which is an alternate way of explaining why rotor-bearing instability occurs
when the lowest rotor-bearing resonance frequency is below the zero-damping cross
over frequency ratio. A similar approach for evaluating net damping capacity of
interstage fluid annuli could be employed if the dynamic coefficients were known.
Also, predictive analyses of feed pump vibration in general would be considerably
advanced with reliable dynamic coefficients. A test rig has been designed and
built to experimentally determine the and [OJ, [BJ
matrix coefficients under [KJ
operating conditions in feed water pumps.

TEST RIG AND GOVERNING EQUATIONS

Under Electric Power Research Insti tute sponsorship, a test rig has been designed
and built to experimentally determine the dynamic coefficients of currently u sed and
newly devised inter-stage fluid-annulus configurations. Testing is currently in
progress and the results will be published when the work is completed.

A conceptual sketch of the test rig is shown in Figure 6, and a detail layout
of the actual rig is shown in Figure 7. The concept employed in the design of this
rig follows directly from the governing equations which relate the interactive

149
dynamic force components and the components of relative radial harmonic motion.
As developed in Appendix A of this paper, twelve independent equations are needed
which relate the force and motion parameters to the twelve dynamic,coefficients.
These equations are summarized below for the'harrnonic.. circular orbi , which is the
vibration mode built into the test rig eccentric-spindTes design;/

(F cos e )/R
x x
= - K
xx
- Q.B
J xx
+ Q D
J xx

(F
x
sin e )/R
x
= + K
xy

Q D
J xy
(6)
(F cos
y
e )/R = - K
yx
+ Q D
J yx
- B Q.
yy J
Y

(F
y
sin e )/R
y
= - Q.B
J yx
- K
yy

Q D
J yy

where, j = 1,2,3

The controlled parameters are X and Y, the single-peak vibration amplitudes

and their respective phase angles x and y. The measured parameters are Fx and Fy,
the single-peak dynamic force amplitudes and their respective phase angles, ex and
ey . As implied by equations (6), test data is needed at three different vibration
frequencies for a given operating condition. That is, one needs twelve independ
ent equations to solve for twelve unknowns. This necessitates independent control
over rotational speed and vibration orbit frequency.

There are basically two experimental approaches one could take: (i) impose
dynamical forces and measure displacements, or (ii) impose dynamical displacements
and measure the forces. With currently available measurement techniques, the
second approach is potentially more accurate, and has been used in our design.
The test rig is configured around a double-spool spindle, with the inner spindle
having an adjustable run-out or eccentricity with the outer spindle (Figures 6 and
7). Rotation of the outer shaft therefore causes the rotational centerline of the
inside shaft to experience a circular orbit with a precession frequency of the
outside shaft1s rotational speed. The rotational speed of the inside shaft is the
test rotational speed. Independent control over vibration frequency and test speed
is therefore accomplished. The same approach is now being used (see Reference (5))
to experimentally determine the linearized spring and damping coefficients for low
specific speed centrifugal pump stages. As the shaft-to-shaft eccentricity is
adjusted, it can be measured with extreme accuracy using an LVDT or even a precision
dial indicator while slowly rotating the outer shaft by hand.

The test ring is rigidly supported in the radial plane by four piezoelectric
load cells (see Figures 6 and 7), two in each of the x and y mutually perpendicular
directions which allow variation of orbit-center eccentricity. The necessary
advantage of piezoelectric load cells is that they are extremely stiff and therefore
keep test ring vibration amplitudes negligible and thereforunnes,saryto mealJre.
Strafn gauge load cells would not be a feasible option here because they require
displacement to sense load.

The test ring is contained within a pressurized chamber. The test ring divides
the chamber into high and low pressure compartments. The difference in pressure
between these two compartments is controlled to the desired test pressure drop

150
across the fl id annulus, with a maximum axial pressure drop of through the
test annul The tes rl ng lS5UPporteaaxTaTT,yOilTTLiTd-fiTril-llydrostatTCtffFUSr
ces w ich introduce no extraneous radial loads. The test ring is attached to the
four load cells by four leaf springs which are soft in the circumferential direc
tion but stiff in the radial and axial direction. This type of construction allows
the fluid reaction torque on the test ring to be equi1ibrated by the testLiJl
, _-
su ort, without introducing extraneous radial forces on the te n The closure
head of tessurized chams easily removed as is-the test ring and test
journal. This provides for quick interchange of the various fluid annulus config
urations to be tested and modified.

It is essential that the radial run-out of the test journals which results from
inner shaft rotation be as close to zero as is possible to manufacture. This was
accomplished in the final machining operation on the test journals by grinding 'them
while they were rotated (inner shaft rotated, outer shaft fixed) in the final
assembled double-shaft spindle.

For any given annulus configuration (i.e., diameter, axial width, clearance
and surface geometry) the basic operating parameters are rotational speed, water
temperature, static eccentricity and axial pressure drop across the test annulus.
The effects of each of these parameters will be determined by varying them through
ranges encountered in actual feed pump applications.

CONCLUSIONS

The importance of inter-stage fluid annuli to rotor dynamical performance of

high-head feed water pumps warrants developmental efforts by the pump manufacturers
in this area. The test rig described in this paper has recently come on-line and
is just beginning to provide reliable data on inter-stage fluid-annulus dynamic
coefficients. Currently used configurations as well as newly devised high-damping
configurations are being tested. By the end of this year, we anticipate the comple
tion of the testing that is presently planned. We expect to be able to recommend
inter-stage clearance geometries which will provide considerable additional rotor
vibrational damping capacity for feed water pump and other multi-stage centrifugal
pump applications.

REFERENCES

l. Makay, E. and Szamody, 0., "Survey of Feed Pump Outages , EPRI Publication
II

FP-754, April 1978.

2. Adams, M. L. and Makay, E., "Development of Advanced Rotor-Bearing Systems for

Feed Water Pumps, Phase 211 , EPRI Publication CS-2027, September 1981.

3. Makay, E., "How Close Are Your Feed Pumps to Instability - Caused Disaster?",
Power, McGraw-Hill, December 1980.

4. Adams, M. L. and Padovan, J., "Insights into Linearized Rotor Dynamics",

Journal of Sound and Vibration, Vol. 76(1), May 1981.

5. Brennen, C. E., Acosta, A. J. and Caughey, 1. K., "A Test Program to Measure
Fluid Mechanical Whirl-Excitation Forces in Centrifugal Pumps", Proc. Texas
A&M Workshop. (May 1980) on Instability Problems in High-Performance Turbo
machinery, NASA Conference Publication 2133, pp 229-235.

151
 Appendix A

GOVERNING EQUATIONS WHICH RELATE EXPERIMENTAL MEASUREMENTS

TO FLUID-ANNULUS DYNAMIC COEFFICIENTS

A.l DEVELOPMENT OF GOVERNING FORCE DISPLACEMENT EQUATION

Postulating harmonic motion, equations for rotor system vibrations can be expressed
using phasor notation as follows.

i(nt+8x) i(nt+8y)
f =F e , f =F e
x x y y
i(nt t:@:x) . i(nt+8y)
x = xe y =Ye

. (A-l)
x =inx y =iny

For the general 1 inear case, the force-motion equations are given in expanded form
as follows.
. ..
f - - K x - B x - D x - K - B Y - D Y
x xx xx xx xyY xy xy
(A-2)
. .. . ..
f =- K x - B x - D x - K Y - B
yx yx yx yy y
yY - D
yy
Y
Y

Using harmonic motion, equations (A-l), the force-motion equations (A-2) then can
be simplified as follows.

f = (- K - inB + n2D )x + (- K - inB + n2D )y

x xx xx xx xy xy xy
(A-3)
f =(- K - inB + n2D )x + (- K - inB + n2D )y
y yx yx yx yy yy yy
-int
Using the phasor form, (A-l), and dividing through by e gives the fo11 owing.

i i
F e 8x =(- K - inB + n2D )Xe x +
x xx xx xx
i
(- K - inB +. n2D )Ye y
xy xy xy
(A-4)
i8y i
F e (- K - inB + n2D )Xe x +
y yx yx yx
=

i
(- K - inB T n2D )Ye y
yy yy yy
i
Recalling from basic phasor convention, e 8 =cos 8 + i sin 8. Therefore, equations
(A-4) take the following form.

152
 F (cos 8 + i sin 8 ) = (- K iQB + Q2D }X
x x x xx xx xx

(cos + i sin <1> ) ( K - iQB + Q2D }Y+:

<1> x x \::- xy xy xy
,-

-f;. (cos <1> + i sin <1> )

y y

F (cos 8 + i sin 8 ) = (- K - iQB + Q2D }X

y y y yx yx yx

(cos <1> + i sin <1> ) + (- K iQB + Q2D }Y

x x yy yy yy

(cos <1> + i sin <1> ) (A-5)

y y

The two complex equations of (A-5) can be segregated by real and imaginary parts
to obtain four real equations. Since there are twelve unknowns (i.e., the stiff
ness, damping and inertia coefficients), experimentally measured inputs to these
equations must be obtained at three different vibration frequencies (i.e.,
Qj, j = 1,2,3) for a given operating condition. This leads to the following
general form of the governing equations.

F cos 8 = [(QD - K ) cos <1> + B Q. sin <1> JX

x x J xx xx x xx J x

+ [(QD - K ) cos <1> + B Q. sin <1> JY

J xy xy Y xy J Y

F sin 8 = [(QD - K ) sin <1> - B Q. cos <1> JX

x x J xx xx x xx J x

+ j
[(Q DX
y
- K X ) sin <1> - BX Q cos <1> JY
Y y y j y

F cos 8 = [(QD - K ) cos <1> + B Q. sin <1> JX

y y J yx yx x yx J x

+ [(QD - K ) cos <1> + B Q. sin <1> JY

J yy yy y yy J Y

8 = [(QD - 0 ) sin <1> - B Q. cos <1> JX

y J yx yx x yx J x

+ j
[(Q D
yy
- K
yy
) sin <1> - B Q cos <1> }Y
y yy j y
(A-6)

where, j = 1,2,3.

As explained in. the main text of this paper, the test rig has been designed to
provide a contrdlled harmonic circular orbit of radius R. That is,

X = Y = R and <1> = <1> -

y x 2 _

Therefore, cos <1>y = sin <1>x and sin <1>y = - cos <1>x. Furthermore, all phase angles
can be referenced to the x-component of vibration (i.e., <1>x = o). Implementing
all these simplifications, reduces equations (A-6) to the following.

153
 " i//
( cos 8 )/R - K - Q.B QD
Fx x J xx /J',y
- ,
J xx
= +
XX "
"

"'\,
( sin 8 )/R Q.B - QD
J x6(
+ K
Fx x Q5j'----: xy J xy
-
=

(F cos 8 )/R - QD - B Q
y yx yy j
K
y J yx
= +

(F sin 8 )/R - Q.B - K - Q D j (A-7)

y y J yx yy yy
=

where, j = 1,2,3

A.2 EXPERIMENTAL ERROR CONSIDERATIONS

Equations (A-7) provide twelve equations in twelve unknowns. The orbit of vibra
tion, R, and its frequency are controlled by the design of the test rig. The
single-peak dynamic force amplitudes (Fx, Fy) and the associated dynamic force
phase angles (8 x, 8y) are measured. The remaining twelve unknowns are the dynamic
coefficients. Close examination of equations (A-7) will reveal that the first
two sets of equations are coupled only to each other, while the third and fourth
sets of equations are individually decoupled from the other equation sets. This
is a result of the basic approach of using a controlled circular orbit vibration.
An additional advantage is thereby provided since one does not haye Jo actuall,Y/
solve a single system of twelve equations. Instead, solution requires indiv;Clua1 '\
solution of Orle system of six equations and two systems of three equations. This
simplification tends to reduce the affinity for amplification of experimental
error in the equation solution step of the overall scheme for determining the
dynamic coefficients.

As fully shown, experimental data is required at three different frequencies for

a specific operating condition in order to recover the twelve dynamic coefficients
from the governing equations. Therefore, in theory, any three sufficiently differ
ent frequencies should yield the same dynamic coefficients for a given operating
condition. The experimental setup therefore inherently provides a means for
determining the overall inaccuracy of the experimentally determined coefficients.
That is, one can obtain data at say ten (or more) different vibration frequencies
and then determine the dynamic coefficients with the data from all unique combina
tions of three different frequencies out of the total of ten (or more) frequencies
available. The difference in answers between the different combinations is the
experimental spread or error. Experimental error will come from two basic sources:
(i) measurement inaccuracies, and (ii) non-linearity in the actual force-displace
ment phenomenon. These two sources of error will tend to work in opposition.
That is, measurement inaccuracies can be minimized by using 11arger" vibration
amplitudes, thus making the measured response forces larger and therefore easier
to measure accurately. However, the larger the vibration amplitude, the more
significant becomes the non-linearities, which are neglected in the governing
equations. Therefore, for each operating condition there will be an optimum
orbital vibration radius which minimizes the experimental error, being large
enough to obtain good measurements, but not too large to produce significant non-
1inearities. The test rig spindle ;s designed to that the orbital vibration radius
can be varied, specifically so that error can be minimized.

154
 AJLL LENGTH 132.5 N.
I'"

I. BEARING SPAN 110.0 IN.

, .
'. ".
Figure 1. - Twele-stage b oiler feed '":." .
.. . ...::
pump rotor; cane run no. 6 station. ., '
"10-
'
-
I " '';';''
' ''
:'",''1 X
Louisville gas and electric. .....:..:.: ..

.
-:':"
-
, '

_IOo11Wb' ---- , . ,:: ::,':'

MILS -
:..:.. -.:>.":
T.
''': -.!=-:'

~ NORMAL RINGS
li X =X SIN[f)t +BxJ
lj =YsIN[()jfeJ

69-.-.x
5-IOi
MILST

_ Figure 4. - Single-frequency harmonic
LOMAKIN EFFECT RNGS orbital vibration of rotor with
respect to stator.

Figure 2. - Currently used feed pump

, wear ring geometries, reference (3).

MUL TIS TAGE

10
BOIL ER FEED PUMP

8
<X

o 7
o
o

OPERATING SPEED
"
w 5
w
0.
WITHOUT
o
WEAR-RING
....
.. EF F E CT
(.)
;::
3 /'
2

a:
o I
....
o
a:

10 20 30 40 50 60 70

IMPEL LER WEAR-R ING CL E A RA NCE - MILS

Figure 5. - Trends of journal bearing

Figure 3. - Lomakin effect on first net damping per cycle of harmonic

'
critical speed of a typical feed motion. n/oo is vibration
water pump, reference (3). frequency-to-speed ratio.

155
 Legend:

1 - Test rotating element 7 - Inner spindle rotor

2 - Test annulus ring 8 - Outer spindle rotor
3 - Piezoelectric load cell 9 - Spindle housing
4 - Hydrostatic axial ring support 10- Support base
5 - High-pressure compartment 11 - V-belt pulley
6 - Low-pressure compartment 12 - V-belt pulley

Fi gure 6. - Conceptual sketch of double-spool spindle fluid-annulus test rig.

Figure 7. - Detail layout of double-spool spindle fluid-annulus test rig .

156
 TESTINJ OF TURBULENI' SFALS FOR ROIffiDYNAMIC CDEFFICIENI'S

Dara W. Childs
Department of Mechanical Engineering
Texas A&M Uhiversity
COllege Station, Texas 77843

John B. essman
Department of Mechanical Engineering
The Uhiversity of Louisville
Louisville, Kentucky 40208

SUMMARY

A facility has been developed for dynamic testing of straight and convergent
tapered seals with the capability of measuring the radial and tangential force
components which result from a circular centered orbit. The test apparatus causes
the seal journal to execute small-eccentricity centered circular orbits within its
clearance circle. Dynamic measurements are made and recorded of the seal
displacement-vector components, and of the pressure field. The pressure field is
integrated to yield seal tangential and radial reaction-force components.
Representative test data are provided and discussed for straight seals.

NOMENCLATURE

A amplitude of orbital motion (L)

C radial clearance (L)

r
C, c nondimensional, direct and cross-coupled damping coefficients,
introduced in Eq. (4)

C, c direct and cross-coupled damping coefficients, introduced in

Eq. (6), (FT/L)

H clearance function (L)

K, k nondimensional, direct and cross-coupled stiffness coefficients,
introduced in Eq. (4)

K, k direct and cross-coupled stiffness coefficients, introduced in

Eq. (6), (F IL)

L seal length (L)

M, m nondimensional, direct and cross-coupled added-mass coefficients,

introduced in Eq. (4)

M, m direct and cross-coupled added mass coefficients introduced in

Eq. (6), (M)

*The work reported herein was supported by NASA Lewis under NASA Grant 3200;
technical monitor, Dr. Robert C. Hendricks.

157
6.P seal differential pressure, (F/L2)

r r components of seal-displacement vector, introduced in Eq. (4), (L)

X' y

component of seal ,reaction force, introduced in Eq. (4), (F)

,
R seal radius, (L)

R =
2VC/v nominal, leal axial Reynolds number
a

R =
RoJC/v nominal, seal circumferential Reynolds number
c

T =
L/V transit time of fluid through seal, (T)

v nominal flow velocity of fluid through seal, (L/T)

A friction factor, defined in Eq. (3)

v fluid kinematic viscosity, (L2/T)

t;, entrance loss factor, introduced in Eq. (1)

(J friction loss factor defined in Eq. (2)

3 '
p fluid density (MIL )

W shaft angular velocity (T-1)

INTRODUCTION

Black [1,2,3,4], in a series of publications incorporating theoretical and

experimental results;, has demonstrated that the rotordynamic behavior of pumps is
critically dependent on forces developed by neck-rings and interstage seals
illustrated in Figure 1. Subsequent experience [5] has demonstrated that the
stability of cryogenic turbopumps is comparably dependent on seal forces. The
test program discussed here was stimulated by stability difficulties encountered
in developing the turbopump of [5], and has the objective of measuring radial and
tangential force components for straight and convergent-tapered seals over a range
of axial and circumferential Reynolds numbers.

Seal Analysis: Leakage and Dynamic Coefficients

Black [1,2,3,4] is largely responsible for developing models for constant

clearance seals. Black's analysis yields a definition of the force acting on a
rotor due to its motion at a seal location, and is based on the following leakage
relationship from Yamada [6] for flow between concentric rotating cylinders:

(1)

where t;, is a constant entry-loss coefficient, p is the fluid density, V is the

average fluid velocity, and (J is a friction-loss coefficient defined by

158
 cr = A/
L C (
2
)
r

In the above, L is the seal length, C is the radial clearance, and A has been
r

(R , R)
defined by Yamada to be the following function of the axial and circumferential
Reynolds numbers
a c

7R /
(8 C 2 3 8
= 0.079 R , R = 2 VC lv, R RwC lv, (3)
R )]
- +
,
A [l =

a a r c r
a

(
),
3
where V is the fluid's kinematic viscosity, R is the seal radius, and w is the

1/7 power
rotor's rotational speed. The friction law definition of Eq. Yamada's
definition for A, is based on an assumed velocity distribution, and fits
the Blasius equation for pipe friction.

{ } :rr {:: : :rr J {::} ::: ST

From a rotordynamic viewpoint, the objective of seal analysis is the defini

: :
tion of seal force coefficients, i.e., stiffness, damping, and added mass terms. In
nondimenSio a rm f o r i f m

where (K,

C, M) and

(k, c, m) are, respectively, the direct and cross-coupled

:} (4)

stiffness, damping, and added-mass terms.

Prior Analytical Results for Constant-Clearance Seals

Test results for the constant-clearance geometry seals are presented and dis
cussed here. Comparable results for convergent-tapered seal geometries will be
presented subsequently. The contents of this section are provided to briefly
review theoretical and experimental results and procedures for this seal
configuration.
Seal analyses which have been published to-date use bulk-flow models in which
the raial variation of the velocity and pressure fields across the fluid film are
neglected. Governing equations consist of axial and circumferential momentum
equations, and the continuity equation for the axial and circumferential velocity
components, U (z,e),U (Z,e),p(Z,e), which are averaged with respect to the radial
z e
coordinate, r. Most reported analyses of seals use a perturbation analysis in the
eccentricity ratio of the form

(5)

where H is the clearance function, and E is the eccentricity ratio. Governing

equations for PO' U ' UZO define the centered, zero-eccentricity flow field, while
eo
th first-order equations f r the variables U ' UZ ' PI def ne the flow fields
: e1 1
WhlCh result from small motlon about a centerea posItion deflned by the clearance
function H (t,e). A "short-seal" solution results if U ' "the pressure-induced
l el
circumferential flow, " is assumed to be negligible in comparison to the shear
induced term U Black [1] initially developed an analytic short-seal solution,
eO

159
but then with Jensen [2] developed a numerical finite-length solution. Correction
factors were developed by Black and Jensen from the finite-length solution to be
used in adjusting the short-seal solution to account for finite L/D ratios. Black's
second refinement of the original theory was to account for the influence of a
change in clearance on local Reynolds numbers [3].

Finally, Black et al. [4] examined the influence of inlet swirl on seal coef
ficients. In previous analyses, a fluid element entering a seal was assumed to
instantaneously achieve the half-speed tangential velocity, Ue = Rw/2. The results
in [4] demonstrate that a fluid element may travel a substantal distance along
the seal before asymptotically approaching this limiting velocity. For interstage
seals in which the inlet tangential velocity is negligible, the practical conse
quences of accounting for this "swirl" effect is a marked reduction in predictions
for the cross-coupled stiffness coefficient, k. Black's experimental results show
clear evidence of the influence of swirl. A short-seal analysis is used in [4],
and the perturbation in Reynolds number due to a local change in clearnace
introduced in [3] is not included.

One of the authors [7,8] has recently completed two analyses based on Hirs'
turbulent lubrication model [9]. Short and finite-length solutions are developed
in [7] and [8], respectively, including all of the various influences introduced in
Black's initial analyses [1-4]. The results resemble, but do not coincide with,
Black's.

Prior Seal Testing Procedures and Results

The pertinent data which must be measured to confirm the seal leakage model
of Eqs. (1) through (3) are AP, V, (from flow rate), w, and the axial pressure
gradientwf.thin the seal. This latter measurement yields (J which in turn yields A.
Yamada's model for the friction factor was based on testing for these variables
over the Reynolds number range (200<R <40,000; O<R <40,000) and clearance to
- c-
radius ratios of (.0106<C /R<.0129) a-
- r -

Various approaches can be taken to the measurement of rotordynamic coefficients

of Eq. (4). For example, if the journal segment of the seal is stationary (i.e.,
r ry rx ry 0), Eq. (4) can be inverted to obtain
X
= = = =

{::} -
- 2
K
2 e
RpflPK
e

Hence, by applying the static load definition (F F ' F 0), and measuring the
x s y
= =

displacement components, r and r , one obtains a combined measure of the direct

X y
and cross-coupled stiffness coefficients. This is predominantly the type of
testing performed by Black, who cites results in the form of "receptance
magnitudes," 1. e. ,

The relative magnitudes of the direct K and cross-coupled k stiffness coefficients

depend on the relative magnitudes of the axial and radial Reynolds numbers.

160
 Specifically, at zero running speeds, k is zero, but increases with R and can
c
exceed K.

Most of Black's testing [2,3,10,11] has been of the static nature cited above.
The second type of test cited consists of analytically modelling a test rotor in
cluding the theoretically predicted seal dynamics, and comparing the dynamic
characteristics of the model with test data. For example, in [2], the test rotor
was rapped and a correlation was made with the observed logarithmic decrement on
the decay curve. In [3], known imbalances were applied to the test rotor, and a
comparison was made with synchronous amplitudes and phase, critical speed location,
and onset speed of instability. Comparisons between rotor model results and tests,
of this nature, are helpful in deciding whether the general seal model is reason
able. However, this type of test-correlation yields limited specific information
about the individual dynamic coefficients. Further, discrepancies in synchronous
amplitude and phase results could result from an inadequate initial balance.

A summary of the test results of references [2,3,10,11] is provided in Table .

1
The correlation in these tests ranges from "good" to "fair." The nature and
results of the test support the following general conclusions concerning the
adequacy of Black's dynamic seal model:

(a) Over the Reynolds number range tested, the prediction of the direct
stiffness coefficient K is adequate for plain and serrated seals,
although less accurate for serrated seals. Black's test results
indicate a divergence between tests and theory for the direct damping
coefficient C as the axial Reynolds number is increased.
(b) Although the data cited generally supports Black's dynamic seal model
over the Reynolds number range considered, it is inadequate to speci
fically verify the proposed relationships [Eq. (4)] for the dynamic
coefficients as functions of the axial and radial Reynolds numbers.

THE UNIVERSITY OF LOUISVILLE DYNAMIC SEAL TEST PROGRAM

Test Section Design

Figure 2 illustrates the test-section design employed in the current seal test
program. Water enters the center of the section and flows axially across the two
rotating test seals exiting at the bottom of the test section. The seal journals
(L 4 in 10.16 cm, D 2 in 5.08 cm) are mounted eccentrically on the shaft
1.27 x 10 -4 m.
= = = =

with a constant eccentricity , A = .005 in= The nominal seal

clearance is C = .020 in = 5.08 mm, which yields C /R = .010. Accordingly, shaft
rotation cause the seal journals to execute circulr centered orbits at the
nominal eccentricity ratio s = 0.25. Axial and circumferential Reynolds numbers
may be specified over the range R s[5,000,30,000], R s[O,ll,OOO] by varying the
c
shaft rotational speed (0-4,000 rm) and flowrate. Shaft-speed is measured by a
once-per-revolution counter, while turbine flowmeters separately measure flowrate
through each seal.

The rotor of Figure 2 is supported in Torrington hollow roller bearings1[12].

These precision bearings are preloaded radially, have zero internal clearances, and
an accurately predictable radial stiffness. The first rotor-bearing critical speed

1These bearings were donated by Torrington through the kindness of W. L. Bowen, whose
assistance is gratefully acknowledged.

161
is predicted to be approximately 12,500 rpm. The end thrust bearing is provided
to react the small axial load developed by the opposed test-seal design.

Instrumentation and Data Analysis

The dynamic instrumentation illustrated in Figures 2 and 3 consists, for each

seal, of Bently eddy-current motion transducers and five piezo-electric pressure
transducers which are distributed both axially and circumferentially along and
around the seal. The circumferential "clocking" of the pressure transducers is
provided primarily as a matter of convenience, since the transducers are provided
exclusively to define the time history of the axial pressure distribution. Since
the seal journal is forced to execute a closed circular orbit at constant speed
w within its journal, the steady-state pressure distribution is constant with
respect to an observed fixed to the shaft, and the circumferential pressure distri
bution at time t, p(z,8) is definable in terms of either past or future time
measurements p(z,t) at a fixed value of 8. The direct extraction of circumferen
tial pressure distributions from pressure time histories also permits the
"reconstruction" of a pressure time history at a given circumferential location
(e.g., 8=0), despite the fact that the transducers are distributed circumferen
tially around the seal as illustrated in Figure 3.

The seal reaction forces at a given time, t = t, are defined by the integrals:

- . L
2rr 2rr

(t ) = f f p(8,z) sin 8 R d8dt = -RL f sin 8 p (8) d8

0 0 0
(6)
2rr L 2rr

Ry(t) = -f . f p(8,z) cos e R d8dz = -RL J cos 8 p (8) d8

0 0 0

where p is the average axial pressure defined by

L
p(8) =
1
-

L
f . p(8,z)dz.
o

The integral of Eq. (6) is evaluated numerically from pressure time histories
corresponding to 8 = 0 measurements, and denoted Pl*(8). The axial spacing of the
pressure transducers has been chosen from Gauss-Legendre quadrature formulas [13]
to minimize the error involved in evaluating this integral. The quadrature
formula for integration with respect to z is;

where

A = 0.23603, A = 0.47862, and A = 0.56889.

l 2 3

Data Analysis Procedure

The dimensional form of Eq. (4) is

162
 (7)

The nature of the test rig is such that steady synchronous motion of the form

r A cos wt , r =
A sin wt (8)
x y
=

results, and for this type of motion, the six coefficients of Eq. (7) can not be
separately identified. In fact, only two independent numbers result from the
steady-state, harmonic test data (r (t), r (t), Rv(t),
X y
(t which is generated Ry
by the test rig for a given axial and circumferential Reynolds number set (R ,R ).
c
For the present study, the two numbers chosen for presentation are the radia t
and circumferential components of the reaction force. From Eqs. (7) and (8),
these components may be stated

RIA
rr x R'I
2
= =

a
A
(9)

TEST RESULTS AND DISCUSSION

Test Results

Static test results for the seals consists of leakage. Dynamic test results
are the radial and circumferential force components, (R lA, RIA), of Eq. (9) as
a function of Reynolds numbers (R ,R ).
a c
Table 2 contai s bot static and dynamic ft ft
test results. With respect to static results, for 0.5, the coefficient =

C 1 + + 20'
d
=

of Eq. (1) is approximately -5% in error at the higher R range (R 24,000) and
a a
+3% in error at the lower R range (R 4,600). Hence, leakage is generally
a a
well predicted by Eq. (1).

An inspection of the Reynolds number sets of 'Table 2 demonstrates that R is

a
held nominally constant, while R is varied, with the result that the Reynolds
c
number pairs (R ,R ) are clustered about the nominal axial Reynolds numbers
[24,760; 18,8501 9;040; 4.580]. These results were obtained by holding the seal
leakage rate constant while varying the rotational speed w.

The dashed lines of Figures 4 and 5 connect the test data points of Table 2
for the +adial (R /A) and tangential force components (R /A) as a function of the
r a
circumferential Reynolds number, R . The test results reflect both the small
changes in R
a
about the nominal va ue R
a

and the obvious changes in R
c
resulting
from changes in w.

163
 Discussion of Radial-Force Component Results

Initial discussions of the test results will deal with results for radial
force components of Figure 4, and the comparison of these results to various
theoretical predictions of references [7] and [8]. The experimental and theo
retical results of interest are denoted as follows in Figure 4:

*---*---* Experimental results.

Finitelength solution, reference [8].

0__ 0 0
__

Short-seal solution, reference [8].

+-+-+ Short-seal solution, reference [7].

<1-0 - K from reference [8].

The theoretical results of reference [8] would, presumably, provide the best
prediction of the radial force component; however, the results of Figures 4(a)
through (e) show that: (1) measured radial forces are roughly twice as large as
predicted values, and (2) the magnitudes of measured forces tend to increase with
increasing R , while theoretical predictions show a decreasing magnitude.
c

As noted in the introduction, short-seal solutions are obtained by neglecting

the "pressure-induced" circumferential flow, while including the shear-induced
flow. Short-seal solutions yield larger radial force predictions, and, as
expected, reduce the discrepancy between theory and experiments of Figure (4).
The short-seal solution of reference [8] is developed under less restrictive
assumptions than that of reference [7], and agrees better with the test data.
The direct stiffness prediction of these two solutions are comparable; however,
reference [7] predicts values for M and c that are approximately twice as large
as those of reference [8]. The large value for M explains why this solution
diverges from both the experimental data and the remaining theoretical solutions
.with increasing R
c

In fact, all solutions predict magnitudes for M which are excessive in com
parison to either test results or experience with predictions of pumps. An
overprediction of M would explain why the theory predicts a reduction in R
with increasing R , while test results show the opposite. The short-seal heo
fi
retical results s ow that M and c arise from the same momentum terms in the
governing equations; hence, if the theory overpredicts M, one would assume that c
is comparably overpredicted. If both these terms are eliminated from Eq. (90),
only the direct stiffness K remains. This coefficient is illustrated in
Figures 4(a) through (e) for the short-seal solution of reference [8].

The coefficient K from the short-seal solution generally behaves the same
way as the test results for changes in R and R , and provides a much better pre
a
diction for the radial force component R than E
he other approaches of Figure 4.
However,
it underpredicts the test resul s on the average by about twenty percent.

Discrepancies between theory and experiment illustrated in Figures 4(a)

through (e) may be explained as a combination of the following factors:
(a) The theoretical developments are based on a perturbation analysis which
assumes that the seal-orbit amplitude A is small compared to the radial
clearance C , and that second-order terms in A/c are negligible. However,
r r
for the current apparatus, A/c 0. 2 5, and second-order effects may be
=

influencing test results. Thi may provide part of the explanation for the

164
 seals being stiffer than predicted.
(b) The fundamental bulk-flow model, as presented by the momentum and continuity
equations, is, in some sense, deficient. An explanation of this nature is
required to explain the apparent major overprediction of the direct added
mass and cross-coupled damping coefficients.

Discussion of Tangential Force Components

Theoretical results for the cross-coupled, added-mass coefficient m typically

indicate a negligible contribution to the tangential force component, i.e.,

Further, in the test apparatus, the shaft rotates in a clockwise sense and the
tangential force would be defined by

(10)

A comparison of the test results with the finite-length solution predictions

in Figures Sea) through (e) shows generally good agreement. If the test results
were curvefitted by a linear curve of the form predicted by Eq. (10), one would
reach the following conclusions:
(a) The cross-coupled coefficient k is smaller than predicted, and the difference
between theory and experiment increases with increasing R . This conclusion
is supported by noting that measured values for R /A are maller than
e
predicted at low values of R

C
(b) The damping coefficient C is slightly smaller than predicted. The generally
steeper slopes for the experimental data curve supports this conclusion.

The finite-length solution of [8] generally provides a better correlation

with experimental data than either of the short-seal solutions. All of these
solutions include the influence of "swirl." Specifically, in contrast to typical
hydrodynamic-bearing analyses in which the fluid is assumed to have an average
circumferential velocity of Rw/2, the acceleration of a fluid particle due to
shear forces is accounted for. For short-seals with high axial fluid velocities,
this means that a fluid particle may exit the seal without acquiring a substantial
tangential velocity. Since the test rig of Figure 2 provides negligible pre
rotation of fluid entering the seal, the theory assumes that the inlet circumfer
ential velocity is zero. Predictions of k are substantially reduced by including
the influence of swirl; hence, test results of Figure 5 provide an additional
verification of the influence of swirl.

SUMMARY AND CONCLUSIONS

The test results presented support the following general conclusions:

(a) Leakage is reasonably well predicted by Yamada's model.
(b) The circumferential force, which primarily represents the direct damping
coefficient C and cross-coupled stiffness coefficient k, is adequately
predicted by the finite-length theory of reference [8].
(c) The radial force is substantially underpredicted (40 to 50%) by the finite
length theory. The direct, added-mass coefficient M appears to be much
smaller than predicted by theory. By inference, the cross-coupled damping
coefficient c is probably also much smaller than predicted.
(d) The direct stiffness K from the short-seal solution of reference [8] has the

165
 same form as the measured radi.al force component but is approxi'll).ately 20%
less than measured values.

A conclusive explanation for the discrepancies between the theory and experi
mental results for the radial force component is not possible; however, it may
result from either (a) nonlinearities due to the size of the seal orbits relative
to the radial clearance, or (b) basic inadequacies in the bulk-flow model.

On the basis of the present results, the following reduced model for seal
reaction forces is recommended for modelling seals in rotordynamics analysis of
pumps:

Note that the added-mass and cross-coupled coefficients have been discarded. The
coefficients of k and C are adequately predicted by the finite-length solution of
reference [8]. The best prediction for K is provided by the short-seal solution
of reference [8]; however, this prediction is approximately 20% lower than test
results.

REFERENCES

1. Black, H. F., "effects of Hydraulic Forces in Annular Pressure Seals on the

Vibrations of Centrifugal Pump Rotors," J. M. Eng Sci., vol. 11, No. 2,
pp. 206-213, 1969.

2. Black, H. F. and Jensen, D. N., "Dynamic Hybrid Properties of Annular

pressure Seals," Proc. J. Mech. Engin., vol. 184, pp. 92-100, 1970.

3. Black, H. F. and Jensen, D. N., "Effects of High Pressure Ring Seals on

Pump Rotor Vibrations," ASME Paper No. 7l-WA/FF-38, 1971.

4. Black, H. F., Allaire, P.E., and L. E. Barrett, "The Effect of Inlet Flow Swirl
on the Dynamic Coefficients of High-Pressure Annular Clearance Seals, "Ninth
International Conference in Fluid Sealing, BHRA Fluids Engineering, Leeuwenhorst,
The Netherlands, April 1981.

5. Childs, D. W., "The Space Shuttle Main Engine High-Pressure Fuel Turbopump
Rotordynamic Instability Problem," ASME Transactions for Power, pp. 48-57,
January 1978.
.

6. Yamada, Y., "Resistance of Flow through Annulus with an Inner Totating

Cylinder, "Bu!. J.S.M.E., vol. 5, No. 18, pp. 302-310, 1962.

7. Childs, C. W., "Dynamic Analysis of Turbulent Annular Seals Based on Hirs'

Lubrication Equation," accepted for publication, ASME Trans-., J. of
Lubrication Technology.

8. Childs, D. W., "Finite-Length Solution for Turbulent Annular Seals, "

Accepted for publications, ASME Trans., J. of Lubrication Technology.

166
 9. Hirs, G. G., "A Bulk ... Flow Theory for Turbulence in Lubricant Films," ASME
J. Lubrication Technology, pp. 137-146, April 1973.

10. Black, H. F., "Empirical Treatment of Hydrodynamic Journal Bearing Perfor

mance in the Superlaminer Regime," J. Mech. Engine. Sci., vol. 12, no. 2,
pp. 116-122, 1970.

11. Black, H. F. and Cochrane, E. A., "Leakage and Hybrid Bearing Properties of
Serrated Seals in Centrifugal Pumps," Paper G5, 6th International Conference
on Fluid Sealing, February 27-March 2, 1973, Munich German Federal Republic.

12. Bowen, W. L. and Bhateje, R., "The Hollow Roller Bearing," ASME Paper No.
79-Lub-15, ASME-ASLE Lubrication Conference, Dayton, Ohio; 16-18 October 1979.

13. Froberg, Introduction to Numerical Analysis, Second Edition, Addison Wesley,

1969.

TABLE 1. - PRIOR DYNAMIC SEAL TESTS

REF. LID (C /R ) x 103 Seal type Cited Results

r

[2] [6,000, 14,000] [a, 3,500] .25 , .5, 1. 10.5 , 12.9 plain a, b, c

[3] [3,000, 20,000] [a, 6,000] .5, 1. 7.23, 10.3 plain d, e

[10] 10,000 [a, 8,000] 1. 10.7 plain f

[11] [3,000, 12,000] [a, 6,000] .232, .5, 1. 7.1 Serrated d

a. Static force-deflection curves for 0 0.8.

b. Resultant stiffness F /r for centered position and w = O.

s

c. Rotordynamic modelling to correlate with test data on the

direct damping coefficient c.

d. Receptance magnitudes for centered position.

e. Rotordynamic modelling to correlate with synchronous test results.

Correlation includes amplitude. phase. critical speed, and onset

speed of instability.

f. Receptance magnitude and phase at centered position.

167
 TABLE 2. - EXPER IMENTAL R ISULTS FO R A CONSTANT-CLEA R ANCE SEAL,
L = R = 2 in. (5.08 cm), Cr = 0.020 in. (0.50
" 8 mm)

M(psi) Leaka (J e w(rpm)

( gpm )
1 24,580 3,160 113. 62.0 1,149. -13,950. 3,507.

2 23,550 5,800 99. 58.5 2,105. -13,190 6,800.

3 26,780 7,950 121. 66.51 1,885. -16,150. 11,870.

4 25,330 8,840 113. 62.9 3,209. -15,500. 12,790.

5 23,545 10,360 106. 58.5 3,762. -15,090. 16,540.

6 18,140 2,878 62.0 45.1 1,054. - 7,886. 2,344.

7 19,990 5,860 65.0 47.1 2,178. - 8,507. 5,821-

8 18,730 8,260 65.0 46.5 3,000. - 8,790. 9,250.

9 18,560 10,660 71.0 46.1 3,871. -11,120. 12,750.

10 13,330 3,040 34.0 33.1 1,103. - 4,125. 2,503 :

11 13,510' 6,360 36.0 33.6 2,308. - 5,014. 4,360.

12 13,060 8,880 35.0 32.4 3,226. - 6,030' &,802.

13 13,610 9,860 41.0 33.8 3,582. - 7,430 9,170.

14 9,010 2,9.20 17.0 22.4 1,060. - 2,276. 1,597.

15 8,850 5,840 19.0 22.0 2,120. - 2,240. 3,210.

16 9,260 8,610 21.0 23.0 3,125. - 2,095. 5,715.

17 4,750 2,700 5.0 11.8 980.4 670. 812.

18 4,410 4,920 3.0 11.0 1,786. 645. 1,475.

19 4,750 8,220 4.0 11.8 2,985. 578. 2,763.

20 4,410 10,630 5.0 11.0 3,859. 417. 3,492.

NECK RING NECK RING

SEAL SEAL

J
,
----'1.----, - - ----
,

INTERSTAGE SEAL

Figure 1. - Neck-ring and interstage seals for pumps.

168
Figure 2. - Test-section lyout.

Itl
l +1 t +
I
'Iz.tl

o
Figure 3. - Instrumentation.

169
 lJNVIl'Y 0' L0UI3V[LL, STRAIGHT SEAL DATA
UNIVERSITY OF LOUISVILLE STRAIGHT SEAL DATA
NOM[NA AX[A EYNODS NUMBER - 4,580.
NOMINAL AXIA REYNODS NUMBER 18,850.
a.
1 800 .
-1121121121.
16121121.

14121121. -2121121121.

12121121.
Z -51211210.

,Xi 1121121121. "

_1 III -41271270.
81210. ..l

-51271210.
60121.
<T
4121121. " -612100.
--1
<r J
21210.
<T -700121.

0. 0
<T -801211!J.
- - -

--
,213121. - -

- . "
*- -
- - _ ....
-
U. -90210.
-412J1ZI.

-612H!'. -11210121121.
- - - - - - - - *- - - -
*- -

-8121'"
-11121"'0. '.*

-1121121121. I
2121121121. 3121121121 4121121121 -121210121
2'1210121.
50122121 6121121121 7121121121 81210121. 9121121121. 1121121121121. 110121121.
25121121. 35121121. 4SI21I2J. S512112J. 65121121. 7512112J. 85121121. 95121121. 11215121121. !i0121121. 4121121121. 500121. 612100. 712100. 8121121121. 901210. 1I2112'0 0. IUI"I2I.
25121121 35121121. 45121121. 5500. 650". 76121121. 85121121. 950121. t 1215121121
C[RCUMFERENTIA REYNoDS NUMBER
CIRCUMFERENTIAL REYNOLDS NUMBER

UNIVERSITY OF LoUISVILE STRAIGHT SEAL DATA

UNIVERSITY OF LOUISV[LE STRAIGHT SEAL DATA
NOMINAL AXIAL REYNOLDS NUMBER ' Q.040.
NOMINAL AXIA REYNOLDS NUMBER - 24,760.
a.
0. -

-21210.Ii:n ,11211210.
-412111LI2II -.2121121121.

-6121121.1211 -31211210.

Z -4121121121.

-8121121.
" "
III
_1
-112112112'1.
'"_1 -Saa0.

-12121121.
-7121130. ' _
_ _

::: 1-----
-'" ---.
iT
"
14121121.
------

iT
-16121121.
J
<T
-1800. --
0
<T -2121'11121.

U.
-22"'12J.

-2400.

-26"'121.

-28121121.

-3121121121
2'121121121.
.- - - - - - - - - - - - - -*- - -

,l
600121
t

0

..
..
.... * - -
-

.5121121121. 4121121121. 5121121121. 6121121121. 7121121121. 8"00. 912100.

.1712J12I il2ll2lrzl 4rz1121 0--: 71211210. 8"121121. 9121121121. 1 f2!"I21I21. 1 t 1210121.

6121,,121.
1210.
251210. 351210. 451210. 55121121. 661210. 75"". 85121121. 6rz10".
36121a. 45 55121" 65,,121 75121121. 85/21121. 95210. t ,,521121
CIRCUMFERENTIAL REYNOLDS NUMBER
CIRCUMFERENTIA REYNOOS NUMBER

UNIVERSITY OF LOUISVILE STRAIGHT SEAL DATA

NOMINAL AXIAL REYNOLDS NUMBER ' 15,5B0.
a.

-SBI2J.12I1

-1121121li1"

-ISI21I2J.

Z -2IiU!lI2l.

" -25121121.
III
J -:51212121.

-3521121.
<T
" -412J12I2I.
*
-

J -45121121. -
- - --
<T -
- -
-- .* .....
-51211210.
0
<T
-66121121.
.....
-
U. -6121121121. *
,

,,
,

-65121121.
,
,
,
-71211210.
'*
-75121121.

-Sl2Il2Il2Iil2l"". 4""121. 5121121121. 6121121121. 71210121. 8121121121. 90121121. HJ ;12I0.

35121121. 45121121. 55121121. 650121. 75121121. 85121121. 95121121.

CIRCUMFERENTIAL REYNOLDS NUMBER

Figure 4. Seal radial-force component versus axial and circumferential

Reynolds numbers.

170
 UNIVERSITY OF LOUISVILLE STRAIGHT SEAL DATA
UNIVERSITY OF LOUISVILLE STRAIGHT SEAL DATA
NOMINAL AXIAL REYNOLDS NUMBER 4,5B0. NOMINAL AXIAL REYNOLDS NUM8ER 18,850.
41211210. 16121121121.

:375121. 151211Z1121.

:3500. 141Z11211Z1.

Z :3250. Z 131Z11Z1121.
H
..... 3121121121. ..... 121Z11Z1121
m m
--1 2750. --1 1112HZIIZI.

25121121. 112121021.
a: a:
..... 225121 ..... QI2l0"'

--1 21211210. --1 800",.

a: a:
175121. 71211210.
I- I-
Z 15121121. Z
W W 6121"'121.
(') (')
Z 1260. Z 5121121121.
a: a:
I- 11210121. I- 41210121.
I.. 760. I.. 31211211Z1.
5""21. 21211Z1121.
25121. llZ1121l21.

121. a.
='-
1 1_-.
2121121121 31210121 41Z1121121. 51211210 6f" 700121 8121"" . 91Z1012J. 11210121121. 1 101210. 21211Z1
2500. 351210. 4500. 551210. 1ZI 6500. 751210. 851Z1t21. 951Z1121. 10512J12J. 50:0 51Z1:1ZI 5121: 12I 51Z1: 12I 5012l 512l : 12I 5121; 12I 5IZ1 b0 I2lI2l.
5 JJ
CIRCUMFERENTIAL REYNOLDS NUMBER CIRCUMFERENTIAL REYNOLDS NUMBER

UNIVERSITY OF LOUISVILLE STRAIGHT SEAL DATA UNIVERSITY OF LOUISVILLE STRAIGHT SEAL DATA
NOMINAL AXIAL REYNOLDS NUMBER 9,040. NOMINAL AXIAL REYNOLDS NUMBER 24,760.
61Z112l12J. lQI2lI2J13.
18131210.
55121121.
17121121121.
501211Z1. 161211210.
Z
H
"- 45121121.
15121121121.
m 00 14121121121.
_1 -' 1300121.
41210121.
121211210.
a: :350121. a: 1100121.
"- or "-
11211211210.
/
-' 30'2HZ!.

-
-'
a: a: Q0121121.
I- 2500. I- 8121121121.
Z Z //
W W 7000. /"
(') 200121.
Z
(')
z 612100. /
a:
I- 151Z10. a: 51210121.

L
l-
I.. 41211211Z1. * ... ... ... ... ... ... ..
1121121121. I..
51211210.

51210. 20121121.
11211Z10.
tZl 0
2121121121 3121121121. 41210121 512100. 6000. 7121121121. 81210121 aa. --9
= .--ea .-+Iaaaa .
a 0
a
,
000.
q""e. i00121 - a
4-0 a-
. --6
. 5-000 a .-+ 700a
0
0 11
251211Z1. 561Z11Z1. 451Z11Z1. 55121121. 65121121. 750121. 85"121. 35e1121. 45e10. 5500. 6500. 750e1. 85e11Z1. 951210. 1121500.
CIRCUMFERENTIAL REYNOLDS NUMBER CIRCUMFERENTIAL REYNOLDS NUMBER

UNIVERSITY OF LOUISVILLE STRAIGHT SEAL DATA

NOMINAL AXIAL REYNOLDS NUMBER 15,580.
11121121121.

11211211Z1121.

z 9012JI2I.

.....
m 8'11210.
-'
71Z1l2l2l.
a:
.....
-'
a: 5211211Z1.
I
Z
W 4121121121.
(')
z
a:
-

21211Z1121.

100121.

l2I
il2l121l2J _
41211210. ----oo.___60121121. 701Z1121. 81211Z11Z1. 912101Z1. 11Z101210.
351210. 4512J0. 55121121. 651Z10. 750121. 8512121. 951210.

S IRC-llMFERENTIAL REYNODS NUMBER

Figur,e 5. Seal tangential-force component

versus axial and Circumferential
Reynolds numbers.

171
 A HIGI-REYN:)LDS-NlMBER SEAL TEST FACILI'IY: FACILI'IY DE:3CRIPTlOO

AND PRELIMINARY TEST n\TA*

Dara W. Childs, Clayton Nelson, and Ted Noyes

partment of M:!chanical Engineering
Texas Jl&M Un iversity .
College Station, Texas 77843

John B. Dressman
partment of M:!chanical Engineering
The University of Louisville
Louisville, Kentucky 40229

ABSTRACT

A facility has been developed for testing the leakage and rotordynamic
characteristics of interstage-seal configurations for the HPFTP (High Pressure
Fuel Turbopurnp) of the SSME (Space Shuttle Main Engine). Axial Reynolds numbers
on the order of 400,000 are realized in the test facility by using a Dupont freon
fluid called Halon (CBrF )' The kinematic viscosity of Halon is of the same order
3
as the liquid hydrogen used in the HPFTP. Initial testing has focused on the cur
rent flight configurations (a three-segment, stepped unit) and a convergent-taper
candidate.

INTRODUCTION

The HPFI'P develop:nent program experienced considerable difficulty with sub

synchronous, unstable vibration problems [1]. As originally designed, the HPFI'P
was violently and destructively unstable at running speeds above approximately
20,000 rpm. The remedy for this situation consisted of the following steps:
(a) the original soft bearing carriers were replaced by stiff bearing carriers, and
(b) the grooving was rerroved from the original stepped interstage seal design.
With these m:xlifications, the HPFI'P has been stable over its design speed range.
The essential contribution of the seal m:xlification to the elimination of the rotor
instability was underlined by tests in 1979 on an HPFTP unit which whirled enthu
siastically after the interstage seal clearances were deliberately enlarged.

Fnom a rotordynamics viewpoint, seal analysis has the objective of predicting

the influence-coefficients of the following equations:

(1)

*The work reported herein was supported by NASA Contract NAS8-33176 from
George C. Marshall Space Flight Center, Alabama 35812; Contract Monitor:
Frank Garcia.

172
as a function of seal operating conditions. Initial analyses and testing for inter
stage and wearing seals were published by Black et al. [2-5]. Black used the
following leakage relationship from Yamada [6] :

6P = [1+s+2a] (2)

where s is a constant entry-loss coefficient, V is the average fluid velocity, and

a is the friction-loss coefficient defined by;

AL
a = -C (3)
r
In this relationship, L is the seal length, C is the radial clearance, and A is
the following function of the axial and circufuEerential Reynolds numbers (R ,R ) :
a c

(4)
R = 2VC lv, R = RuC lv
a r c r

where 'J is the fluid I s kinematic viscosity, R is the seal radius, and w is the
rotor's rotational speed.

Childs has subsequently developed analyses [7,8] for constant clearance seals
based on Hirs' turb ulent lubrication equations [9] . These results are similar to
Black IS, but differ in some respects. Their principal advantage is in a direct
derivation from a single set of governing equations.

Fleming [10] proposed that the direct stiffness of seals could be increased
by either (a) introducing a step in a constant-clearance seal, or (b) using a
convergent-taper geometry. Childs [11] completed an analysis for convergent-tapered
seals which defined all of the dynamic coefficients, and derronstrated that the
seal taper significantly reduces the seal damping and cross-coupled stiffness
teITIlS, while increasing the direct stiffness. Hendricks [12] found in testing
straight and convergent tapered seals at large eccentricities in liquid hydrogen
that a slight taper reduced local seal cavitation at the seal exit, and eliminated
the possibility of a local negative stiffness developing.

Black's analysis was used initially in predicting the rotordynamic coefficients

of the HPFI'P interstage seals [1] . His experimental results were for plain
(constant-clearance) seals having srrooth and serrated surfaces, and extended out
to axial Reynolds numbers on the order of 20,000, as compared to 500,000 for the
HPFI'P. Given that the HPFTP seals differ substantially. from those tested by Black
in both configuration and operating conditions, the decision was made to implement
a test program to directly measure the forces developed by seals, which are geo
metrically similar to the HPFTP seals and operating at the same Reynolds number
conditions. The development of a facility to make these measurements together
with a discussion of some preliminary results is the subject of this report.

173
 FACILITY DESCRIPTION

Test Section

The test section design is an extension of an earlier design used in a

University of Louisville test program [13], and is illustrated in Figure 1. Fluid
enters in the center and discharges axially across the two test seals. The seals
on the right and left illustrate, respectively, the current stepped configuration
and a proposed convergent-tapered configuration for the HPFI'P interstage seals.
The current test section differs from that enployed earlier [13] in that it
assembles in a mid-plane position, which pennits testing of the "slingers" shown
in the figure.

The test-section rotor is supported in Torrington hollow-roller bearings* [14),

which are extremely precise, radially preloaded, and have a predictable and re:peat
able radial stiffness. The test seals are nounted eccentrically to the test
rotor; hence, rotor rotation develops a rotating pressure field that is constant
with res:pect to a rotor-fixed observer. Axially-spaced kulite strain-gauge pressure
transducers are used to measure the pressure field which is developed by seal
rotation. Seal notion is measured by MrI capacitance probes in the vertical and
horizontal planes. Inlet and discharge measurerre:nts are made of tenperature and
pressure for use in defining the density and viscosity of the test fluid.

A 10-hp variable-speed electric motor is used to drive the tes t section fran
five rpn to 5300 rpn. A notion transducer, actuated by a rotating ten.:...toothed wheel,
provides a signal for a counter which defines the rpn of the test section.

John Dres sman of the University of Louisville designed the test section and
Supervised its construction.

Test-Fluid Selection

The very high Reynolds numbers encountered in the HPFI'P result from (a) the
extremely low kinema.tic viscosity of liquid hydrogen, (b) the high-shaft rotational
speed, and (c) the very large t:,p' s developed by the HPFI'P stages. The average
properties of hydrogen within the HPFI'P interstage seals are

-5
= 1.16 10 Ns/rn2, p = 1. kg/rn2
7 2
x

-
v = 1.63 x 10 7 rn2/sec.

For corrpa.rison, the nominal pro:perties of water at 25C are

-4 3
= 8.9 x 10 Ns/rn2, p = 1000 kg/rn

-
v = 8.9 x 10 7 rn2/sec.

Since the Reynolds numbers are inversely proportional to v, an ideal test fluid
should have a low absolute viscosity and a high density.

* These bearings were made available from Torrington, Incorporated through

the courtesy of W. L. Bowen.

174
 The test fluid selected to rreet these criteria is broontrifluo thane,
CBrF ' which is manufactured as a fire extinguisher fluid (Dupont FE 1301 or Halon)
and refrigerant (Freon Its fluid properties at
3
l3B.I). 25C are [15,16]

Ns/m2, p 1570 kg/m3,

-4
= 1.54 x 10 =

-7
v = 1.0 x 10 m2/s.

This liquid actually has a lower kinematic vis<x>sity than liquid hydrogen, and has
the additional advantage of being nonflarrmable and nontoxic. The vap:>r pressure
of Halon is approximately 200 psi at room temperature.

Seal Geometry

As illustrated in Figure 1, the current flight gearetry uses three constant

clearance seal segmants separated by two steps. The d:i.m:m.sions of the test seal
segments are

L. (em) R. (em) C .
1. ... 1. r1.
(nm)
1.4986 5.0292 0.5080
1.5265 4.9682 0.5080
1.8034 4.9098 0.4826

The d:i.m:m.sions of the tapered seal are

L = 4.996 em, R = 5.0292,

C C . 0.45 72 rom
ro r1.
= 0.5715 nm, =

where C and C . are the entrance and exit clearances, respectively. These
1.
d:i.m:m.siiSRs yiela the following definition for the tapered parcureter:

(Crori) / (CroiCri) 0.111.

q
= =

Relatively speaking, this is a slight taper which should [11] slightly increase
K, while slightly decreasing the remaining coefficients.

Mechanical System Layout

Figure 2 illustrates the flow loop used to provide specified flowrates through
the test-section seals. A six-stage Goulds punp provides the flowrate required.
Loop flow rate discharges fran the pump, and then may split with part of the flow
going through the test section, and the remainder proceeding through control
valves 1 and 3. Two control valves in series are required to absorb the full
output pressure of the purrp without cavitation. The bypass flow node is used for
total test-section flowrate less than 100 gpTI., which represents the lowest flowrate
operating p:>int for the pump . Valve 4 is closed in the l;>ypass node. As the
required flowrate increases above 100 gpTI., valve 1 is closed and valves 2 and 3 are
progressively opened. Valve 4, which has a larger capacity than the remaining
valves, is opened to achieve ma.xi:rm.lm flow conditions. Flowrate through the test
section seals is neasured by Fischer-Porter vortex flowmeters.

175
 The test-section fluid is circulated through a heat exchanger, which is
supplied chilled water by a Trane chiller. The chiller capacity is augmented by a
2000 gallon water tank which is buried outside the test facility.

The ambient system pressure is maintained by the accumulators illustrated in

Figure 2. They are also used to retIDve liquid Halon from the test sec-tion for
replacerrent of rotors. Similar accumulators are used to transfer fluid back and
forth from the 2000 lb Dupont delivery tank. In fact, all of the Halon can be
pumped back into the delivery tank using the accumulators.

The filters illustrated in Figure 2 have a ten micron limit for particles. The
complete flow system is stainless steel except for the pump body and the heat
exchanger so that particle contamination in the test fluid is minimized.

Control System

The axial and circumferential Reynolds numbers are the quantities to be con
trolled in the seals. These variables are determined by the pressure and tempera
ture measurerrents within the seal (which define p and ]1), the seal rotational
speed w, and the flowrates through the seals. Control is supplied by means of a
Data General Nova computer. Control signals are generated, based upon the dif
ference between a measured R and a specified R , and cause a change in the
a
Masoniellan control values. Active control is flot entirely closed-loop_ The
operator specifies the number of control cycles to be evaluated by the computer.
The computer calculates the running speed that is required to achieve a specified
R , and the specified test section speed is set manually.
c

DATA ACQUISITION AND ANALYSIS

Data Acquisition System

The transient test data to be recorded consists of the seal rrotion r (t) ,
X
r (t) and the pressure measurements p. (t). Five pressure measurements are made for
tl\etapered seal, with their spacing pecified by a Gauss-Legendre quadrature
fonnula [13]. Four additional pressure measurements are made on the stepped seals
to measure the pressure immediately upstream and downstream of the steps. These
data are recorded using a 17-channel Physical Data, Inc. recorder. Egch channel of
this unit has its own A/D converter with a sampling rate up to 2 x 10 Hz, and
4096 storage locations.

The physical data system is controlled from the D. G. Nova computer, which
enables the recorder, comnands data capture, and then controls dumping of the data
into disk menory of the Nova. Transient data for control of the system is obtained
directly through AID units on the Nova.

Data Analysis

As outlined in [13], transient data analysis consists of alignment of the

pressure signals (which are measured at various circumferential locations), and
integrating the pressure signals to obtain the seal force components. The test
rig causes a synchronous precession of the seals, and the resultant rrotion may be
stated:

176
 r = A cos w t, r = A sin w.t (5)
x y

The six coefficients of Eq. (1) can not be separately identified for this type of
rrotion. In fact, if Eq. (5) is substituted into Eq. (1), and the coefficients of
sin wt and cos wt. on both sides of Eq. (1) are e:::ruated, only two independent
nmnbers can be identified. For the present study, the two nmnbers chosen for
presentation are the radial and circumferential components of the reaction force.
From Equations (1) and (5) these components ma.y be stated:

R'/A -
-
IrxRl =k-Cw-mw
2
e 2
A
(6)

R /A =
I rRI Mw
2
- CW - K .
r 2
=

Since the coefficients depend on cr, which in turn depends on R = Ru lv, one can
not generally vary w to obtain additional independent equation for solution of
the coefficients. Stated differently, changing w also changes the coefficients.
An exception to this rule is provided by very high values of R for which cr is
a
insensitive to changes in w.

TEST SERIES

Three rotors were initially manufactured for use in testing the general seal
configurations of Figure 1, viz'" the current flight configuration and a tapered
seal alternative. These rotors differ only in the nominal seal eccentricity;
specifically, they have the following eccentrities and eccentricity ratios:

(a) Rotor 1, A = .005 in, A/C = 0.25,

r
( b) Rotor 2, A = .010 in, A/C 0.50, and
r
=

(c) Rotor 3, A = 0.0 in, A/C 0.0.

r
=

The zero e8centricity rotor provides static leakage and pressure gradients only.
Rotors 1 and 2 provide insight as to the influence of seal dynamic eccentricity on
test results. 'rhe above rotors have been tested over the following nominal axial
and circumferential Reynolds nmnber ranges:
R : 50,000, 75,000, 150,000, 325,000, 400,000
a

R : 40,000, 65,000, 90,000, 115,000, 140,000

c

tIDre specifically, a matrix of tests has been carried out such that all of the
five values of R are obtained for each value of R
c a

Theoretical results [5,7,8] have derronstrated that the cross-coupling tenns

of Eq. (1), i.e., k, c, m, depend on the fluid tangential velocity entering the
seal. The standard test rig provides minimal pre-rotation of the fluid entering
the seal. Hence, rotor 2 was nofified by shrinking on an active "slinger" at the
entrance to both seals to provide significant fluid pre-rotation. Slingers
incorporated radial slots and an axial clearance of 0.010 in. for a radial distance

177
of O. 75 inches prior to seal entry. The power requirerrents to rotate this rotor
are considerably higher than the other rotors, and data can only be taken for axial
Reynolds numbers out to 150,000.

Two additional rotors have been manufactured to examine the influence of

surface roughness on both sealing capability and the seal reaction-force components.
These rotors have an "a1m:::>st" circumferential grooving pattern with a nominal depth
of 0.001 and 0.002 inches, respectively. The original, zero eccentric rotor, was
machined to yield an 0.002 inch surface roughness pattern. All of these rotors have
been tested over the COI'fPlete Reynolds number range.

Additional planned tests include machining surface roughness into the housing,
with the first and second machining operations yielding depths of 0.001 and 0.002
inches, respectively. The following rotors will be tested in these two surface
roughened housings:

(a) Rotor 1 (srrooth); A = 0.005 in.,

(b) Rotor 4 (0.001 inches depth); A = 0.003 in.,
(c) Rotor 5 (0.002 inches depth); A = 0.005 in., and
(d) Rotor 3 (0.002 inches depth); A = 0.0 in.

Testing of helically grooved seals is proposed for the corning year which will
include a systema.tic variation in taper angles, nominal clearances, and groove
depths. Initial testing for these seals will be at zero eccentricities.

PRELIMINARY RESULTS
Reduction of the test data is still in a preliminary stage; hence, the results
which are available for discussion are limited at present to sane static data
( - leakage and pressure gradients) and force magnitude results for the two seals
of rotor 1.

Static Results
-
Static results were obtained for the stepped and tapered seals of rotor 3, viz.,
the zero-eccentricity rotor. Figures 3 and 4 illustrate the steady-state pressure
gradients for the stepped and tapered seals, respectively, for a range of axial and
circumferential Reynolds numbers. Each frame of these figure sets illustrates the
influence of changes in R with Ra held constant. The strain-gauge pressure trans
ducers used to obtain thee results have an accuracy specifications of .1% of full
scale, which arrounts to approximately 1 psi. This lack of accuracy is much rrore
noticeable at lCM flCM rates and shallCM pressure gradients than at high flCM rates.
'lb sane extent, this explains the rrore "ragged" nature of the pressure gradients in
the low flow conditions of figures 3 and 4 as canpared to high flow conditions.
Each data point in these figures corresponds to an average of 200 samples.
The pressure gradients of figures 3 and 4 are seen to be rrore sensitive to
changes in % at low values of , Pecoming relatively independent of Ry... at higher
values of Ra' This result would"1>e expected, since from Eqs. (2) and (j)

30
' 045
(L)R -9/4[1 + (7%)2]-5/8
8Ra '
=

3R C a
c

which shCMs a marked drop in sensitivity with increasing R '

a

178
 An apriori analysis of the stepped seal for leakage would use Eq. (1) for each
step yielding the results

pV?
1
6Pi = ); i = 1, 2, 3
(1 + i + 20i

3
/:"p = L: /:"p.1
.
1=1

where Ai is the area of each annulus. Nor:ma.lly, one assumes that the entrance loss
factors are constant, indepeQdt of Ra and %' curl range between 0.1 and 0.5. At
the inlet, the te:rm (1 + represents a velocity head loss due to fluid accel
d
eration plus an entrance loss representai by i . The following numbers for (1 + )
1
were obtainErl fran the data of Figure 3.

Stepped Seal. Vata

(1 + ) (1 + ) (1 + 3)
Ra 1 2

382,000 1.129 .435 .484

323,000 1.129 .432 .531

150,000 1.255 .432 .371

75,000 1.681 .537 -.178

50,000 1.913 .916 .999

'!hese results are fairly regular and constant over the higher Reynolds number
sets, but tend to be erratic over the two lower Reynolds number sets. '!he results
for Ra = 75,000 curl 50,000 should be viewai with sane skepticism, because of the pre
viously mentioned accuracy of the pressure transducers. '!he rrost unexpected result
fran Figure 3 is the relatively small pressure gradient across step 1, the entry
step, as canpared to the rema.ining two steps. Figures 3 (a) curl 3 (b) shov.r a small
pressure rise imnediately before the first step, while figure 3 (c) shOlrlS a flat
pressure gradient across the first! step. A zero pressure gradient, a = 0, would
yield a praiiction of zero stiffness contribution from the first step.
For a tapered seal, the steady-state leakage relationship is [llJ

for small q. '!he experimental results from figures 4 (a) - (c) yield the follov.ring
entrance-loss result:

179
 1 + s

434,000 0.926

325,000 0.996

150,000 0 878
.

While these losses are constant, independent of %'

this is not the case in figures
4 (d) and 4 (e) where the losses increase steadily with Re. In fact, an inlet pres
sure increase was measured for %
16,890 and %
= 49,981. The tapered-seal re
=

sults above for high Reynolds numbers are consistent with Fleming's assumption lO[ J
of zero s.

Dynamic Results
Only a very llinited number of dynamic data sets have been reduced. The only
immediate conclusion which has been drawn fram these results is that the force mag
nitude R t I = + (Rii Ri)
is canparable for both the current stepped seal configur
ation and the proposed tapered seal. A canparison of their radial and tangential
forces and a ccrnparison to theory will be forthcoming shortly.

CLOSING STATEMENTS
A test facility has been developed which has the capability for testing annular
seals at Reynolds numbers which are canparable to those developed in cryogenic turbo
pumps
Generally speaking, the facility rks the way it was expected to and is
beginning to yield the types of static and dynamic test data th at are required to
. support theoretical predictions of seal rotordynamic coefficients and leakage.

180
 REFERENCES

1. D W. Childs, "'the Space S'liuttle Ma i n Engine High P re ssur e Fuel

.

'turbopump Rotordynamic Instability Prob lem, " ASME 't ra n J of

E n gin ee ri ng for Power,pp. 48-57, Jan. 1978

2. H. F .Black, "Effects of Hydraulic Fo rces in Annu lar Pressure Seals

on the Vibrations of Centrifugal Pump Rotors," J.ri.Eng. Sci.!
Vol.ll, no.2, pp.206-2l3,1969

3. H.F.Black and D.N.Jensen , "Dynamic Hybrid Properties of Annular

Pressure Seals, " Proc. J.MECH. Engin., Vol. 184 , pp. 92 - l00 , 19 70

4. H.F. B lac k and D.N.Jens en, "Effects of High Pressure Ring Seals
on Pump Rotor Vibrations,"ASME Paper No. 7l-WA/ff-38, 1971.

s. H.F.Black, P.Allaire, and L. Barrett, "'the Effect of Inlet Flow

Swirl on the Dynamic Coefficient o f Hi gh-P r essu re Annular Seals,"
Ninth International Conference in Fluid S ea l in g, BHRA Fluid Engineer
E ngi n eeri ng, Leeuwenhur st, 'the Netherlands, April, 1981.

6. Y. Yamada, "Resistanc e of Flow through Annulus with an Inner Ro ta ti ng

"
C ylind er , B u ! .J .S .M.E ., Vol. 5 , no.18 , pp. 302-310,1962.

7. D.W. C hilds , "Dynamic Analysis of 'turbulent Annular Seals Based on

Hirs' Lubricati:on Equa1:ions,'i accepted for p ublic ati on , --
ASME
'tr ans:J. of Lubrication T echnology .

8. D.W.Childs,"Finite-Length Solutions for Rotordynamic Coefficie nts

of Turbulent Annular Seals," accepted for publication, ASME Trans. J.
of Lubrication TechnologY.

9. G.G.Hirs, "A bulk-Flow 'th eory for 'turbulence in Lubricant Films,"

ASME J. Lubrication Technology. pp. 137-146, Apr il 1973.

10. D.P.Fleming, "High Stiffness Seals for Roter Critical S p eed ' Co ntrol ,"
ASME Paper 7 7 -DE't -10 , De s i gn Engineering T echn ical Conference,
Chicago,IL., 26-30 Se ptemb e r 1977.

U. D.W. Childs,"Convergent-Tapered Annular Seals: Anal ysis for Rotor

dynamic Coefficients." Fluid Structure I nt erac ti ons in Turbomachinery
presented at the ASm wi nter annual m!eting, 15-20 Novemb e r 1981.

12. R.C. Hendricks,"Some Flow Characteristics of Conventional and Tapered

high-Pressure-Drop Simulated Seals," ASLE p r ep r in t No. 79-LC-3B-2,
pr esen ted at the ASLE/ASME Lubr i cation Confe renc e, Dayton, Ohio;
16-18 October J.979

181
13. D.W. Childs, J.B.Dressman, and S.B.Childs, Testing of T urb ulent
'

Seals for RotCllrdyn'tuDi.c Co eff i cients ," NASA Con fe rence Publication
2133,Rotordynamic Instability Problems of High Performances
Turbomachlnery, p roceed ings of a works hop held at Texas A&M Univ.,
1 2- 14 11ay 1980, pp. 121-138, proceedings p ub lished September 1980,

14. W.L. Bowen , and R. Bhateje, "Th'e Hollow Roller Bea ring . "AS ME Paper,
No. 79-Lub -15, ASME-ASLE Lubrication C onf erence ; Dayton,Ohio; 16-18
O ctober 1979.

15. H. Stephen and K.Lucas. Viscosity of Dense Fluids , p.59,P1enum Press ,

New' York-London

16. Anon., T-1301,Thermodynamic Properties, Dupont Fire Exti nguisher

P rope rt ie s , E. I. Dupont De
' Nemours c. Co., Wilmington, Deli 1966.

182
 Figure 1. -
Test-rig assembly.

SERL 1 DRTR
AVG RXIAL REYNOLDS = 382,98.

Cl FILE SEAL221031.RD RC 139857 RA 39C1270

'" FILE SEAL231031.RD RC 114747 RA 3e3!5
+ FILE SEAL241031.RD RC 89931 RA 378744
FILE SEAL251031.RD RC 64982 RA 377426

5ERVO

+ CONTROL
...... VAI..VE':;

OI!>Ci\AP.'iE.
ACC"MULATORS

'"
'"

C
"

'" SEAL 1 PROFILE

'"

'"
'"
"'fa.oo 0.40 0.80 1.20 1.60 2.0(1 Z .40 Z'.80
AXIAL LOCATION I IN)

Figure 2. -
Test-facility layout.

Figure 3(a).

183
 SERL 1 DRTR SERL 1 DRTR
AVG AXIAL REYNOLDS . = 323001. AVG AXIAL REYNOLDS . = 75305.

Cl FILE SEALl71031.RO RC 139809 RA 3235G7 fILE SEAL061031.RD RC 139912 RA 75910

'" FILE SEALl8103I.RD RC 114779 RA 320515 '" FILE SEAL071031.RD RC 114779 RA 75162
+ FILE SEALl91031.RO RC 89888 RA 322849 + fILE SEAL091031.RD RC 89986 RA 74652
X FILE SEAL201031.RO RC 65031 RA 322346 X FILE SEAL101031.RD RC 64942 RA 76225
FILE SEAL211031.RO RC 39947 RA 325786 FILE SEALl11031.RO RC 39976 RA 74578

o
o

_0

:i:
0..

o o
o o
o
o
w .,

: :::=;=====
SEAL P ROFILE

t-:=::=
.OO z':4Q---z.ao-"
0.40 0.80 1.20
AXIAL LOCATION
1.60
(INI
2 . 00 .oo 0.40 0.80 1.20
====;===
1.60 2.00
s- __ ___
2.40 2.80
AXIAL LOCATION (INI

Figure 3(b). Figure 3(d).

SERL 1 DRTR
SERL 1 DRTR AVG AXIAL REYNOLDS . = 49927.
AVG RXIAL REYNOLDS . = 149168.

Cl FILE SEAL011031.RO RC 139755 RA 49828

SEALl31031.RD RC 115074 RA 150306 115021 49691
"'FILE SEAL021031.RD RC RA
SEALl41031. RO RC 90085 RA 148369 90093 RA 49774
+ FILE SEAL031031.RD RC
SEALl51031.RD RC 65172 RA 149365 64997 RA 50870
X FILE SEAL041031.RD RC
SEALl61031.RO RC 40030 RA 148631 RA 49471
FILE SEAL051031.RD RC 40048

o
o

o
w
"''''
::om
",m

'"
W
",0
o..

o
o o
o
m
o

0>

g PROFILE

b-::
m.oo --= :=-0.80
=- -
2.40
-"
:t-::::=:
==;:::=-:r::-
0.40 1.20 1.6Cr 2.00
AXIAL LOCATION !INI
.OO 0.40 0.80 1.20 1.60 2.00 2.40 2.80
AXIAL LOCATION (INI

Figure 3(c). Figure 3(e).

184
 SERL 2 oRTR SERL 2 OR R
RVG RXIRL REYNOLDS . = 434487. RVG RXIRL fE,NOLDS " = 150886.

!r'
Q) FILE,SEAL01 1032.RD RC 140143 RA 432319 Q) FILE'SEALI21032.RD RC 140002 RR 149563
"FILE SERL021032.RD RC 114806 RA 431793 " FILE,SEALI31032.RO RC 114855 RR J 5154!
+ FILE SERL031032.RD RC 89832 RA 434150 + FILE,SERLI41032.RD RC 90063 Rh 149 J 18
X FILE SERL041D32.RD RC 64983 RR 432116 X FILE,SERLI51032.RD RC 64923 RR 151027
1>FILE SERL051032.RO RC 39983 RA 442058 ,,"u """ ,<" RC 40074 RA 153180

o
00
m

'j \
o
o
o
m
m

, \
1
-"
;I I

",0
,,-0
w
Ng;

LI
o
o
'";1
1

SEHLj2ROFILE
__________
"- "
' ' 2r:-O=-=0 4 :----="'
- ".-0 O-
.,CC
0.80 1.20 1.60 2.00 2.40 2.BO "b'.oo 0 .40 0.80 1.20 1.60
RXIAL LOCATI ON 11 N) RXIAL LOCRTION (IN:

Figure 4( a) Figure 4{c).

SERL 2 oRTR
SERL 2 oRTR RVG AXIRL REYNOLDS . = 75166.
RVG AXIRL REYNOLDS = 325656

Q) FILE,SERLI71032.RD RC 139905 RA 74777

Q) FILE,SEAL061032.RO RC 140079 RA 324572 "FILE,SERLI81032.RO RC 114912 RA 74013
"FILE,SERL071032.RD RC 114861 RA 326364 + FILE,SEAL191032.RD RC 89791 RA 75615
+ FILE,SEAL091032.RD RC 89904 RA 325388 X FILE,SEAL201032.RD RC 64956 RA 75589
FILE,SEAL101032.RD RC 64963 RR 326301 FILE,SEAL211032.RD RC 40050 RR 75836
0
0
0
N
m

o
0 o
0 o
N
0
m
m

0
0
_0

a..
-
0
w
",0
:;,0
u;m
u;
W
",0
a..
0
"'le
...J
a:
wo
u;
0
N

0
0
0
m
'"

0 SERL PROFI
0
0

4, .00 0.40 0.80 1.20 1.60 2.00 2.40 2.80 0.40 0.80 1.20 1.60
RXIRL LOCRTION I IN)
2.00 2.40 2.80
AXIAL LOCATION I IN)

Figure 4(b). Figure 4(d).

185
 SEAL 2 DATA
AVG AXIAL REYNOLDS . = 49212.

CD FILE SEAL231032.RO RC 140537 RA 49552

"" FILE SEAL241032.RO RC 114619 RA 47769
+ FILE SEAL251032.RO RC 90015 RA 50082
X FILE SEAL261032.RO RC 64890 RA 48676
FILE SEAL271032.RO RC 40108 RA 49981

o
o

o
o

.OO 0.40 0.80 1.20 1.60 2.80

AXIAL LOCATION

Figure 4(e).

186
 VIBAATION AND DESTABILIZINJ EFFECTS OF FLOATINJ RINJ SFALS IN CDVlPRESsc:RS

Mark F. Bnerick
COmpressor and CUstom Pump Division
Allis-Chalmers Cbrporation
Milwaukee, Wisconsin 53201

I . INTRODUCTION

Operating experience on a compressor commissioned 12 years ago has pre

sented an interesting history of sporadic increases in shaft vibration.
Initial operation was satisfactory with low levels of vibration. However,
after some time the shaft vibration level increased to several mils. Ini
tially this was believed to be due to rotor unbalance from deposits formed
in the passages due to process upsets. After cleaning up the rotor, opera
-
tion was again satisfactory. In time the vibration level again
increased. It was then found that the rotor vibration was primarily sub
synchronous. Further investigation revealed that the original seal design
was subject to wear and was no longer properly pressure balanced. A
modified seal design was installed and it has operated successfully for
the past six years.

Subsequent analysis has provided a better understanding of the seal destab

ilizing effects on the rotor and motion of the seal which has been con
firmed'by test data on the current seal design. These will be briefly
presented.

NOMENCLATURE

MW molecular weight

P pressure

Kyx Kxy cross coupled stiffness damping properties for seals

Cxx Cyy principle damping properties for seals

14' angular velocity (rad/sec)

D diameter

P. oil viscosity

L seal or bearing length

c clearance

F force, radial
P

FA force, axial

187
It coefficient of friction

Q aerodynamic or seal destabilizing

( logarithmic decrement

II COMPRESSOR DESIGN

The subject machine is a 5-stage, vertically-split, centrifugal compressor

with floating ring oil film seals in refinery service. (See Figure 1)
The gas is a diesel distillate (MW = 8.2) and the process conditions are
as follows:

Inlet Pressure P = 2896. K Pa (420 PSIG)

1
Discharge Pressure P = 4178 K Pa (606 PSIG)
2
Flow =
322 Kg/min (739 LB/MIN)
Driver Size = 1118.6 KW (1500 HP)

The compressor has 38.1 cm (15") diameter impellers and operates at 12,320
RPM in 8.89 cm (3.5") diameter tilting pad journal bearings. The seal
diameter is 11.43 cm (4.5 in.)

III OPERATING HISTORY

Commissioned in 1970, the machine had a history of minor, occasional

vibration problems. A pattern developed which was noted by the
Allis-Chalmers Field Service and Repair Group:

The machine generally operated smoothly following service or

maintenance.

Increased shaft vibration would develop over time (6 months).

Seal oil flows would sometimes increase substantially, resulting
in operation of the auxiliary seal pump to keep up with the
increased flow.

Upon disassembly, the unit would be fouled with ammonium chloride

deposits in the aerodynamic passages. (See Figure 2) The presence of
the deposits and the resulting unbalance was initially believed to be
the cause of the vibration.

The floating bushing would be quite worn on the axial face

resulting in high axial forces on the seal housing. (See Figure
3 for arrangement.)

Finally, in 1975 the operator reported that vibration levels had become
unacceptable and noted that the machine behaved differently with each of
the two rotors (main and spare). One rotor performed smoothly with
enlarged radial seal clearances (8 mils vs. 2-3 mils design), even though
it was fouled and a balance check indicated it was out of balance. The
other rotor ran rough with design seal clearances, despite a touch-up
balance. The apparent contradiction between vibration experience and the

188
machine balance condition strongly suggested that the vibration problem
was non-synchronous in nature. The Allis-Chalmers service group discussed
the problem with the Compressor Engineering group and a study was
initiated.

Field vibration spectra were obtained by Allis-Chalmers on the balanced

rotor at several locations using displacement probes and accelerometers.
See Figure 4a.

The data showed:

1. A subsynchronous vibration signal was present at all locations

checked. (See Figure 4b & c)

2. The frequency which would change with a slight variation in bearing

and seal oil temperature, varied slightly from 80.6 to 81.6 hz, (4836
to 4906 CPM).

3. Accelerometer data (integrated to yield displacements) showed

casing motions which were significantly lower than shaft
amplitudes indicating that the probes were measuring actual
shaft motion, and not a foundation resonance.

Vibration amplitudes at various locations are shown in Table 1.

TABLE I. - VIBRATION AMPLITUDES AT 4850 CPM (FILTERED)

AMPLITUDE A IBI A OBI OB2

l 2
mm .0127 .051 .009 .009 .074
(mils) (.5) (2.0) ( . 35) (3. 5 ) (2.9)

The synchronous component of shaft vibration was less than .0127 mm (.5
mils). This coincided with the customers comments about the performance
of the two rotors. An increase in the synchronous vibration of the out
of-balance rotor could still result in lower overall vibration levels if
the sub-synchronous component present in the well balanced rotor were
eliminated.

IV ANALYSIS OF PROBLEM

At the time the machine was designed, analysis was limited to undamped
critical speeds, so an updated rotor dynamics analysis was performed using
improved rotor dynamics programs in use at Allis-Chalmers in 1975 which
included:

Undamped critical speed map showing intersection of

undamped critical speeds with bearing stiffness curves.
Mode shapes.
Elliptical orbit synchronous response analysis.
Stability analysis.

189
The critical speed analysis indicated that the machine was operating
between the 2nd and 3rd modes, see Figure 5a. The mode shapes show sub
stantial motion at the bearings for the second mode indicating it should
be well damped (see Figure 5b). This is confirmed by the response
analysis (Figure 5c), note that the 2nd mode is well damped and that the
amplitudes produced by an unbalance distribution based on the API
residual unbalance limit are quite low (less than .002 mm, [.8 mils]).
The response analysis showed a 1st resonance at 4100 RPM, and the 1st
critical speed on test was 4086 RPM.

Baseline (no destabilizing) stability analysis showed acceptable stability

with a log decrement of .169 at 3845 CPM for the 1st Y-mode in forward
precession. (See Figure 5d.)

Since the machine was stable under baseline conditions it was then
desired to evaluate the rotor's sensitivity to destabilizing forces. To
approximate their destabilizing effect the stiffness and damping proper
ties of the seal ring were estimated by assuming that the seal stops track
ing the shaft ("locks up") and behaves as a non-cavitated concentric plain
sleeve bearing.

Under these asumptions the properties are given approximately by:

Kxy = Kyx = f CAr D)C- ()l

tAT C =.,.C = 2 Kxy
xx yy

From this calculation the properties developed for design conditions are:

5
Kxy = Kyx = 1077 N/CM (6.148 x 10 lb/in.)
Cxx = Cyy = 1681 N-S/CM (960 lb-sec/in.)

The principle stiffness terms Kxx and Kyy for a concentric seal are negli
gible. Note that these properties are highly sensitive to variations in
3
clearance (inversely proportional to C ), and that quadrupling the clear
ance reduces the properties by a factor of 64. Thus the enlarged clear
ances at the seals found on disassembly could have allowed operation even
if the seals were locked up. This explains why the unbalanced rotor with
enlarged clearances operated with less vibration than the balanced rotor
with design clearances which had bounded whirl.

More sophisticated calculation schemes exist to develop the seal complete

stiffness and damping matrices for various assumptions about the seal
lock-up eccentricity. However, the results are sensitive to the
assumptions about whether lock-up results in increased or decreased
journal loading. Reduced journal loading will change the natural
frequency of the rotor because the effective bearing span changes with
transfer of the load to the seals. Due to the uncertainties associated
with the assumption of a lock-up eccentricity, the simple concentric seal
properties were used in this case.

190
Various values of seal destabilizing were input into the stability
analysis to evaluate the system sensitivity. See Figure 5e. The
stability analysis shows zero log decrement with aerodynamic destabilizing
of 2329 N/CM (1330 lb/in) distributed among the impellers and 38,530 N/CM
(22,000 lb/in) at each seal. This is substantially less destabilizing
than would be produced by the locked up seals with original design clear
ance. Thus the seals can produce sufficient destabilizing to drive the
rotor into bounded whirl under lock-up conditions.

The frequency of the analytically predicted unstable mode was 3850 CPM as
opposed to 4850 CPM in the field. Phase information from the field test
data indicated that the shaft ends were in phase, so it was concluded from
the limited available information about frequency and mode shape that the
1st 'y' mode of the rotor was unstable (bounded whirl) and was being
driven by the seals. Several other conclusions can be drawn from the
operating experience and analysis:

1. Stable operation of this rotor with the design seal clearances is

only possible if the seal "tracks" the shaft and doesn't lock-up.This
implies that any destabilizing produced under tracking conditions must
be substantially less than that present under lock-up, in fact less
than 38,530 N/CM (22,000 lb/in).

2. Stable operation is possible if the seal locks up, if the

seal clearances are abnormally large (resulting in much smaller
hydrodynamic destabilizing forces).

3. The difference in the frequency of the unstable mode between analysis

and field data is possibly due to the development of principle
stiffness terms (Kxx, Kyy) at the seals due to an eccentric lock-up
of the seal ring which transfers bearing load to the seals, thus
reducing the effective bearing span of the rotor, and raising its
natural frequency from 3850 to 4850

4. The frictional force (F ) which restrains the seal from moving

R
radially (and determines lock-up eccentricity) is a function of the
pressure induced axial forces (F ) on the seal ring and the
A
coefficient of friction ( ,) between the seal and its housing.

Review of the axial forces (F ) on the seal at the design pressure

A
with no seal wear show a relatively small value. See Figure 6A.

2 2
(D - D ) = 792 N (178 lbs.)
2 1

However, as wear occurred on the axial face of the seal, the outer
diameter of the contact face (D )increased. For example, if axial
2
wear on the ring was .0254mm (.001") one fourth of the chamfer would
be removed, and D would increase from 11.53 cm (4.54 in.) to 13
2
cm (5. 118 in.).

191
 At this point the axial force would be 8985 N (2020 lbs.) or 11.3
times the original seal design value. The radial force could increase
by more than this if the coefficient of friction increased with wear.

Thus the original design was highly sensitive to both the friction
coefficient and wear so that following some initial wear, the wear
rate would accelerate until lock-up occurred.

4. Since the seal parts had shown substantial wear during earlier
maintenance and service inspections, the compressor performed
well after maintenance, and the stability analysis showed good
correlation with experience, seal modification was selected as the
best methotl of resolving the problem.

V SOLUTION:

Such a seal-induced instability can be solved by two types of seal

modifications:

1. Allow the seal to lock up, but reduce the hydrodynamic forces
produced by changing the geometry of the seal in the following ways:

a. Reduce seal effective length - reducing the effective

length by adding grooves to the seal surface reduces
the stiffness and damping produced but this is at the expense
of reduced film thickness and therefore seal centering
capability which can increase the possibility of seal rubs.

b. Increase seal clearances - This reduces the stiffness

and damping properties but increases oil flows dramatically
which is desirable.

Increased clearance due to wear allowed operation in this case

despite a locked-up bushing. The auxiliary seal pump ran all the
time. In addition the customer added a thir pump to keep up.

2. Balance the axial forces on the seals as closely as possible so

that the seal doesn't lock up to begin with. Reduction in axial
forces makes the seal less sensitive to the coefficient of friction
between the seal and its housing (which can increase with wear) and
reduces the seals tendency to wear.

The second type of modification was used to solve this problem. Figure 6a
and 6b show the original and modified seal bushings.

Note that the following modifications were made.

1. The residual axial force on the bushing was reduced by balancing

the pressure induced axial force.

192
 a. a-ring was removed from bushing end reducing friction.

b. Face relief was remachined to better control pressure

equalization.

c. Pressure balancing axial hole added.

d. The taper was removed to make the new design

insensitive to wear.

2. These changes don't affect the hydrodynamic performance (i. e., leak
age or film thickness) of the seal, only the force required to move it
radially (FR).

Since the new seal has been installed, seal wear has been negligible and
the sub-synchronous vibration problems have been eliminated. The modified
seal has performed successfully for over 6 years.

Allis-Chalmers has over 20 years experience in the design and application

of oil film seals. The current Allis-Chalmers standard "Trapped Bushing
Seal" features:

A "dual" bushing which encompasses both the inner and outer seal
in one ring for reduced axial length. (See Figure 7)

Low residual axial force on the seal which effectively reduces

the potential for lock-up and seal sensitivity to friction and
wear. (See Figure 8)

VI SEAL VIBRATION TEST PROGRAM

A test program was subsequently conducted in the Allis-Chalmers test

facility to verify the motion of the A-C standard dual bushing at design
pressure level to insure that the bushing tracks the shaft without
lock-up. All seal vibration data were provided by Mr. P. G. Shay, the
Supervisor of the Allis-Chalmers Compressor Test Facility. Two displace
ment probes were mounted 90:' apart in the cage surrounding the bushing to
determine the amplitude of bushing motion and the relative phase lag
between the shaft motion and bushing motion. (See Figure 9, note the
epoxy-filled relief around the probe tips.) The oscilloscope traces show
the seal orbit to be circular with a phase lag of 45 behind shaft motion
and amplitudes slightly less than the shaft amplitude (See Figure 10.)
The vibration spectra show the seal motion to be predominently synchronous
with only small traces of non-synchronous motion. Shaft vibration is
entirely synchronous. (See Figures 11A-D.)

Based on this information, it may may concluded that this arrangement

results in minimal destabilizing effects as the bushing is able to freely
track the shaft motion. Seal induced hydrodynamic forces are dissipated
in seal motion and not applied to the shaft. The seal is also insensitive
to wear on the axial faces.

193
VII CONCLUSIONS:

1. Residual axial forces in seals can influence seal and shaft vibra
tion. Some small level should be present. However, seal lock-up
should be avoided.

2. Restraint ("lock up") of bushing results in high levels of

destabilizing forces and it is therefore better" to err on the low side
with respect to axial (pressure induced) forces in the event that seal
wear increases the friction coefficient significantly.

3. Normally tracking seals exert only minor destabilizing effects.

4. The seal design should be relatively insensitive to wear on its

axial face to prevent accelerating wear rates.

REFERENCES:

1. Rouch, K. E., Kao, J. S., "Reduction in Rotory Dynamics by the Finite

Element Method". ASME Paper 79-DET-70.

2. Alford, J. S., "Protecting Turbomachinery From Self-Excited Rotor

Whirl", Journal of Engineering and Power, Trans. ASME Series A, Vol.
87, October 1965, pp. 333-339.

3. Lund, J. W., "Stability and Damped Critical Speeds of a Flexible

Rotor in Fluid Film Bearings", ASME Paper 73-DET-103 - 1973.

4. Kirk, R. G., Miller, W. H. "The Influence of High Pressure Oil Seals

on Turbo-Rotor Stability ASLE Preprint 77-LL-3A-l.

ACKNOWLEDGEMENTS:

1. P. G. Shay, Supervisor, Compressor Test, Compressor Engrg.,C&CPD

2. K. E. Kraemer - A-C Field Service and Repair, C&CPD.

3. J. H. Hudson, Supervisor, Mechanical Design Engineering,

Compressor Engineering, C&CPD.

4. Keith Rouch, Advanced Technology Center, Allis-Chalmers.

194
 I
I

11_

Figure 1. - 5 stage barrel compressor.

195
 Figure 2. - Fouling deposits in compressor.

1 - BUSHING
2 - DIAPHRAGM
3 - IMPELLER
4 - STATOR

--- -'----

Figure 3. - Seal arrangement.

196
MOTOR H

COMPRESSOR
AcceIeromeIers:
A, Verticlll on COUpling guard Displacement
A2 Verticlll on outboard bearing cap IB; 30" c/OCkwlae from vertical as viewed from motor end
A(I Verticlll com casing inlet end
on OB, 30 clockwlae from vertical as viewed from motor end
A4 Vertical on gear box O 60" counterclockwlae from vertical as viewed from motor end

Figure 4(a).

Figure 4(b). - Vibration spectrum inboard probe I.

Figure 4( c ) . - Vibrtion spectrum outboard probe I.

197
 CRITICAL SPEED MAP

"b
-__ +- __ L- ____-LLW

<t,
----r-rTr---rrnnT--r-rTTTrM-
lrf lrf . Hi
SUPPORT STIFFNESS

Figure 5(a). - Critical speed map.

ROTOR MODE SHAPE AT CRITICAL SPEED

BRG .l BRG 2
o
c::>
I I
...;

o
o

'0.00 '20.00 110.00 GO.OO 90.00 100.00

DISTRNCE RLONG ROTOR - INCHES

BRG BRG CRITICAL SPEED

NO. STIFF 1 3.926 RPM (1)
2 115.93 RPM
1 . '50X105
.t.
3 1?03 RPM +
2 .50X105 '! 2260? RPM x

Figure 5(b). - Mode shapes.

198
 PROBES:

--O\J;c/O OOQ) STA l THRUSTBRG.END

O
Qc::?c? (J 0 0 0 0 STA COUPLING END

J
"'I
01
I
'" I
'"
1

0
'"

q
ri STATION UNBALANCE ANGLE LOCATION
Ul
.-' 2 .197 O. Coupling
r 14 .134 16. 1st Stage
I 18 .037 90. 2nd Stage
>.
22 .037 90. 3rd Stage
ro
'" 26 .037 90. 4th Stage
C1
30 .037 90. 5th Stage
0
'" 32 .134 164. Balance Piston
45 .120 180. Thrust Collar

4.00 9.00 12.00 16.00

ROTOR SPEED WDD RPM

Figure 5(c). - Synchronous response.

o FWD
o MODE 2 LOG DECREMENT = .17 RPM = 384?

o
m

.0
.-Jm
1.U .
!

o
....,

'0. DO 20.00 tlO.DO GO.DO

i NGTH

Figure 5(d). - Shaft whirl mode.

199
 liii "
' ... F ., tjf';. 1:1:;: 1';,:., ,,, . ':::'
I'!i if r: :;', I> : . ' . ,: W, .,
""tilt. !: FCC'; , .. ljii I". Ui
j;:: b" r. .r L .. n'!" , .'.
I".H'. ..+ ...... : i.
.1
:: "

:;U; iH:;' I.TH . ii '

' .' i .l: :"1 .. ' .. ::ii:
'fil'f:;":.t:., ,: ... . ' '''' ':i
W
',' Ji' I ": : 1,:; L
l'd';I" PY ",,:, ' Ii' : I I' .. .. L ....J
,m
I
1" i. : ." I
. ..

Z . ::'iI"',.:;' .','., .. . .. " .: K': " . I ! iL

w
:, "

Ul::".' "''1' ' , ; ..

. .
.' I.' ' '''1'' ..

'''.' ." .:: ,
I
kW:i :" r::::; I""; " . ,'. . r: .ll . ... ' ' CJ)
W I ' I '" ['2': i
II: ;;: . ' I',. . , 'i if :;
.' . I I l ." I
o W:U: ,' . ,' I .: 1 .. ' i '. iX .'. I, ",1;; UtF:0
W
.
'i":: .' I r: I'..
. 05
. ...
o
. :I. i! i ! :0. 1 I F;;
(!) '. I:, ,, ', . " . , . . . I
1\
' , . ' b""'!
i' I> .. .' '
o
....J : : " .. ...
,... I

o r;.LlW logO( ' .: inlVlIGn .:xlnMn . 2p

60){)1);:i
"'P >":;';:.
I::., ."
: , . I:... ... .. .. , ....
.. J' .i ... ! ' . " . ' ......... , : ....... \ ... . . " . ,:,HI
nii' " ,,',. ,.. I

.. <I

. .....
::I"I;i;;': i ,.' : '. .. ." ... . .... T
:,li: <:'::iii! i', : ',' I .' : , ""' , "., . , ,i,
I::> .. " ' .. I . ',,'
:: i:;', I .... '. I .. 1;; ....... ' , .;. . : '.
r:,:i Jf+P:;:;Jii " : : : , : : : ': . .. r' .
'
. .
. ': :;,:;1;1:;.:: :T: . : ; . " . : Iii ,,'

TOTAL DESTABILIZING Q (LB/IN)

Figure 5(e ). - Log decrement vs. destabilizing.

T CHAMFER
, LENGTH
-----

4.54 OIA. (02) 4.48 OIA.

! (01)

Figure 6(a). - Original bushing. Figure 6(b). - Modified bushing.

200
Compact Design - allows shorter
bearing spans for higher critical
PROCESS
speeds of the compressor rotor. GAS
Sleeve (impeller) with interference fit
under bushing - protects shaft and
simplifies assembly and disassembly.
Requires only a jack/puller bolt ring.
TO
Spacer fit at initial assembly - no INNER
field fitting of parts. OIL DRAIN
ITEM

1. ... ...Shaft
. . . .
DESCRIPTION
CD
2 . ... . .Impeller
. . . . .

3 . ... ....Stator
. . .

4. . ...Stepped Dual Bushing

. . . . . .

5. .......Bushing Cage
' "

6 ... . ..Nut
. . . . .

7 .. . ...Shear Ring
. . . . .

8 . .. . .Oil/Gas Baffle
. . . . . .

9 ... . .....Spacer Ring

. .

TO
RESERVOIR

Figure 7. -
Trapped bushing seal arrangement.

-----.---
- -

I ttmmtmf PD :t;
ltfftltlftl/!E"ltttftlttJfrtt
PD PD
Ps
tt TOTAL FORCES ON DUAL BUSHING
RESIDUAL FORCES ON DUAL BUSHING
t\:'
I I
Ps - SEAL OIL SUPPLY PRESSURE
P D - OUTER LEAKAGE PRESSURE

Figure 8. -
Pressure induced forces on dual bushing.
201
 Figure 9. - Probe arrangement for trapped bushing seal vibration test.

1 Mil

I I Upper: Bushing Orbit Upper: Bushing X

1 Mil Lower: Shaft Orbit Vibration Y

Lower: Shaft X

Vibration Y
13050 rpm 1 MIL Per Division

F ig u r e 10. - Shaft and dual bushing vibration data.

202
 31-----+-i----t--
I ----'---t---Tl-------r
I I
! II I

11
I I ' I

0:-

I
2
fs
I

r---I--_+
I

1e---- -1 1---
'
-- - -->--

I
. .

I I
Figure ll(a). - Vibration spectrum - shaft (horizontal).

3t----+----r-1-i
, I --1 r-r-
i +---+--lii! I
.
!
0:- I i

I
2-f---
!!},

i I I I Ij

a
1J----
I I

a..-....._
.. _"--____-=--_
----....M-._....____
.. .JJ
Fi gure ll(b). - Vibration spectrum - shaft (vertical).

203
 r I -r---Jr--r-r" -T- r--l,-ll
it
cL
rT
21 i '
-
i
s: ,-------.
I I i
j-----r------+-
i I
.. -r---t-'I -- 1-------1
' i
-;; () i
I II !i I ! ;
- -

==
g ! i j
.

I t- '---+-iii
::E I I
I
r---- i ----t-----t----
"-'--i-" ---+------lI - ------ I I I
1

, i, II I L_UJ
Figure 11(c). - Vibration spectrum - seal (horizontal).

I
r---t---i-----r--1.---- :------ ---i--11 r
I ! I I

I
I I 1 I I' I II !
i
I I
I
,--'--'1 . .-.j I
I

'-- --------ta..---
4 I ! I
I
()g -----r------
::E i I
I '

j --------i------
: -t-------L !

I , . , II
. --

I I j I i 1 ! ! !!
,I i ! ! I I I
I
i I I I !
I
I
I' !
1

Figure 11{d). - Vibration spectrum - seal (vertical).

204
 F.I.(M7 INJ:XX::E) :FC:RCE OF IABYRINlli SFAL

Takuzo Iwatsubo, Naoto Mbtooka,

and Roj i Kawai
Faculty of Engineering , Kobe University
Rokko, Nada, Kobe, 657 Japan

SUMMARY

This paper deals with theoretical analysis of flow induced instability

force due to labyrinth seal. That is, a approximate solution is given for the
partial differential equation representing the flow in labyrinth seal and it is
compared with the finite difference method which was proposed in the previous
report in order to verify the accuracy of both methods. Then the effects of
difference of inlet and outlet pressures of the seal, deflection of pressure and
mass flow from the steady state, rotor diameter, seal clearance, seal interval
and seal number on the flow induced force of the seal are investigated and it is
known that some of these factors are very influential on the flow induced force.,

INTRODUCTION

In the previous report (1), the fundamental equation of flow in the labyrinth
seal was derived by consi ering the effec t of the variation of gland cross section.
The equation is numerically solved by using the finite difference method. Then
the spring and damping coefficients of the labyrinth seal are calculated, and by
using the result, the stability of a rotor system having labyrinth seal is
discussed on this coefficient by using the energy concept. Furthermore, experi
ments are executed to observe the flow pattern in the gland and to study the
characteristics of the flow induced forces in the labyrinth seal.

However, further theoretical investigations are required for the accuracy of

numerical calculation and more detail description of labyrinth seal behaviour.
Then in this paper the equation is solved by another method and the calculated
results are compared with the solution of finite difference method. After that
influences of deflection of steady state, pressure difference between inlet and
outlet, labyrinth seal radius, seal clearance, seal pitch, number of seal cham
ber, seal strip height and divergence and convergence seal on the flow induced
force, phase angle and leakage flow rate are investigated.

SYMBOLS

t time

w = Rsf

RS radius of labyrinth seal

p pressure in seal chamber

205
c peripheral velocity in seal chamber

f density of gas

q mass flow rate in axial direction

angle from x axis

f cross section of seal chamber

2. length of strip pitch

h strip height

radial clearance of seal

friction shear stress of stator surface

friction shear stress of rotor surface

U' length of acting surface of shear stress ( stator )

u" length of acting surface of shear stress ( rotor )

R gas constant

T absolute temperature of gas in seal

"t specific weight of gas

1f" flow coefficient

n ratio of specific heat

W rotating speed of rotor

Wn critical speed of rotor system

u=WRs peripherical velocity of rotor

K coefficient of viscosity

J) coefficient of kinematic viscosity

Reynold's number

friction coefficient ( stator )

X' friction coefficient ( rotor )

subscript * steady state

206
 EQUATION OF MOTION

Derivation of Equation of Motion

Derivation of the equation which describes the labyrinth seal flow is

followed the previous report (1), where the following assumptions are taken and
schematic flow is shown in Fig.l;

(1) Fluid in labyrinth seal is assumed to be gas and its behavior is assumed to
be ideal.

(2) Temperature of the fluid in the labyrinth seal is assumed to be constant.

(3) Cross section area of the seal gland is assumed to be constant in spite of
the deflection of rotor and time derivative of cross section area is only consi
dered.

(4) Change of flow state in the gland is assumed to be isentropic change.

Thus the fundamental equations with respect to continuity and momentum for the
ith seal element are as follows;

8f) + f {i) t (l6-i.i"I-ifn) = 0 (1)

.1-. (pi,Cb + (fLo h Ci.) + ( a... . C. fL. c. ) + ,.,.1 Ul _ """ "U:I - '0 Pi- f.
LT l.. tJ {J l.. [.-1
=
(2)
_

Vi. (. I.. j. "

) i... dW t L.
dW "

As temperature of each gland is constant, the equation of state in the ith gland
is given as;

where
Yt, fj, ,.
=

Next, denoting the gas flow velocity through labyrinth nozzle ( seal clearance )
S, the relation between axial flow rate and pressure in the gland is given by
the the rmo dyn ami c al energy equilibrium condition as,

Po n-I
s= 2g----p Vl(l-()-n- J
n -I 1 1'L (4)

where n is isentropic exponent, v specific volume and subscripts I and 0

mean inlet and outlet of nozzle, respectively. Assuming that the flow is adiabatic
change, i.e. vo=(Pr / Po)Vt' the flow rate through the nozzle G [ kgf / sec ] is
given as;
'b_ I .(...LJt )W-
2. R 1,\+
=d. )n=rJ.. F
J
0_
G= 2g-1L J:t (--.Q.. )71
'\to 'Vl 1't n -I '\It f1. PI (5)

207
where F is nozzle cross area ( area of seal clearance ) and subscripts ( i,a )
correspond to the gland number of the labyrinth seal. Applying the above rela
tion to flow in the labyrinth seal, we assume that the labyrinth seal is a
series of the nozzle, pressure difference between each gland is small and enthal
py of each gland is nearly equal. Then leakage flow rate through labyrinth seal
G ( kgf /sec ) is given by using the relation

(6)

Applying Eq. (6) to the flow through the ith seal element, mass flow rate for
unit area is given as;

cJ. t;
J RT /R.-:'I ;2 _ 1)2
J"L

Rewriting Eq. ( 7 ), the following equation is obtained ;

where
(8)

Linearization of Fundamental Equation

For the linearization of the equation, the perturbations of 'pressure,

peripheral velocity and flow rate from those of the steady sta te and also mean
value and perturbation of seal clearance are introduced as

R. i..( I + )
= Ci,= C*-i. (I + ti,)
(10)
Cf5i.= *i. (I + ) 6i. 8*i. (I +"0/)
=

where ,cand q*are pressure, peripheral velocity and axial flow rate of
steady state in the ith gland and i ' t and are nondimensional perturbation
terms of pressure, peripheral velocity and axial flo-w rate. Also 01"

and .''0/*
-
are mean value and perturbation of seal clearance.

Assuming that the rotor is whirling along elliptical orbit, tVt is repre
sented as

(11)

208
 Then the cross section area in -the ith gland is represented as

fi. = {hi. + b*i. ( - ( + "o/i.) + hL-t1 + 6*i.tl (I + "ljri.+I)} i (12)

As the change of state in the gland is isentropic, the following relations are
obtained;

Pri.. Po.
f- A (13)
- -

P;1- -

therefore

afl =nR (14)

gt 'pi, 9t
Substituting Eqs. ( lO ) (14) into Eqs. ( l ) and (2), and denoting the time
*-
-

derivative and spatial derivative ( : ) and ( ) I, respectively, the

.

following equations are obtained for the ith seal element;

f""i. . + Cltitt, 1;1 + Cfi.. t g: RTFit&ti.i"1 E , + RTr8.i. (COtt,,+1

2., i" C2.. ).
'1 ", C,"... , t;",
_

Ott
...
n nRs c, i.. Rs h %it'i. *
1') 2
:z. 2
- R T'Rtt.-1 ?J*i.fI..; + Q R T"2o*i.. r-tfL-i ('R2
2. 2. 2. c. 2
R T b1ti.... 1 (R a ) d" 'R2 )
t..+1
_ _ .

-tfi.-I- -let.
2 Ri.- *
it\.- oft
%/Rt.. i.-I
2 'P*t- *
.
}-RTf{2
=-

*P*i--
b*i:tl (R:- Rt:t-I)t a+lcoswtcoS'P+bi.+ISinwtsincp J
t;2. b* (_I- R){aL-c.oswtcoSCf+ bi.Si nwtsH1'f}
+ {-aLWsiVlwtC.OSCP + bi.WCoswtsinf-lA,:tIWSir1wtsinftb"t,WC.oswtsin<f}
(15)

(16)

209
For the n stage labyrinth seal, Eqs.(15) and (16) are represented in matrix form
as;
lfU! +V ul'+ AUI <G.[Llil-COS(tf'twt)+ a*COS(f-wt)-bCoS(Lf+wt)
=

+ b,d:.OS (<f -wt + QJ2a.,.S;n(i.f+wt)+a"1(SiYl(r- wt) + bifSin

ltJ'+wt)
+ b Sin (If -wt
... ()j[ ettCOS(<p+wt) + Ba(J)SCLf- wi) -8bLOS U-p+wt)
+

+ ebCOsCLf- wt t ([,4[-Ba sin(t.p+wi) T9aSin(tp-wt)+ BbS;n(f +wt)

+ 9bSiYl (CP-wt (17)

where PitO = Po (inlet pressure) and PM = Pn (outlet pressure).

ANALYSIS

Analysis of Steady State

Setting the perturbation terms t and ti. zero in Eqs.(15) and (16) in order
to obtain the pressure and peripheral velocity of each seal stage in steady
state,
q
.+l
1
= q
.1
=

C".
(18)

q W - q /I \ . + T I U.' - T "U . " = 0 ( 19 )

i+1 i iUol 1-1 1 1

As the multi labyrinth stage is considered as a serial nozzle, the axial flow
rate in the steady state is obtained as follows

() L
1) - = r:f.o- L. J
n I '0' 0'
r i,-I ) [.-1
[( )1. - ( .J=J....)n+lj n (20)
_ n n "11 '
't-I ri..-I

where P_I is the pressure of (i-l)th chamber which is taken into account of
draughty flow. he pressure is given by Ref.(3) as;

(21)

where A is determined by the shape of labyrinth seal;

O
A
.== Tee (22)
l,.. i.. Cc. +tanB
where C c. is coeff.icient of vena contracta.
From these equations, state variables i.e. the pressure and flow rate satisfied
the equilibrium condition, is obtained by iterating Eqs.(18) and (19). The
peripheral velocity is obtained by using Eq.(19) in the same manner.

210
 Analysis of Dynamic Force

Re fe rring to the left term in Eq.(IT), the s olut ion s are assumed as follows

Pi.-::: 1" P-!('i" {AHSin(Lf-tUJt)+B+i,COS(f+wt)+A_SH1(tp-wt)tRi.COS(CP_wt)} (23)

Ci. C*i, +C?ti.l EtLS

== elf"" wt)+ F+i"COS(c.p+wt)+ESin (f-wt)+ Ecos (If+wt)] (24)
Representing the above coefficients in matrix form as

A+I Btl B-1

fil::: i' I , d/.2= (/31= F+I J liz = f-a
A+Yl B+n B-I'I
E+VI Ft-n F-n
o 0 0
the solution is written as;

Substituting E q (2 5 ) into Eq.(rr), and separating it to the elements cos U f +wt ) ,

.

sin ( Cj' +wt ) , cos ( cp -Wt ) and sin ( Cf -wt ) , the coefficients are represented as
follows;

WlfC#1 + 1+
:=
+$.1 +{J5"b*+QJ?eb (26)
- u51Tfil - V//1! t + $4 efA t 'iJb + ([;8 fJb (27)
- wl2 +VdJ.2 HidA1t+ HI3Bat- H,b* + H78b
=
(28)
wl2 - + H148a+ HI, + His eb
(29)
where ([;, "" G8 and !HI rv are coefficient matrices which is composed of the
matrices lB , It , IF and Solving Eqs. (26) , (2T) and E qs (28), (29)
. simultaneous
ly,
'JA8, '2 and 2 are obtained. Then velocity and pressure in the gland is
calculated by substituting the coefficients into Eq. (23), and the flow induced
force for whole seal is obtained by integrating the pressure for each gland.
Representing the x and y direction force by Pand P ,

n r2T..
Px.:=: - Rst-I JD R i".Q cosCfd if
(30)
p Rs t P1tp.;ti, '1
=-
(, smrd'f
1..-::.1 Jo

The flow induced force F and phase angle from x axis are described as

(31)

Spring coefficients and damping coefficients are obtained from Eqs.(30) and (31)
in the following matrix form;

211
 x C.tX.Cx 0
Y t
c:x:. Cj!1 (32)
eX. o Cexx. cex
e C9X. Cey

When Eqs.(30) and (31) are rewritten to Eq.(32) in the above, the following form
is used as ;i.'
i.-- 'i, C{* coswt + 2i.a1fSj nwt -t- jboffCOS wt + 4,-bsm wt't H el). coswt
=

+bLea Sinwt n,ebcoswt ;'i.8bSinwt

+ +

Using the above expression, the following relations are obtained between coeffi-
cients;

Ken= K8 K;x. ::;,-Kx. Kex =+<8X.

Ce= C8 C;t. = - CjX C&X= -Ce9Y.

NUMERICAL EXAMPLE AND DISCUSSIONS

The labyrinth seal having three teeth, which is shown-in Fig.2, is used as
a numerical model. Details of the labyrinth seal is shown in Table 1 and for
the parameter survey each element of seal dimension is independently changed in
order to investigate the influence of its seal size on the flow induced force.
In order to evaluate the seal'force, circular orbit (a* = b* in Eq.(17)) are
used as a whirling of labyrinth seal, and its radius a* = O . lmm . For the calcula
tion the following equation is used as the friction coefficient between rotor
and stator;
for stator surface;
P(U
Ri- =
l)
- C*d

where D is wetted perimeter length considered the rotating Reynod's Number

RL
<. f 200 1 Xi.. =64/Ri.' RL 1200 A'= J 0.3164 Re-O.2;
for rotor surface;
DW (.jl-i-)
Rei. -=

j,.J
-

Two methods are used for the numerical calculation, one is the method which is
shown in previous section, the other is one which is the finite diference method
that was proposed in Ref.(l). At first calculated results for these two methods
are compared with and as the result it is known that these two results agree.
Then parameter surveys of labyrinth seal are carried out to see the influence
on the flow induced force which exits the instability.
In this paper critical speed of a rotor system is assumed to be 5000 rpm,
and rotating speed of the rotor is represented in nondimensional expression as
W/Wn({D (rpm):rotating speed of the rotor). F and FAI shown i n figures mean the
absolute value of the force induced by labyrinth seal and its phase angle from x
axis.

212
a) Accuracy of Two Calculation Method
Fig.3 shows the spring coefficients and damping coefficients which are
obtained by two methods i.e. approximate method and finite difference method. In
the figure, full line and broken line show the coefficients for inlet pressure Po
=196.13(KPa) and 147.I(KPa) respectively and outlet pressure Pn=98.07(Kpa) which
are obtained by the approximate method and doted line shows the coefficient for
inlet pressure Po=196.13(KPa) and outlet pressure P3=98.O(KPa) which is obtained
by the finite difference method. Also sign (-) on line means negative value.
From this figure full line and dotted line coincide very well so each calculation
method has good accuracy.

b) Influence of Deflection of Steady

In this paper pressure, peripheral velocity and flow rate in steady state
are iteratively calculated, howeve Kostyuk and other authors obtain them by
another method and there are some discrepancies between both steady state
values. Moreover, steady state values in field and the calculated values may
have discrepancies. So it is important to investigate the influence of deflec
tion from steady state on the accuracy of the induced force. Then Fig.4 shows
the flow induced force and its phase angle when pressure is a little deflected
from steady state where the number on lines correspond to the number in Table 3.
The calculated data are shown in table 2. From the calculated result, it is
known that even if the flow rate is deflected 20% from the steady state, flow
induced force and phase angle are not so influence and these curves become like
one line, so this figure is neglected. On the other hand if the pressure is
deflected 0.5% from steady state value, they are influenced very much. Also it
is known that the induced force is not influenced by rotating speed of rotor but
if it is separated to component, influence of rotating speed is remarkable as
shown in Fig.5. From the above discussions it is concluded that in the calculation
of flow induced force, steady state values must be calculated with high accuracy.

c) Influence of Pressure Difference Between Inlet and Outlet

Fig.6 shows the influence of pressure difference between inlet and outlet
on the flow induced force when inlet (entrance) pressure is kept constant value
(209 atm) and outlet (exit) pressure is changed. The figure shows that as the
outlet pressure becomes large, i.e. the pressure difference between inlet and
outlet becomes small, the flow induced force b ecomes small. As the flow rate
decreases, the induced force p ropo rti onall y decreases.

d) Influence of Labyrinth Seal Radius

Fig.7 shows the influence of labyrinth seal radius on the flow induced
force, where the seal radius is valiable and other dimensions are fixed at
constant values. Here flow rate means total flow rate as shown in the expression;

From the figure the flow induced force proportionaly increases as the seal
radius increases. Therefore,the flow induced force due to labyrinth seal can be
easily evaluated by the similarity law, when the seal radius is known.

e) Influence of Seal Clearance

Fig.8 shows the relation between seal clearance and the flow induced force
and the leakage flow rate (leakage mass flow rate per unit length), where as the
clearance increases, the induced force F decreases and the flow rate increases
proportionally.

213
f ) Influence of Seal Pitch
Fig.9 shows the influence of seal pitch on the flow induced force, when
strip height is kept constant. From the figure it is known that as the seal
pitch increases, the induced force increases, but the leakage flow rate Q does
not so much changed. Thus the large pitch is not good from the view point of
the flow induced force.

g ) Influence of Number of Seal Chamber

Fig.10 shows the influnce of chamber number on the induced force and the
leakage flow rate, where inlet and outlet pressure is constant and each chamber
dimension is also same. From the figure the induced force increases like hyper
bolic as the chamber number increasing. On the other hand leakage flow rate
decreased like exponentially. Therefore optimum number of chambers may be
determined when inlet and outlet pressure are known.

h ) Influence of Seal Strip Height

Fig.ll( a )( b ) shows the influence of seal strip height. In this case the
induced force, phase angle and leakage flow rate do not influenced so much. But
if this result is represented in the spring and damping coefficients, the influ
ence due to the strip height is remarkable. Fig ll( c ) shows the spring coefficients
k and k
xx xy

i ) Influence of Divergence and Convergence Type Labyrinth Seal

Fig.12 shows a comparison of seal types, that is, comparison of straight
through type, convergence type and divergence type. In this calculation there
is little discrepancies of the flow rate Q, so its figure is neglected. From
Fig.12 ( a ), it is known that the flow induced force for the convergence type is
the smallest of all and divergence type is the largest inspite of the same
leakage flow rate. From Fig.12 ( c ), the flow induced coupling stiffness kx is
similar characteristics, but convergence type is excellent from the viewpoint
of st abili ty .

CONCLUSION

In this report the flow induced force due to labyrinth seal which is some
times the cause of instability is studied for the special model. That is, the
labyrinth seal of shroud of steam turbine is used as an example and its flow
induced force and stiffness and damping coefficients are theoretically calcula
ted by two method. Then influences ot deflection of steady state, pressure
difference between inlet and outlet, labyrinth seal radius, seal clearance, seal
pitch, number of seal chamber and seal strip height on the flow induced force
and leakage flow rate are investigated. As the results, it is known that the
above factors are very influential on the flow induced force. However further
theoretical and experimental investigation are reQuired in order to obtain a
more precise description of labyrinth seal characteristics.

REFERENCES

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

Turbines or Compressors. NASA 2133,May 12-14, 1980.

214
 TABLE 1. - LABYRINTH SEAL DIMENSIONS FOR NUMERICAL EXAMPLE

SEAL RADIUS Rs 1 (m)

ROTOR NATURAL FREQUENCY Wn 5000 (rpm)
STRIP HEIGHT h 10 (mm)
SEAL CLEARANCE 0 1 .3 (mm)
STRIP PITCH 1- 30 (mm)
NUMBER OF CHAMBER N 2

SEAL TYPE straight-through

type

TABLE 2. - GAS CONDITION FOR NUMERICAL EXAMPLE

OPERATING LIQUID air

LIQUID TEMPERATURE T 784.15 (K)
ENTRANCE PRESSURE Po 20482.0 (kP a )
EXIT PRESSURE Pn 19678.4 (kPa)

TABLE 3. - DEFLECTION OF PRESSURE AND FLOW RATE FROM STEADY STATE

PRESSURE OF EACH STAGE (kPa) FLOW RATE

PO PI P2 P3 (kg simi)
STEADY STATE VALUE (1) 20482.0 20166.3 19936.7 19678.4 6.8443
DEFLECTION OF (2) 20482.0 20267.2 20036.4 19678.4 6.8443
PRESSURE Pl,P2 (3) 20482.0 20065.5 19836.9 19678.4 6.8443
DEFLECTION OF (2) 20482.0 20166.3 19936.7 19678.4 7.5287
FLOW RATE (3) 20482.0 20166.3 19936.7 19678.4 6.1599

215
 casing

POo'
c,-, c. c

UtAl
rotor

(a) (b)

' i+1
B'B AA "+ C}c'&w
aw

I I i

i-l

" Jw
(C) (d)

////////////
PO Pl P2 P3
r---
Inlet Outlet
- -
'---

( e )
777777/777/7

Figure 1. - Cross section of labyrinth seal Figure 2. - Labyrinth seal model

and definition of the coordinate. for numerical example.

216
 Kxx=Kyy
3 Cxx=C yy
10 -'L======== ============ -
( )
e
-----------------------
"
( -) '10
\.)
E
o

. 10
....
2
j::b----------------I
Q)
o

m
-I

- ..... 10
I ____________________
10'

Kxy=-Kyx'
-
10 b----------


- -. --.-
----- -- -
---- ( -)
- --
-----
- --
-- - ---.
.
. .
--
.... - -- ----- -
/ .. .. ......
......
-
--
...

10

:
...

... b----------------
10----------------------
./ ----
..

,
"
. u

100------1--2--
-
I

1 2 3
NONDlMENSIONAL FREQUENCY NONDlMENSIONAL FREQUENCY

(a) (b)

( -)
"10 1
--------------------4
. Kxx9= Kyy9
"'0 -I
"'0
I'd 1'd10
'
'-
,
0
, 1 o
1::--- -----...., Q)
( -) Cxx9=Cyy9
o

..
10-- ------
--- - ---- -
: Kxy9':-Kyx9: _- ...... ----- -
--
. ... --- ----------
. -4 Cxy9': -Cyx9
.- -- . .
10
.

, '"
.............

o -
10
-I .. .... . ..
o
.

b---- ------
.-

;- ------
. u ./
,
':' ,'
.
.

-I ":" -s
10---
-:
-----2--3 100---------2--3
NONDlMENSIONAL FREQUENCY NONDlMENSIONAL FREQUENCY

(c) (d)

Figure 3. - Accuracy of two calculating methods and influence of pressure

difference between inlet and outlet.

217
 5000 -----, 10

4000 (3 ) 8

7
-0
(3)
III 6
I.... (1)
'-"

5 r-
(2 )
u. 2000 4
(2 )
-& 3
(1)
1000 2

0-------- 0 k- ----r-l--c---
NONDIMENSIONAL FREQUENCY NONDIMENSIONAL FREQUENCY
(a) (b)

Figure 4. - Jnfluence of pressure deflection from steady state on

the flow induced force.

E 5 (3).------------
1 ,0 1-----(_- -----
)
-..",...2"'-
( ..., )--- -- .- ..-
. . ... . . .
.-
...
.. -:

.
..,.m

.
. -
.

. ..
...... . .
/' /'
. . . .. .. . . ..
. .
. .

/' . .
.

. .. .
.
. .. (-)
..
4 I .
.

10 --------------------

10 1.-------L.-----2L--
- --3
0

NONDIMENSIONAL FREQUENCY

Fi gure 5. - Influence of pressure deflection from steady state on

the cross c oupling stiffness.

218
 10000 20 10
w/Wn = 1.0
9000 Po=20482.0 (kPa) 9
' " ..
....
8000 " ." . 8
" .
. ..
15
7000 ... 7
"
"C
1:
,... "q. "- rd
Z.6000 0
'-'
r..
6

OJ

5000 10 '-' 5
l.L
4000 4
'. -e-
3000 '." 3
5
2000 D"
2
1000

q @ q 0
EXIT PRESSURE EXIT' PRESSURE
(a) (b)

Figure 6. - Influence of pressure difference between inlet and outlet.

5000 10
OJ/wn=1.0
9
4000 100 8
,....,
E ,....
"- "C
7
0 rd
Z3000 '"'
r..
6

OJ

5
I.&. 2000 50 4
-e-
3
CI
1000 2

0
131::1 1 2
ROTOR RADIUS Rs(m)
.:
ROTOR RADIUS Rs(m)
(a) (b)

Figure 7. - Influence of labyrinth seal radius.

219
 2500 20 10
w/Wn= 1.0
9

2000 8
'15 ..

E 7
...... "0
It!
(I)
Z 1500 l-
S
F .....
C'l
q
"
10 5

u... 1000 4
-e-
" 3
5 IT
500 "
"
2
,.'

,. '

0 0
CLEARRNCE d (mm) CLERRRNCE Jemm)
(a) (b)

Figure 8. - Influence of seal clearance.

2500 10 10
W/Wn= 1.0
9 9

2000 8 ,...
8
q '"

.
. . .
. . ... . .
. .. . . . . ..
. . . . . . ... ....... . . . . .
? E
"0 7
. . . .
.. .
...... It!
(I)
Z1500 6 l-
S
C'l

5 ...., 5
F
1.J...1000 4 4
-e-
3 3
IT
500 2 2

0 0
00 0
SERL PITCH tCmm) SEAL PITCH ! (mm)
(a) (b)

Figure 9. - Influence of seal pitch length.

220
 5000 10 10
UJlwn 1 . 0
9
=

9
4000 '
.
8 ,.., 8
.... 7 "e ,..,
7
. ,
til
"'D
'"
z3000 '"
' .
6 L
..., 6
. OJ
.
.
..
.
.
5
..

5
..
..
..
...,
..
. . .
.
..
..

2000
.

LL 4 4
-e-
3 3
D'"
1000 2 2
1
0 era 0
NUMBER OF CHAMBER N NUMBER OF CHAMBER N
(a) (b)

Figure 10. - Influence of number of seal chamber.

1000 ...---- ---- ------..., 10 10

900 F 9 9
800 8 8
700 .... . " .............................................................. 7 "e
,..,
"'D 7
, '"
q
til
Z 600 6 L
..., 6
..... OJ
UJ IWn 1 .0
500
.

5
=

l- 5

lL 400 4 4
-e-
300 3 3
D'"
200 2 2
100 1
0':1 0, :J ll!::! l:J 20
STRIP HEIGHT h(mm)
(a) (b)

Figure 11. - Influence of seal str ip hei ght on the flow induced
force and phase ang le.

221


7
10 Kxx-=Kyy
E
"
Z
'"'10 r---- -- 2500 r---------------
straight-through type
divergence type
convergence type

5
10 - 2000
>-
x Kxy=-Kyx ,
4
10 --- Z1500
-/f-)
20 L.. 1000
5 10 15
STRIP HEIGHT h(mm)

(c ) 500

--- ------ ---- -- ---

Figure 11. - Influence of seal strip
hight on flow induced cross-coupling. 1 ;
NONDIMENSIONAL rREQUENCY
(a )

10
. ..
straight-through type . ...
... .
....
. ...

. .

divergence type
. . . ...
_
. ..
. .. . ..

9 convergence type .
.
..
.. .
. .

,...
8
4

,... 10----------------

"'C 7
III

/</ /
L
'"' 6
----------------- ---
3 ___ ---
5 10 F------------
---- --------
".--- - ---
4 /'
-e- /'
3
0 I=:"f'/-'--------------
2

1
I
0, 1 2 100--------2--3
NONDIMENSIONAL rREQUENCY
(b) NONDIMENSIONAL rREQUENCY

(c)
Figure 12. Influence of divergence and convergence type labyrinth seal.

222
 CIICUV.IFERENTIAL PRESSURE DIS1RIBUfIQ\lS IN A IVllJEL LABYRIN'lH SEAL

Y. M. M. S. Leong and R. D. Brown

Heriot-Watt University
Riccarton, Edinburgh EH 14 4AS
Scotland

SUMMARY

Leakage flow through labyrinth glands had long been known to affect rotor
stability but its effects hav.e been often underestimated. A research programme
to isolate and study this cause of instability has been initiated. Circumferential
pressure distributions are measured in the labyrinth glands with geometry approp
riate to the high pressure labyrinths in large steam turbines. Knowledge of this
pressure distribution is essential as it is this unequal pressure field that results
in the destabilising force. Parameters that are likely to affect the pressure
distributions are incorporated into the test rig. Some preliminary pressure profiles
are presented.

INTRODOCTION

OVer the last few decades the development of steam turbines and gas turbo
compressors has resulted in a steady increase in the energy density of the working
fluid. Thus the energy available for various mechanisms leading to unstable motion
also increases. The increase in power implies an increase in length of turbine
generator sets as these machines operate at a fixed speed (3000 or 3600 rpm). This
makes the machine less rigid and the operational speed often lies above the first
and second criticals. High level nonsynchronous vibrations have often been attribu
ted to instability ariSing from lightly loaded bearings. The light loads instability
problem is accentuated by the presence of a large number of bearings supporting a
typical rotor with rigid couplings. Operational practice ensures that each bearing
takes a load and/or the use of more stable bearing confiqurations, e.g. lemon-bore.
In extreme cases tilting pad bearings may be used. If the instability arises in
the bearings then the problem will be eliminated. However, there are other sources
of instability which may prove to be significant. Steam whirl associated with
differential clearance at the periphery of a blade row has long been suggested as a
possible source. The possibility of labyrinth seals inducing instability has not
been given the same weight of attention. The destabilising forces were considered
by Pollman to be inSignificant (ref.l). Furthermore it has been concluded that
instability arises if and only if the labyrinths converge, Alford (ref.2), or diverge
Spurk (ref.3) which of course led many to conclude as highly unlikely to occur in
practice. It is also argued that the pressure in the labyrinths would equalise and
hence cannot maintain the unequal pressure field which is essential in producing
the lateral forces acting on the rotor.

Recent experience has indicated that the dynamic effects have been under
estimated. Based on operational steam turbines, Greathead (ref. 4,5) has brought

223
 to light the influence of the labyrinth sealing glands, in particular at the high
pressure end ad j acent to the working fluid inlet, on the sub-synchronous response
and the instability threshold. The response is strongly load dependent and has a
major frequency component greater than half speed. Evidently the resultant
destabilising force re-excites the lower criticals . A further interesting observat
ion here was that the replacement of the standard journal bearings with tilting
pad j ournal bearings had detrimental consequences on the response. This indicates
that an instability which does not originate from the bearings canno t be eliminated
by the use of tilt pad bea rings .

NOTATION

p absolute pressure n number of labyrinth stages

absolute pressure at entry eccentricity ratio

absolute pressure at exit r radius of rotor

mean axial flow U rotor surface velocity

axial flowrate per unit c circumferential velocity of fluid

circumference in gland

flow coefficient v axial velocity of fluid

kinematic viscosity Re circumferential Reynolds number

( = (U-c)h/2v)
A change of flow coefficient at
l,2
narrow, widest clearance Re axial Reynolds number
a
( = 2v 6)/v)
.Jp density
Ta Taylor number (- Re / (h /r
R gas cons tan t
6 peripheral angle, measured from
T temperature widest clearance

horizontal coordinate ( . r6)

radial clearance

z axial coordinate
h height of labyrinth fin

Subscripts
f cross-sectional area of gland

i i-th gland
L' wetted perimeter of gland per
unit circumference
mean value

L" we tted periter of rotor per

unit circumference

224
 CURRENT STATE OF ART

Reports on labyrinth induced instabilities date back as early as the 19508

(ref.6). A physical explanation of the occurance of self-excited vibrations through
clearance flows at the blade tips and interstage glands was presented by Thomas
(ref.7). It is already recognised that ese vibrations were very much load
dependent. Alford introduced further understanding to the problem in his report on
aerodynamic exciting forc::;es on jet turbo compressors and turbines (ref.2). The
occurrence of a circumferential variation of static pressure acting on the cylindri
cal surface of a rotor particularly within labyrinth seals was put forward. The
whirl frequency was identified as the fundamental frequency of the rotor and rotor
support system whose whirl speed was about 40 to 50 percent rotational speed. The
whirl amplitude increases with power output. It 1s concluded that unequal radial
clearances at entry and discharge have significant effect on the excitation. Based
on a single chamber labyrinth a stability criteria was arrived at, with converging
labyrinth glands being destabilising. This stability criterion was contradicted by
Spurk (ref. 3) in that diverging labyrinths are destabil1Bing instead. In common with
Alford the rotational motion of rotor is neglected in the stability analysis and the
reason given for the contradiction is due to the further neglect of circumferential
flow within the labyrinth by Alford. Further literature (ref. 8.9) indicates that
lateral forces can be produced by parallel clearances when eccentricity exists, with
or without the occurence of spiral flow: and these lateral forces lead to unloading
of radial bearings.

Until 1975 approximately comprehensive experimental work on multi-stage laby

rinths is non-existent, and at best the destabilising forces are only defined
qualitatively. Among the first attempts to analytically quantify these forces
in multi-stage labyrinths was by Kostyuk (ref. 10) A set of fundamental equations
correlating leakage flow, circumferential flow and pressure distribution was arrived
at by applying compressible gas dynamics and conservation of energy and momentum.
Simplified solution to the equations was forwarded for cases with tilt although the
form of solution arrived at predicts no unequal pressure distribution for cases of
parallel eccentricity between rotor and labyrinths. The application of Kostyuk's
equations were extended by Iwatsubo and Kurohashi et al (ref. 11, 12). Iwatsubo
applied a finite difference method on the equations with the spring and damping
coefficients of the labyrinths numerically determined. With a simpler approach
Kurohashi et al applied an equivalent clearance to take into account of the variable
flow coefficient through the annular clearance of the labyrinths to arrive at a
pressure distribution and set of stiffness and damping coefficients. No swirl or
spiral flow was assumed nor considered. Although some experimental work was
performed the published data were found to be rather limited especially on the
pressure distributions. Pressure measurements were made at only two opposite
locations within a labyrinth chamber hence with the assumption that the pressure
peaks at such location.

e experimental work of cenckert and Wachter (ref. 13,14,15) represents the

first serious attempt to investigate the leakage flow induced forces on multi-stage
labyrinths. It was an extensive experimental programme on labyrinths on a high
pressure air rig. Various geometry, number of stages, inlet conditions, eccentricity
and realistic peripheral velocity of rotor were catered for. Pressure profiles were
indeed obtained on cases with a parallel eccentricity through the labyrinths and
swirl has an increased forcing effect on the rotor. However the published results
were mainly of integrated forces and only few pressure distributions were published,

225
pressure distributions with rotor rotation were not presented. As the interest of
the authors was towards gas turbo compressors the published work was based on
geometry and dimensions not appropriate to steam turbines.

It is this lack of directly applicable experimental results in open literature

based on realistic seal geometry with full rotational speeds that necessitates the
need for further research. This research programme on labyrinth seals was a natural
extension of existing work on rotor dynamics in the Department of Mechanical
Engineering at Heriot-Watt University. Work on rotordynamic instability included
the effects of high pressure ring seals on turbomachinery vibrations, fluid force
induced vibrations, boiler feed pump instability and a general rotor dynamic
computer programme evaluating forced response and stability. It is ultimately
hoped that the programme would incorporate labyrinth seal forces to extend its range
of applicability.

OBJECTIVE

The main objective of the rig is to isolate the labyrinth induced effects from
other rotor dynamic effects that can possibly arise. A rotor diameter approximately
half that of a typical steam turbine shaft is used. Nevertheless the rotor and
labyrinth dimensions should meet the requirement that the flow parameter of Reynolds
number and Taylor number are of the same order of magnitude both in the axial and
circumferential direction. As it is felt that the rotor peripheral'veocity is of
importance in the shearing of the leakage flow, a realistic velocity must be attained
on the rig, in this case of 100 m/s as in operational steam turbines. Consultation
on the above data and seal geometries had been made with United Kingdom based
manufacturers and public utilities.

The approach taken is to measure the static pressure distributions in the

labyrinth chambers as the flow passes through the labyrinth and to match these
measured distributions with theoretical predictions. An alternative method is to
attempt to measure overall force and/or stiffness coefficients directly. However
the literature has a number of examples where considerable scatter was demonstrated
in attempts to measure stiffness and damping coefficients. Moreover circumferential
pressure distributions give a better inSight into the nature of the problem. It is
interesting to know how these pressure distributions are affected by eccentricity,
clearance, speed and a host of other parameters. When an adequate match has been
achieved between experimental and theoretical circumferential pressures, the
required dynamic coefficients can be obtained theoretically. The integrated
pressure force for overall perturbed conditions yield the appropriate stiffness
and damping coefficients.

DESCRIPTION OF RIG

The experimental rig consists of a main casing enclosing the rotor and
labyrinth assembly, figures l, 2 ,3. The main casing is made up in two sections with
the upper section easily removable to enable the labyrinth assembly, especially
combination and stepped labyrinths, be built up conveniently. Slots are milled on
the casing adjacent to the labyrinth assembly to facilitate the mounting of pressure
tappings in the labyrinth. A vertical rotor is mounted on roller ball bearings

226
within the casing. 'lhe top bearing housing is bolted on to the main top cover plate
and the bottom bearing housing on the main platform on which the casing is bolted.
Drive is by means of a horizontally mounted 8.5 kW variable speed DC motor, with the
driving shaft attached to a 1:1 right angle conversion gear box. 'lhe rotor is
driven by a timing belt from the gear box with the necessary step up in speeds
achi ved through the pulleys giving speed ranges from 0 to 7600 rpm .

Consistent with the need t o isolate other rotor dynamic effects a rigid rotor
is used. 'lhis eliminates the influence of hydrodynamic bearings as well as making
it easy to preset and maintain predetermined eccentricities. Mounting it vertically
eliminates the need to consider gravitational effects in the experiments and
subsequent theoretical work. Grease lubricated self aligning ball bearings are used
at the top to allow for small relative misalignment between the rotor and main
casing for tilt investigations. To take the high downwards thrust a matched pair
of angular contact ball bearings is used in tandem. 'lhe bottom bearings are oil
lubricated from nozzles directed on to the cages and races allowing them to be used
at the higher speeds. A high pressure thrust compensating dummy piston could have
been incorporated into the design but was not as conservation of the available
working fluid was a major consideration. In order to achieve the peripheral
velocity of 100 mIs, a drum of 240mm in diameter is mounted on to the main shaft to
form part of the rotor. 'lhe drum is interchangeable and the xperiments will involve
the use of a plain drum and grooved drum for various labyrinth combinations. 'lhe
axial length of the drum is 160mm to accommodate the axial length of a complete
labyrinth assembly. 'lhe main shaft and drum are first balanced individually and
finally as a completed assembly to ISO G 1.0 (1 qm-mm per kg rotating mass)..

'lhe working fluid is fed into a diffusion box in the main casing. Within the
diffusion box deflection vanes and perforated holes spread the fl ow. From here the
flow enters a plenum chamber where the flow is further settled. 'lhe working fluid
used is compressed air at a gauge pressure of 5.52 bar although higher pressure or
higher density compressed gas can easily be adapted for use on the rig. 'lhe flow
rate through the rig is measured by an annubar. It has four dynamic pressure sensing
holes facing the stream, and the dynamic pressure being averaged by means of an
interpolating inner tube. A further pressure sensing hole measures the static
pressure less the suction pressure which eventually gives a differential
pressure representing the mean velocity along the tube. 'lhis measured flow rate is
compared with the extensively used and tested semi-empirical labyrinth leakage
equation of Vermes (ref.16) as well as equation 1. (ref. 10,12). Axial and
circumferential Reynolds number and Taylor number of 33 x 103, 63 x 103and 17 x 103
respectively can be attained on the rig as compad to" typical values of 222 x 10 ,
3
124 x 103 - 216 x 103 and 14 x 103 - 35 x 103 respectively on operational united
Kingdom steam turbines. Nevertheless one of the main current constraints on the rig
is due to the limitations of output from the compressors of the Department. Just
prior to entry into the labyrinths the flow enters a set of swirl ring mounted on top
o
of the abyrin
a assembly. 'lhese rings are interchangeable with vane angles of 45 ,
o
30", 15 and o . 'lhese swirl rings can hence pre-induce a spiral flow prior to
entry nto the labyrinth assembly. As the rotational direction of the rotor can be
reversed the case of the pre-swirl in opposite direction to the rotation of the
o
rotor is also possible. 'lhe 0 vanes are used as a flow straightener when no pre-
swirl is required.

Depending on dimensions, the labyrinth assembly is built up from interlaying

individual seal fin and chamber or as in integral individual labyrinth, figures
4,5. 'lhis allows interchangeability of seals with different geometry, sizes and

227
clearance. The current sizes used for straight through labyrinths are 4.064mm
(height) 6. 35Qmn 'axial length) and 9. 525DD1l X 9. 525DD1l being the largest size to be
x
used; with clearances of 0.254mm - 0.635mm. Stepped and cOmbination seals of
4.064DD1l/4.62Omm x 6.35Omm are also used. Further combination of labyrinth dimensions
can be catered for by mere manufacture of the fins required. The number of stages
can be altered by removing or adding appropriate individual fins. 'lbe current
number of stages considered total up to 12 labyrinth chambers, although more can be
added. CUstom made pressure fittings are mounted on recesses in 5he individual
labyrinth fins to measure static pressure circumferentially at 30 intervals in each
and every stage down the labyrinth stack.

The eccentricity of the rotor with respect to the labyrinths can be set at any
predete%1Dined location. 'lbe labyrinth assembly together with the main casing as a
whole is moved laterally across the main table platform, with the rotor and its
bearings housings being fixed in its original location. 'lbe movement is achieved
by means of lateral finely threaded jack screws, with the main casing moving in
guide blocks. To minimise the effort required to move the job, four sets of ball
bearing thrust pads are brought into play raiSing the main casing and assembly off
the table platform whilst being moved. Lateral adjustable pre-compressed springs
are mounted against the casing to eliminate any backlash on such small moveme nts.
The eccentricity is checked by means of displacement proximators as well as dial
gauges. Other than parallel eccentric movement a relative tilt between the rotor
and labyrinths can be set. Bere again the rotor is left undisturbed and the main
casing with the labyrinth assembly tilted. This is so achieved by raising the main
casing off the platform and an angled tilt pad inserted betwn the main casing
and platform.

The experimental parameters are

(i) seal geometry and types: straight through, stepped and combination labyrinths
(ii) number of stages
(iii) relative tilt between labyrinths and rotor
(iv) flow rate and inlet conditions, including swirl
(v) eccentricity
(vi) rotational speed of shaft.

At each setting of these parameters static pressure measurements are taken as

the leakage flows through the labyrinths. The current total of 144 pressure
distribution measurements of the labyrinth assembly as well as static mld dynamic
pressure measurements at the inlet and outlet are fed to a bank of mechanical rotary
swi tches These switches allow 5 pressure lines, on the inlet ports, to be served
by 1 pressure transducer at the outlet port of the switches. At each inlet port
setting a scan of all the pressure transducers are taken and sequential switching
. of open inlet ports gives a complete pressure mapping. As static pressure measure
ments are of interest here this arrangement would be more cost effective than
electronics based scanning valves. 'lbe working fluid inlet and outlet temperature
from the labyrinth assembly are also monitored by thermocouples to validate the
theoretical assumption that the two temperatures are constant. The scanning of the
pressure transducers as well as thermocouples and proximators are activated via a
data transfer unit, giving output in forms of read.1.Ags on the digital voltmeter and
paper tape for final processing on a digital computer.

228
 SUMMARY OF THEORY

and the working

Assuming an isothermal process
in a multi-stage labyrinth glands
fundamental equations
yuk (ref.lO) derived a set of
fluid to be a perfec t gas, Kost
as follows:

(1)
>=

aP a(P c ) 1 (2)
i i i
+ (q - q + )
i i l
... -
-

ax f
a1; i

aC
ile. i qi+1
1. + (c - c )
P + Pi ci i+ 1 i
i a f
-

ax
--
t i
ap
2 2 i (3)
+ K' P c - K" P (U- c ) -
i
-

i i i ax

)" L' )."L" (4)

sign c
t
Ki .. Sign (U - c )
K' i
i
... -

2f 2f
i i

(5)
).' = fn(Re' ) ). "
.. fn (Re")

(U - c )h
cihi i i
Rei Re"
- ...
=

2v v

(6)

The analytical solution to the above set of equations is simplified by

defining an equivalent clearance change (ref.12)

(7)
(1 + )Yi
Y(e ) i
=

!i(t.1l2 - t.lli) (8)

where i
0:

229
Subsequently giving a pressure distribution

. 2
Dr. c Or
- i
- -

- cos 6 + sin 6
Pi (6) Pi
=

f. f.

1
Dr. cos 6
2 [K'C +.K" (0 - c) (9)

f.

q*
where 0 =

2rrr
i
[ (10)

A possible alternative approach is to apply modified bearing theory to

labyrinth chambers. Pinkus and Etison (ref.17) has shown that the Reynolds Equation
in a modified form can be used for finite- cylinders rotating within moderate
clearance ratios, say between 10-3 to 10-1 which cover likely values for labyrinths.

ADDITIONAL MATERIAL PRESENTED AT THE WO RKSHOP

All experimental data presented in figures 6 17 are for straight through seals
-
with mean radial clearance 0.63Smm, axial pitch 9.525mm and labyrinth depth 9.525mm.
Figure 6 shows the pressure reduction as the working fluid is throttled for a
concentric rotor. The resulting circumferential pressure distributions are shown
in Figure 7. The expected unbiased pressure distributions are obtained except for
slight deviation in chambers 6 and 7. This indicates that any biased pressure
profiles would be due to eccentricity, speed or other experimental parameters and
can be taken as significant.

Figures 8-11 give the circumferential pressure distribution down the stages
for cases of parallel eccentricities; = 0.4, 0.6 and u = 94.3, 37.7, 0 m/s.
A restoring force type of pressure distribution with a slight trough is observed
in the first chamber. However subsequent chambers give a distribution with minimum
pressure just before the minimum gap. These give rise to radial negative spring
forces. The profiles are similar in form but with values scaling up with increased
eccentricity. Apparently peripheral speeds do not affect the distributions except
for the first chamber where the trough increases with speed. Figure 12 shows the
pressure ratio for all chambers on an absolute scale for a case seen earlier
= 0.6, 94.3 m/s. It is pointed out here that the experimental points are averaged
values of readings taken continuously for each pressure measurement. Scatter are
within + 2% and extremely good repeatability is obtained. It is obvious here that
unequal-Pressure field dos exist for parallel eccentricity cases. The obtaining
of pressure profiles beyond the first chamber shows the limitation of the 'equivalent
clearance' pressure prediction of Kurohashi which gives a zero differential
pressure field beyond the first chamber.

If the pressure distributions are numerically integrated as defined in the sign

convention, figure 13, radial and transverse forces can be obtained. Note that with
the sign convention adopted a positive F indicates a negative spring force.
y

230
 Figure 14 gives the radial and transverse forces in individual labyrinth
chambers for u : 94.3 mls and = 0.4, 0. 6. The first chamber has a restoring
radial force, and beyond this chamber negative spring forces are evident in all
stages. These radial forces vary with stage number and eccentricity in a
consistent manner, in particular an axial wave variation is clearly present. The
transverse force tends to be small and fairly insensitive to eccentricities.

The pressure distributions, as seen earlier, are fairly insensitive to speed

and obviously so would the resulting forces, figure 15. Figure 16 gives the radial
and transverse forces for several eccentricities. The radial forces are dependent
on eccentricity and the transverse forces much less so. Figure 17 gives the
summation of radial and transverse forces plotted against eccentricitY,with the
linear relationship of F with eccentricity seen.
y
The above results presented are some of the experiments done with the recent
commissioning of the test rig. Obviously the full experimental schedule is yet to
be undertaken and it is hopeful that meaningful results will be obtained. The
effects of these forces will be incorporated into a model of a real rotor using
the rotordynamic computer programme of the Department in the near future.

REFERENCES

1. pollman, E., Schwerdtfeger, H. and Termuehlen, H.: Flow Excited Vibrations in

High Pressure Turbines (Steam Whirl). Trans. ASME , J.Engng. power, April 1978,
pp. 219-228.

2. Alford, J.S.: Protecting Turbomachinery from Self-Excited Rotor Whirl. Trans.

ASME, J. Engng. Power, October 1965, pp. 333-344.

3. Spurk, J.H. and Keiper, R.: Self-Excited Vibration in Turbomachines resulting

from Flow through Labyrinth Glands. C.E. Trans. 6785. Translated from
Ingenieur Archiv. Vol. 43 (1974), pp. 127-135.

4. Greathead, S.H. and Bastow, P.: Investigations into Load Dependent Vibrations of
the High Pressure Rotor on Large Turbo-Generators. I.Mech.E. Conference on
Vibrations in Rotating Machinery. September 1976.
5. Greathead, S.H. and Slocombe} M. D.: Further Investigations into Load Dependent
Low Frequency Vibration of the High Pressure Rotor on Large Turbo-Generators.
I.Mech.E. Conference on Vibrations in Rotating Machinery, September 1980.
6. Den Hartog, J.P.: Mechanical Vibrations, McGraw-Hill Book Co. Inc., New York,
fourth edition, pp. 295- 296

7. Thomas, H.J.: Unstable Oscillations of Turbine Rotors due to Steam Leakage in

the Clearances of the Sealing Glands and Blading, C.E. Trans. 6281. Translated
from A. I. M. Bull. Sci. (1958) Vol. 71, Nos. 11, 12.

8. Rosenberg, C.S., Orlik, W.G. and Marshenko, U.A.: Investigating Aerodynamic

Transverse Forces in Labyrinth Seals in Cases Involving Rotor Eccentricity.
C.E. Trans. 7083. Translated from Energomashinostrojenie Vol. 8, 1974,
pp. 15-17.

231
 9. Orlik, W.G., et al: 'lb.e centring Effect in Labyrinth Type Seals and its Effect
on Low Frequency Vibration of Turbo-Machines. C.E. Trans. 7104. Translated
from Energomashinostrojenie Vol. 10, 1975, pp. 25-29.

10. Kostyuk, A.G.: A Theoretical Analysis of the Aerodynamic Forces in the Labyrinth
Glands of TUrbomachines. 'lb.ermal Engineering Vol. 19 1972 (11), pp. 39-44
(Teploenergetika 1972, 19 (11) pp. 29-33).

11. Iwatsubo, '1'.: Evaluation of the Instability Forces of Labyrinth Seals in

Turbines or Compressors. Symposium - Workshop on Rotordynamic Instabllity
Problems in High-Performance TUrbomachinery. Texas A&M Oniv., May 1980.
.
NASA Conference Publication 2133.

12. Kurohashi, M., Inoue, Y., Abe, '1'. and Fuj ikawa, '1'.: Spring and Damping
Coefficients of the Labyrinth Seals. I,Mech.E. Conference on Vibrations in
Rotating Machinery. September 1980.

13. Benckert, H. and Wachter, J.: Investigations on the Mass Flow Induced Forces
in Contactless Seals of 'l'Ilrbomachines. Proceedings of the 6th Conference
on Fluid Machinery, Budapest 1979, pp. 57-66.

14. Benckert, H. and Wachter, J.: Flow Induced Coefficients of Labyrinth Seals for
Application in Rotordynamics. Symposium - Workshop on Rotordynamic Instability
Problems in High Perfomance 'l'Ilrbomachinery, Texas A&M Oniv., May 1980,
NASA Conference Publication 2133.

15. Benckert, B. and Wachter, J.: Flow Induced Constants of Labyrinth Seals. I.Mech.E.
Conference on Vibrations in Rotating Machinery, September 1980.

16. Ve:rmes, G.: A Fluid Mechanics Approach to the Labyrinth Seal Leakage Problem.
Trans. ASHE, J.Engng. Power, April 1961, pp. 161-169.

17. Pinkus, o. and Etsion, I.: Analysis of Finite Cylinders Rotating Within
Moderate Clearance Ratios. Trans. ASHE, J. Fluids Engng., June 1976,
pp. 156-162.

232
 -
FLOW
MfASUAEl1ENT

DATA
"TRANSFER
UNIT

WORKING
FLUID

THERMOCOUPLES

Figure 1. - Schematic layout of test rig an d 'instrumentations.

Figure 2. - View of overall test rig.

233
 o 50 100 mm
I I

Figure 3. - Sectioned assembly drawing of test rig (main assembly).

234
 ---- -----

(a ) . - Straight through labyrinths.

---+--- ---
---- ----- ------_.-

(b) . - Combination labyrinths.

Figure 4. - es of built-up labyrinth stack.

Figure 5. - View of built-up straight through labyrinths.

235
 e .0 ma12

II 0 mls p... /p.,' 003S

'--I"---r----I--'
=

1.0 P1- p..

I o Swirt Q
.. .295 bars
P.-P.. -p;:p;

ii '\'
p. - p. Po
"I I"r--
I I
1.0
. 9 ----+--- ____

.8 I
"
'" ..9
I I

"l-- t'\.--
P1
I !
'- '--/
", p,'1'
11
. '' -0
.8 , P3 '
1--""'---
'
1 i
or-'
I
"+-
'
I
. --i-
i

I" P
.. 0 mi.
I I !
I . :, I I '! .-- Ii
\ i ,
!
.6 I 0 Swirl
.7 : - i
. ._--r--.-
! . : : i
iNo.of_I.,m.12
I
---- +-

Ps I
,
-

i I
0 1- ,
....... ,....... ,I --- ._,....... ..- ..-.-.
p....tlPi I I I
0.35 I,

, : I I
til
til .5 P : 2.9 bar
l:l
.6
1= a-.:t=
' :d+-- I --
.....P....l.- 1 i_I i ll-
.

' I
i

tl :;, '
Ilo
--
\:. ' : 0.835_. til
til i I I

-
P7

I'"J ill t--. ,: "-'-t-"-"---'- -'''I

(,.0.)
i
-------.i..-----'---i- ----- Ilo .5 ' -r------

I
.4 ! .
0\
r---l-- ..
.
i P,
i----I---" 1
i I .--- .
I
! I
4
i : I
.3
1 \
P9
'
T .1
'I .+== . " . __
I
, I
3

-
1\
t.!"
.21 , I

: --1
' 1
I ' I i , " :
I '.....,.,Ij a-'

-.. J- : I
: I !
.2 ; ' ---,. --_._------+-----
,

, :"\.-
.1 ..---
l

;
---_. _ -- - -
J

-
,

1
, i P" I i
'-t--'
, '
foI::u....""'-'
. '.,..-. , ,I ..
'
____
I L...___
o . "
r--
_

o 2 3 4 5 6 7 8 9 10 11 12 I
LABYRINTH CHAMBER NUMBER

o 60 120
I 180 240
I
300 360
0
PERIPHERAL ANGLE
Figure 6. - Pressure expansion through
the sages, concentric rotor

Figure 7. - Pressure distribution. concentric rotor.

 o
3
I :
t----i-- . '
..;.--1
.
.
'
: :0' i
'I
-+I Po' 2.932 bars
o

3oU--! aj

; '.
I
I .,-.-
i ' :

L J
I) :i ', ;; I
! \ I. '
Ap
P
15 ; ; 15 !'.
Cop
(Ref.)
(Reference)
I :

_':-'.'
: : 2.640
0
"
2,669
0 '

ri pni '.

]L
30 u=9It.3m/s 30 I uz94.3m/s
, 2 '

.
:
Co' 0.4 'i :
. , , t .0.6
15 0 1S
Y.
i

1';: ' !
i ;

I 'P.
....... s
Swirl. ;
'T'...... .12 i'-
.
...... , i
.
m
o 2.620
30 r _ ..

30
I
: 2.587 =12

:
,
m

.Q
e
15 ......
.'
'
,i
..
P",1p 0.345
.Q
e 15 ' V
",
I
PM/PO = 0345

2.468
' . ,./ i
o
.... ....... , _

'
, '"""---
,-- 2.499 '
'
r' 1 ;

" .I
30 30

;:) I P4
.

CIl ;:)
'

CIl

U
"

---
CIl ...... ""..-

'" 15
f---

. '

.
CIl

'"
15 ' Y '
.
, "' ,
"
'

2.368

!
o .'
....... .-...._._--
, ,""'--__=_...... ',
...

30
' ::.1
...
30 t- i ; Ps

:
'
j
j
fil
N
'

15 P
s
W
... t _

E --
1::
.....
S, ...
15 " ' ,/"r
.
.... ..
2.248
, Q
, ,
.

o
'

2.271
30 30 .
' '
i
i

I
!
i
I

;
I
, P6
.
'
'

15
l-- 15

't---:-.r
-
:

j
1'
,
:
", ;
.

t",
o ....... __ . i
30

, /r
30

l--{,
_

f'"
' p

i.
i
15

,,
o
" --. ...... ' .
, 1,990

15
r- I I

P
s
30
o

r- ' "
,
.
I

I. "
"
.'.

'
Pa

; ---'
15 .; ...... .
""
. '

i
o
'......
_ .-.
__ 'I

300
I

360

60
il
l I 120 110 240
I I
300
i
360
120 180 240 o
PERIPHERAL ANGLE
o 80
PERIPHERAL ANGLE

Figure 8. - Pressure distribution. paralll Figure 9. - Pressure distribution, parallel

.
eccent ricity . eccentricit ;
 '
30 -J.- -A.I '-'F
o

. -1 I
.
.
- ..
Po

I
; . 1.837 bars Ap

30
0 -:--r-'r--r'ra..!-.'
I .,-ar-..J.
.. 2.819 bars

Ap 15
L1 ' l
"L 11
P1
, (Ref.!
15
-"r
'I
1
1..0'
Pl
I
I
I
,"-
["" (R.f.!

I 1 iiiI ,L-i ';"'11

! '

; : /IS
-' 2557 i --' 2.538
, ,: I: ,
i '
..
I

I
i

I
o

.....
u 37.7 ",Is

-
..
P2
30 I 0
....
... : , e.0.6 3
a

i : I""'" &.
'J-,
, .

-t
Swirl 0 15 i Swirl=O
. I ",.12
p
15
'l ./i ,i
1 :1 i i:.
i '" .12
:

r-i 1513
. I
,___
I

2.1.97
I j
i
I
;

-"1
I
o

30 i ! 3
!
j i i: P31
i I p

I I ,';ar
'

".,
--+"
15 i 1

1---_.:..-....... 1
t--... , i
f ....... 5
I

' !/ 1 I

r'
2.399 2.382

-_t
I 0 : ! I '
fl!Q 30 ; : !
! P4 0
I ; p11
.... lUOl
3

G'
; l

"L-
15

; V!
til
i.
I
. , , :

'-----.!

:.
: r!
"
en lS
fl!

,
I

2.292
;
___

1 -+
, ,..-- I .

I '
. ___

N fl!
e.
:z
3

15
-......
,
,
; , ,

!
I

i
.-- :
1
rs ....;..-. _
" ;

i:: r. i i
i .r :;.i. .i'

h
I
: 2 .1 8 "-!--l
VJ

5
rr:I

Eo I
00 rx.. ; - I & : to.

2.110

'' i : r--L Ii:

!

l.
0

30
II
. . iIp
:
':
. ........ 1 -t1 1047
j.
lA-
15 i
1 - I I _I
:

: . ' i""'" ,

: f" ' 1
1057 i'--
'

30 f a , '
'

P7/
"-:
i

I.

: /
:
r;'
,I
15
/.
/' ; I
1"
, :
;

j' : ! i--t-
: ..--' ,
.t-.-. 1.924

i ' '''
i
'

30 30
I,'.
i

t
i I .
i '-!'- - ' !
.-....- , , P Pa
"' !
j
'
15
: ,. ! I
i
15 i __ r-
""," ....... -r--
I

,
1
4....---r
'
: 1
1.787 . : ''l'--.+-" . 1.778

I i
60 12Q 180 ,240 300 360' 60 '10 180 140 300 ;60

Figure 10. - Pressure distribution, parallel Figure 11. - Pressure distribution, parallel
eccentri city. eccentricity.'
 f .0.6 m . 12

P,. u .94.3 m/, p'/p. 0.345

P Pili

Fy

Pfli Swirl .0 p 2.91 bars

1
j i 0 I !
...-......... -t.,....... , ...... .,.- ......
-=

1.0 t .-

. =J --
,
.9
1
. I '- ' '. '-- .
fT
_ 2 .
_P
_.:
.8
,..... - -
----.
.. :
Fx
.------
+ I I
i

.
s .7 t== -t--!.-
.
&--f-4--. -

i I--L_'"_...:.------
.1

I/)
I/) .8 Jo;;o::; .; ' .
...6---... ....... -.
.-...-...

N 9
W ' : ' . II

=1
\0 . .
t __.ti__. ,-'-

t=
.7
.5 1 ________
'
. 8
... -- t-:--.--:-
: 2!l
I ' ...,.9....
.
.... .. ----..............---
C

J
T

F r to p COS a de
Y
,3 1' -"--__

, . :
_____

J
'
1--_ ---10
.......... -' ...... .:.....-- 21t

. 2 t- ----- --------; Fx r to p Sin a de

11 1 , ........
0
J.--..-,- . ,,--, ,
.1 1 --I
I

o
eo 120 180
I '
Figure )3 Sign convention for eccentricity
.

PERIPHERAL ANGLE and forces.

Figure 12.
 F., ESJ
i
c OJ.

N 7
[Zj e:Q.6
e 10
o !. 0.4
5 1ft :12
a f.. 0.6
Slfirl 0
----l
=

4 8 0.35
p./P. =

!
3
F., c
a
2 6
N a :

0
Q
o 4

-1
LF/""4:26-4N U : 94.3 mls
m : 12
-2
\" ",6
: 41.0 N
2
L Fy S"irl = 0

-3 P..IP, : Q.345

1 0 I 1 1 2 i 3 I 4 I 5 i 6 I 7 i 6 I 9 110 i 11 i 12 o 2 3 4 5 6 7 8
LABYRINTH CHAMBER NUMBER x103 RPM ( 103 RPM ' 1257 MiS)

N
+=' RADIAL FORCE IN INDIVIDUAL LABYRINTH CHAMBER
o

4
0 ! = 0.4
Fx
'" = 12 a !':Q.6
F.
0
W E :0.4
2
Slfirl =

N
\' [7'.2.2 P./p' :0.35
4
N : -4.5 N
4 u :94.3 m/s
L Fx e.: 0.6

[Fxo
m =12
3 = 4.2 N
Swirl = 0

2 P./p, :0.345 o a 0 C
0-0 Q a

.
- -- ---- .......!..- -'-
-2
---',
o XX7'jV_OVV:7'j \/v"/<: xj"/\'1x 7<,v"'7 <:J\7U
1

3 4 5 6 7 8
2
-2
o
x103 RPM

o I 1 2 3 4 5 sl7 8 9 101 11 12

LABYRINTH CHAMBER NUMBER Figure 15. - Fy F (chamber 7)

x in typical chamber
\ Figure 14. - Transverse force in individual
versus speed.

labyrinth chamber.
 o a. 0.&

L_ J
u.O m/s ..... -...,
10
F,
e= 0.6
m .12
L. ... ...j
. . . 0.4

[J
Swirl =0
N
:- .. ,

to= 0.2

8 /p. "0.35
r--
.---

r--
r--

..
--

-- --- .. 4________________________ ==____-,

4
.. -

L . .. . ....
. r n
.
.
. .

,
Fx u=Om/s toO.8
m=12
0
; __
10
_

___ '" 1:.0.6

11
.
4
. p-.'
--- - .

. ... ... .0.4

:-. r . j
N
.

- ---' t ... ..... 1

. .
0.35
I
a
2 2

r F-- l
-ll rl
[_J : o. 2

0
n
-2
H
[ 1 ..... ...
. .
-2_____________________________ ____

3 I 4
I I
CHAMBER NUMBk R
5 6 i 7 I 8 I
I
9 110 I 11 112 i
-4
u
rjJi ! I
(bl. Transverse force.
I 0.1 I 1
-

1 2 3 1 4 I 1 I I i
CHAMBER NUMBER
5
6
7 8 9
10 11 12

if

(a). - Radial force.

F igu re 16. - Radial and transverse forces in individuial labyrinth chamber.

60
IF
0 [F.
(N)
u=o mi.

50 rf
m=12
D
&wlrl=O

PIA'Po = 0.35

-10

.0 .2 .4 .6 .8 to
ECCENTRICITY

Figure 17. EF ' EF vs. eccentricity.

y x
-

241
 NIEASUHBVlliNT AND EVALUATILN OF S'WlhL-TYFE FLUrv IN LABYhINcIH SEALS

OF (X]\lVENTICNAL TURBINE STAGES

.. L. Hauck
. Institut fur Thermische Kraftanlagen
Techn ische Uni versitii t !Vlunchen
8000 lVlunchen 2, Feoeral Republic of Germany

SUMMARY

Previous studies of rotor whirl instability pointed out two different

mechanisms inducing lateral forces within non-contacting labyrinth-type
seals. Ev aluating the part of exciting forces that results from leakage
losses is possible by means of variable flow contraction in the gap of
a seal with variable clearance at the circumference. The dominating
part of exciting forces has its origin in unequal pressure distributions
at the circumference of the seal with a shaft in out-of-centre position.
The pressure distribution is mainly influenced by swirl-type flow at seal
entry.

The installation of a computer controlled cantilevered pitot cylindre

within the seals of two different test stages enables the measurement
of flow conditions within every cavity of the labyrinth. The effects of
load factor and rotor eccentricity were determined on flow conditions in
test series for both stages. The results indicate that swirl-type entry
flow follows the rules of potential swirl. Within the labyrinth cavities
two spatial separated flow areas are to consider. A dominating flow in
peripheral direction nearly fills the space between the sealing strips
and the ceiling of the cavity. Below this flow, an area of axial mass
transport is situated, with a slight peripheral component, limited on
the nearest surroundings of the seals gap and the rotor surface. Between
both flows, an exchange of energy takes place. Within the gaps, flow
direction depends on axial velocity and therefore on variable flow
contraction. A balance of energy within the seal and the cavities
interprets the results of lateral force measurements as an influence
of friction at the sealing strips surface and the rotating shaft surface.
Stages with their blades put together in buckets by means of shrouding
segments are particularly influenced by the rotating speed of the shrouding.

One method to describe the flow conditions more closely is dividing

the leakage flow in stream tubes with varying cross sections at the circum
ference. This enables a one-dimensional view on the flow along every
stream tube. Additionally, two flow areas within every cavity must be taken
into account. Therefore, the local profile of the tube is unequal to the
geometric boundary of the labyrinth. Other effects, as rotating shaft, flow
.contraction and compressible flow through the labyrinth may be considered
as well.

242
 INTRODUCTION

Non-contacting seals are already known as a possible reason of self-excited

rotor vibrations in_turbomachinery. A large number of papers (ref. 1,2,3,4),
based on theoretical approaches and experimental results, pointed out two
different mechanisms as origin of exciting lateral forces.

Exciting Forces due to Leakage Losses

A change of leakage losses at the circumference, caused by the variable gap

resulting from rotor eccentricity, leads to a change of blade force at the
circumference (ref. 1). The resultant exciting force is given by the integral
of local blade force dU at the circumference

Q = 0
f2Jl
dU cos\(> = - 2 ojr..sJ\(
U 2Jl cos \(> d4> (1)

where \(> is the circumference angle. Following equation (1) tangential

forces can be expressed by isentropic tangential force Uis and the leakage
loss sP The leakage loss is determined by the geometric boundary of the seal
and the performance of the stage at a certain operating point. This is taken
into account by the constant factor C.

(2)

Additional, the leakage losses depend on local clearance s and the flow
contraction factor a. Both change at the circumference.

While geometric clearance can be easily put up, the determination of

varialbe flow contraction is still difficult. Therefore, flow contraction is set
up as a global constant factor a*applying to centric rotor position. Flow
contraction causes a higher uniformity of gap clearance at the circumference
with a rotor in out-of-centre position. This is considered by the effective
eccentricity e .
eff

(3 )

The uniformity factor v (ref. 5) is put up using flow contraction in

F
the narrowest and the widest gap.

(4 )

According to these relations, exctng force can be calculated by using the

isentropic tangential force, the eccentricity, the uniformity factor and the
flow contraction.

243
 (5 )

Exciting Forces due to Pressure Distribution

If fluid flow through the labyrinth is superimposed by a peripheral

flow component, a dominating lateral force arises from the pressure distribution
in shrouded stages (ref. 3,6,7,8). With dp for the local pressure, z the
coordinate in axial direction and b the width of the shrouding, the exciting
force is given by

b 2J1

Ia!
b 2J1

Q =
-
o Ia! dp sin =
-
o p(z,'P) sin r d dz (6)

Calculating the pressure distribution is problematic, since the location

of maximum pressure depends on swirl-type flow at seal entry. A complete solution,
like in the case of leakage losses, is not yet possible.

DESCRIPTION OF TEST APPARATUS

Investigated Stages

The flow patterns within the labyrinth seals of two test stages were
measured, to check existend methods of calculation and set up better resolutions
of calculating exciting forces.

Figure 1 shows the investigated turbine stages. The stage marked with the
number I is a reaction type stage. As may be seen from the blades profile, stage II
is a impUlse type one. Besides this, the shrouding of stage II is closed, while
the blades of stage I are put together in buckets of three or four blades. Further
data of the investigated stages are gathered in table 1. Both stages haVre stepped
seals that are similar in its geometric shape.

Distribution of Measuring Positions

Figure 2 shows the labyrinth seal of one stage. The cantilevered pitot
cylindre illustrated in the figure was used for flow measurements. The revolving
cylindre drive was fixed in ball-bearings and designed to measure flow patterns
in every cavity of the labyrinth seal. The structure of the drive can be seen in
figure 3. The gear box includes two different drives. Revolving the cylindre is
done by drive (c), while driJve (d) moves the cantilevered piitot cylindre to ilts
axial position. Moving from one cavity to the next is done by drive (a). Before
revolving the gear box" the pitot cylindrp must be lowered into the box. Then the
who e drive is raised by three pneumatic' motors, revolved by drive (a) to the next
position and moved back by the pneumatic motors.

In this way, a net of measuring positions is distributed over a sectional

area in the seal. Figure 4 shows the position of the measuring points within the
seal. S everal eccentric positions of the rotor could be adjusted at the inves-

244
tigated stages. The resulting variable gap clearance at the circumference
influences the number of measuring posi
. tions. The whole net of measuring
positions can be divided in two areas as shown in picture 4. Those positions,
that are located within the cavity are used for all measurements. Previous
studies supposed a fundamental influence of flow in the clearance on the whole
flow patterns in the seal. Therefore, additional measuring positions were
provided in the surroundings of every throttle gap. The number of positions
depends on the adjusted eccentricity. The geometric clearance varies
between 0,1 and 0,7 mm (0,004 and 0,027 inch). To enable the measurement
of flow conditions as exact as possible, measuring positions were distributed
over the clearance with a distance of 0,1 mm (0,004 inch) from each other.

Cantilevered pitot Cylindre

As already mentioned, a cantilevered pitot cylindre was used for the

measurements of flow conditions. The cylindres diameter is 1 mm (0,039 inch),
that of the measuring hole is 0,1 rom (0,004 inch). Figure 5 shows the structure
of the measuring unit. It consists of the pitot cylindre, a miniature pressure
transducer and the cylindre carrier. Measurement is done by revolving the
cylindre and picking up pressure data simultaneously. The position and the
level of the total pressure in the stagnation point, as well as the static
pressure at the measuring position is obtained from the pressure distribution.
Thus, intensity and direction of flow velocity can be derived in an area
perpendicular to the cylindre.

Test results may be influenced by a wall close to the cylindre. Therefore,

tests were made in a special prepared wind tunnel. Deviations of the results due
to a nearby wall have been taken into account. The whole test apparatus was
controlled and watched by a desktop computer. In addition, this computer
gathered test and position data.

IMPULSE TYPE STAGE

The result of a test period can be seen in figure 6 with the rotor in
centric position. The figure shows the measured vectors of flow velocity in each
position, described by arrows. Measuring areas are illustrated by dashed linies
in the labyrinth scheme. The arrows length shows the level of velocities Its
direction is equivalent to the flow direction of the measured cylindre flow.
The entry flow into the first cavity follows the rules of potential swirl.

Spatial Distribution of Velocity

Figure 7 shows the spatial distribution of flow velocities in a

qualitative illustration. As shown in the figure, a dominating flow in
peripheral direction fills the labyrinth cavities. This core flow is overlapped
by a heavy, screw-like flow near the walls of the labyrinth. The three
dimensional distribution of the flow velocity is determined by measurements over
the height of the cavity, because only two dimensional velocities are measured

245
by the pitot cylindre, The directibns of flow velocity near the ceiling of the
cavity and the rotors surface, together with the rules of mass continuity point
out a "vortex-screw" rotating with high peripheral velocity within the cavity.
The rotating direction of the vortex is determined by the leakage flow and the
geometric boundary of the cavity. In the second cavity of a stepped seal, as
illustrated in the figure, the leakage flow dashes against the step of the
shruding. The flow is turned round at this point and raises, until it touches
the ceiling of the cavity. The "vortex-screw" revolves counter clockwise due to
the additional pressure drop to the next cavity. Caused by the set-back of the
shrouding in the third cavity, the expansion of the leakage flow is limited on
the nearest surroundings of the rotor surface. Above this f19w, a dominating
tangential flow nearly fills the space between the sealing strips and the cei
ling of the cavity. A "vortex-screw" revolving clockwise in this cavity, grows
by jet mixing and impulse exchange.

Axial Flow in the cavities

Figure 8 shows the distribution of axial velocities in the cavities of the

labyrinth. Measurements were made near the ceiling, in the middle of the cavity
and at the shrouding by changing the eccentricity of the rotor. Up stream
velocities are marked by negative axial flow components. They are caused by
"vortex-screws". The rotating direction of the yortex is determined by velocity
near the shrouding and the ceiling of the cavity. The influence of rotor
eccentricity on flow patterns in this region is unimportant.

Velocities at the Gap Clearance

Near the gap clearance, velocity distribution depends hard on rotor

eccentricity. Especially the distribution of axial velocities along the clearance
is modified, while tangential velocity is not influenced by rotor position.
Figure 9 shows the distribution of axial velocity and flow angle in the gap
clearance. The first measuring position is not situated within the clearance,
because a minimum distance to the rotating surface must be regarded to secure
the pitot cylindre. For the same reason, the velocity close to the shrouding
could not be measured. The influence of the seal clearance on the velocity can
be seen in the figure. Additional the security distance to the cylindre is shown,
with a rotor that is 0,3 mm (0,012 inch) out of centre. The velocity distribution
in the smallest clearance, according to e = - 0,2 mm (0,008 inch), changes
continuously with increasing clearance. This points out the influence of varying
flow contraction. The flow angles, an important factor of calculation, are shown
in the lower figure. As already mentioned, tangential flow does not change by
eccentricity. Therefore, flow angles are essential determined by axial velocity
and can be evaluated using leakage flow and flow contraction.

Exitation Coefficient of Impulse Type Stage

Exciting forces can be defined as dimensionless coefficient too.

Q lUis (7 )
ell"

246
 The exitation coefficient K2 is given by the quotient of related exciting
force Q/U. and relative eccentricity e/l", where 1" is the length of the rotor
s
blade.

In figure 10, this coefficient is plotted over the power of the stage,
represented by the load factor 'll = 2his/u2, with his as the isentropic
enthalpy difference and u the circumferential speed. The resulting force on
the rotor of impulse-type stage is illustrated in curve (a). Forces due to
pressure distributon and to leakage losses are both included. The exitation
coefficient induced by pressure distribution was measured and is shown in
curve (b). The part of leakage losses cannot be measured directly. Therefore,
curve (c) shows the calculated exitation coefficent of leakage losses. Near
load factor 3.5, a remarkable behaviour of excitation coefficient can be seen.
A close examination of this behaviour points out that the speed of rotation is
equal to the peripheral flow at labyrinth entry. Beyond this point, velocities
of labyrinth flow and the speed of rotation are different. Thus, the course of
exitation coefficient can be explained as influence of shrouding friction.
This effect has not yet been regarded as important, especially for seals with
short axial length.

Balance of Energy

Based on the results of measurements, a balance of energy within the seal

and the cavities is set up. Figure 11 illustrates this balance of energy for
different operating points of the impulse type stage. A clear supply of energy
by the shnouding can be seen at load factor 2 and 3, whereas elimination of
energy dominates at load factor 4. Calculating the losses as well as the
velocities and the pressure in the cavities, enables the evaluation of energy
flow. Friction losses can be determined from the mass flow of the "vortex-screw"
and the pressure drop caused by friction. The distribution of energy losses in
"the cavities and the shrouding, shown in the picture, base on this evaluation.

REACTION TYPE STAGE

The energy supply of the labyrinth is influenced by the construction of the

shrouding. The blades of reaction type stage are put together in buckets of three
or four blades by means of shrouding segments that are separated
. from each
other by a gap.

Figure 12 shows the flow pattern within the labyrinth seal of this stage.
Because of the similar geometric shape the flow in this seal can be compared
with stage II. A dominating tangential flow overlapped by a "vortex-screw"
results as well. But it is additionally influenced by a jet flow that injects
perpendicular to the shrouding. This jet flow reaches to the ceiling of the
cavity and causes pressure oscillations picked up during additional measure
ments. The oscillations at high frequency level do not influence the force
due to pressure distribution directly, but the jet flow, injected from the gaps
of the shrouding, works like a rotating blade. Thus the swirl of flow in the
cavities is influenced.

247
 Exitation Coefficient of the Reaction Type Stage

As the jet flow depends on the rotating speed of the shrouding, the
influence of shrouding velocity on exciting forces increases. In figure 13,
the exitation coefficient is plotted over the load factor of the reaction
type stage. Because of the slight entry swirl, exitation coefficients are
generally lower. In spite of this, a heavy influence of shrouding velocity
on measured resulting exitation is to be seen. Near load factor four, the
measured tangential flow is equal to the rotating speed of the shrouding, The
lowest exitation coefficient is measured at this load. Obviously. the injection
flow is able to accelerate the swirl of entry flow in a way, that raises the
exciting forces by lowering power. At high load factors, influence of injection
flow on entry swirl prevents an increase of exitation with the load number.

CONCLUDING REMARKS

As shown by the results of investigations, high tangential velocities,

varying flow contraction together with its influence on flow angles, and the
effect of friction on the shrouding are to be especially considered by the
calculation of exciting forces. A method that considers these influences is
given by ref (9). In this case, the labyrinth flow is described by dividing
the leakage flow in stream tubes that are joint against one another at the
circumference.

The shape of a stream tube is principally illustrated in figure 14. At

every cross section, the equations of impulse and energy are resolved
iteratively, using the continuity of mass flow. A part of the working fluid
is branched at the entry o f the seal. The shape of the stream tubes within
the seal depends on various conditions. The leakage flow is added to the
working fluid at outlet of the seal. This kind of view gives a description
of complete flow patterns, as usually present in labyrinth seals, by means of
a one-dimensional flow.

The test results enable an extension of this method. Two different flow
areas, relating to the flow within the cavity and the leakage flow are to
consider. New statements are needed do describe the local profile of the
stream tubes that.are not identical with the geometric boundary of the labyrinth.
Additionally, the influence of shrouding velocity must be considered by friction
and velocity at the rotor surface. Besides this, the chang e of flow angles due
to flow contraction is important. Because of a slight pressure drop within
every cavity, calculation in the cavity can be done using the rules of
incompressible flow. The pressure drop from one cavity to the n ext however
must be considered. Therefore, compressible flow should be supposed for
calculations in the gap. Using these additional considerations, the accuracy
of this method would raise essentially. Actual work is don e on this topic and
will be described in a later paper.

REFERENCES

1. Thomas, H.-J.: Instabile Eigenschwingungen von Turbinenlaufern,

angefacht durch die Spaltenstromung in Stopfbuchsen und Beschau
felungen. Bull. de L'AIM (1958) Nr. 11/12, S. 1039-1063

248
2. Alford, J. S.: Protecting Turbomachinery From Selfexited Whirl.
Jornal of Engineering for Power, Trans. ASME, Series A, 38 (1965)
pp333-344

3. Lomakin, A. A.: Die Berechnung der kritischen Drehzahl und der Bedin
gung fUr Sicherung der dynamischen Stabilitat des Laufers von hydrau
lischen Hochdruck-Maschinen unter BerUcksichtigurig der Krafte, die in
den Dichtungen entstehen. Energomasinostroenie' 4 (1958) H. 4, S. 1-5

4. Thomas, H.-J.: Urlichs, K.; Wohlrab, R.: Lauferinstabilitat bei ther

mischen Turbomaschinen infolge Spalterregung. VGB Kraftwerkstechnik
56 (1976) H. 6, S. 377-383

5. Hauck, L.: Vergleich gebrauchlicher Turbinenstufen hinsichtlich des

Auftretens spaltstromungsbedingter Krafte. Konstruktion 33, (1981)
H. 2, S. 59-64

6. Childs, D. W.: Dressman, J. B.: Testing of Turbulent Seals for Rotor

dynamic Coefficient. NASA Publication 2133, May 1980

7. Beneker4H.; Wachter, J.: Flow Induced Coefficients of Labyrinth

Seals for Application in Rotordynamics. NASA Conference Publication
2133, (1980)

8. Jenny, R.: Labyrinthe als Ursache selbsterregter Rotor-Schwingungen

bei Radialverdichtern. Technische Rundschau Sulzer (1980) H. 4,
S. 149-156

9. Urlichs, K.: Durch Spaltstromungen hervorgerufene Querkrafte an den

Laufern thermischer Turbomaschinen, Diss. TU MUnchen 1975; s.a.
lng.-Arch. 45 (1976) H.3, S. 193-208

249
 TABLE I. - DATA OF INVESTIGATED STAGES

stage I stage II
-I
Speed m1n 4000 5000

Diameter of stage mm 420 420

Number of blades
nozzle vanes 50 63
wheel blades 50 99

Length of blades
nozzle vanes mm 40.6 3 8.6
wheel blades mm 42.0 40.0

Flow angles
al degrees 17.3 12.9
degrees 17.3 20. I
{32

Clearance
nozzle vanes mm 0.48 0.40
wheel blades mm 0.43 0.40

l-lidth of shrouding
nozzle vanes mm 22.4 36.0
wheel blades mm 19.0 27.0

Degree of reaction 0.45 0.20

250
 Figure 1. - Invetigated stages.

Figure 2. - Labyrinth seal and measuring unit.

251
 rive
ol ving d
a rev m otor
umat ic
b pne d riv e
lving
c r evo n g dr iv
e
a i al movi
d
sing
e hou

ear bO'
of the g
ture
Struc
e 3.
FIgur
_

252
 gap1 gap2 gap3

I
a I I

cavity1 cavity2 cavity 3 cavity4

. . . ...
. . ....
.
.

. i i .
.

L! --:,! ....r--
..
i I
-.L-.L-
...
__

a smallest clearance
b widest clearance

Figure 4. - Distributio of measuring positions.

16 17 ..
-I-
A<' x )( 'X' ')( I

I
I

" "- ,, "-

-
VA
't

.
I iii
/f t
"-
i-- I
"- "-'" "- "I

I
!
b c a

a pitot cylindre
b cylindre carrier
c pressure transducer

Figure 5. - Structure of the measuring unit.

253
 [ca.
I u \\11\\ \ jjjjj 1CLt
1 "' - - ""1
llllil \ 1 \ \
;.tr=L!;
l\lIJ\\\\\h
Ir-I-'I

\ 1\ \\ h Jtcl,
I 1 \ \ l
Ir-J-I

cavityl
l \i R
cavity 2 cavity 3 cavity4
Jtcl,
Figure 6. - Flow velocity within the cavities (measured).

254
 Figure 7. - Spatial distribution of flow velocities (stage II).

t-:7....--
. /;:;;:::;;;;;Ceiling
of the cavity

middle of the cavity

75
plane of measurements
50

25
Vi'
"- shrouding
E 0
2
-25 throttling gap
u

-50

-75

Figure 8. - Distribution of axial velocity in the cavities (measured).

255
 70

t
60

_ 50

E
40
0
u shrouding at
e=0.3
30

20

10

0
-0,1 0 0,1 0.2 0.3 0,4 0.5 0,6 0.7
clearance [mml ----
1900

t
1800

1 70 0
(f-
4>
cox

<7 1600

1500

1400

13 0 0

1200
-0.1 0 0,1 0,2 0,3 O.L. 0,5 0.6 0.7
clearance [mm] ---

Figure 9. - Axial velocity and flow angle in the gap (measured).

256
 3,5

3,0
....
N
::s:.:: 2,5
CXa
L- b .... "
0 -a....-,
u
a 2,0
0+-

c
.2
..... 1.5
a
x
Q)
1,0

0,5 '17
......"'../ . ..
c . ...""........
0
2 3 4 5 6 7
load factor tV

a resulting exitation (measured)

b exitation by pressure distribution (measured)
c exitation by leakage losses (calculated)

Figure 10. - Exitation factor and load factor (stage II).

1\1=3 1\1=4
1\1=2

figure 11. - Balance of energy and load factor (stage II).

257
Figure 12. - Spatial distribution of flow velocities (stage 1).

2,0

1,6
\0

\
N

L-
0
-
u 1,2

c
-

c
0 0
-
c 0,8 .....
-
-0..7""' 1---'1--
'x b
(11

0,4
., ... .,...... J . .. '0.

c!.

o
2 3 4 5 6 7
load factor tV
a resulting exitation (measured)
b exitation by pressure distribution (measured)
c exitation by leakage losses (calculated)

Figure 13. - Exitation factor and load factor (stage I).

258
i-l,k i.k i -1.k

.... k
n-1.k i.1 i.2
n.k

..

Fi gure 14. - Shape of stream tube.

259
 EXPERIlViENT lVlFASURHV.lENT OF ALFUID '8 FacE

IN AXIAL-1"1...0\' TrnOOV.w.::HINEHY

John lYl. Vance

Uepartnt of lVchanical Engineering
Texas A&NI University
College Station, Texas 77843

Frank J. Laudadio
Nilltorola Corporation
Fort Lauderdale, Florida 33322

ABSTRACT

In 1965, J. S. Alford published a theory and mathematical model [lJ which pre
dicts that circumferential variation of blade-tip clearances in axial-flow turbo
machinery will produce cross-coupled (normal to the eccentricity) aerodynamic forces
on the rotor. Ever since then, the theory has been used (without experimental veri
fication) by rotor dynamicists as one of the few mathematical models available to
calculate the cross-coupled aerodynamic stiffness coefficients required for rotor
dynamic stability analysis.

This paper presents the results of experimental measurements made on a small,

high speed, axial flow test apparatus to verify the existence of Alford's force and
to investigate the validity of his mathematical prediction model.

INTRODUCTIO N

There are many mechanisms which cause rotordynamic instabilities in turbines

and compressors. An instapi1ity in turbomachinery occurs when the shaft precesses
about the bearing centerline (whirling) with unacceptable amplitudes, at frequencies
nonsynchronous with shaft speed. Instabilities can become a costly problem since
they limit the operation of the turbomachineryby reducing the allowable speed.
They can be destructive if the instability threshold speed is exceeded.

There are documented cases showing the severity of instabilities in turbo

machinery. Doyle [ 2J presents two cases where an insta b ility has required exten
sive rotor modification, resulting in lost production and heavy maintenance expen
ditures. One of the cases mentioned is the 1964 Ekofisk gas reinjection problem
which has become a classical example of how costly an instability can become. A
centrifugal compressor was designed to reinject the gas by-product from crude oil
production. Oil production was stopped for several months due to the instability
problem with this compressor.

260
 In mathematical terms an instability occurs when an eigenvalue of the equation
that describes the motion of the rotor has a positive real component. In engi
neering terms, the nonsynchronoQs amplitude of rotor whirl (any amplitude which has
a frequency not equal to running speed) grows with time until the system exceeds
its operating limits. In many instances the rotor orbits can be large enough to de
stroy bearings and seals.

Computer programs are now available [3J which can predict whirling frequencies
and threshold speeds, provided that accurate values for the destabilizing coeffi
cients are used as input. The destabilizing coefficients usually take the form of
a linearized cross-coupled stiffness term [4J but sometimes are expressed as neg
ative damping [5J, depending on the mechanism producing the instability.

One mechanism, which is a type of self-excited instability, is the aerodynamic

excitation due to circumferential variation of blade clearance in axial-flow com
pressors and axial-flow turbines. An objective of the work reported here was to
quantify a term in the mathematical equation due to Alford which predicts the mag
nitude of the aerodynamic cross-coupled stiffness, along with verifying the form of
the equation.

Alford actually presented two mechanisms which can cause severe rotor whirl in
axial flow compressors and turbines. One is due to the circumferential variation
of static pressure action on the cylindrical surface of a rotor, particularly with
in labyrinth seals. The other is due to circumferential variation of blade-tip
clearance (the topic of this paper). Alford presented both seal deflection criteria
and bladed-disk torque deflection criteria for use as design guides for a stable
rotor system.

Investigations of rotordynamic problems in steam turbines have shown that un

balanced moments on wheels and shafts, caused by steam forces, are a source for non
synchronous rotor whirl. Steam whirl was first described by Thomas [6J in 1958 and
is similar to the phenomenon hypothesized by Alford. Winter [7J performed model
tests in 1968. He used ball bearings in his test rig to eliminate the shaft whip
ping due to hydrodynamic bearings, and applied external damping elements to the
bearings since ball bearings have little inherent damping. Winter proved the pre
sence of a load-dependent instability but was not able to quantify his results.

THE TEST APPARATUS

The test apparatus used in the experiment consists of a flexibly mounted bladed
rotor driven by a variable speed electric motor, and a movable shroud (see Figure
1). The blade O. D. is 6". The motor is rated 1/7 HP at 10,000 rpm.

An offset of the bladed rotor within the shroud clearance circle is produced by
translating the shroud laterally. The shroud is mounted on slides, and is trans
lated laterally by turning a threaded rod.

The lateral stiffness of the rotor bearing support is made low enough so that
very low levels of Alford's force (3-10 grams) produce a measurable rotor deflec
tion. The rotor deflection is measured with an eddy-current proximity probe and

261
calibrated to lateral force in grams/mil. One of the most challenging and time
consuming aspects of the project was the measurement of rotor deflection (DC signal)
in the presence of a relative large AC signal produced by rotor runout, synchronous
vibration, and ball bearing roughness. This was finally accomplished by passing the
probe signal through a very effective low-pass filter.

The speed of the rotor can be varied independently from the torque, over a range
of 0 to 7000 rpm, by changing the voltage supplied to the electric motor. The torque
is varied by adjusting the air velocity at the rotor inlet with an auxiliary blower.
This method allows the rotor speed to be changed while holding the torque constant,
or vice versa. Many combinations of seed, torque, and shroud eccentricity can thus
be obtained.

Rotor speed is measured by feeding a once-per-revolution keyphaser signal into

an electronic counter.

The aerodynamic torque is measured with a strain gage dynamometer. The driving
torque of the electric motor is entirely reacted by a steel rod in torsion, with
torsional strain gages calibrated to torque.

EXPERIMENTAL PROCEDURE

For each set of data taken, a base measurement was made with the shroud in the
center position (i.e. with the lateral eccentricity nominally zero). With the shroud
in this position, the rotor was run up to the selected measurement speed and measure
ments were made of rotor deflection, speed, and torque.

Immediately after completing the base measurements, an eccentricity was imparted

to the shroud (adjusting screws translate the shroud laterally) and new values of
rotor deflection and torque were measured at the same speed as for the base measure
ment. Since the direction, as well as the magnitude, of shroud offset was found to
have an effect on the measurements, the procedure was repeated for an equal shroud
eccentricity in the negative (opposite) direction.

Large quantities of data were taken and averaged, since the repeatability was
found to be relatively poor (typically 20% for two successive measurements of the
same variable under the same conditions). The poor repeatability was partially due
to apparent aerodynamic instabilities (surging) under some conditions of speed and
torque .

RESULTS

The results of the experiments are shown in Figures 2 through 13. The lines
connecting the points in the figures were derived using a least squares fit.

Figures 2 and 3 show how the measured Alford force varied with rotor eccentric
ity. The only difference between-the two graphs is in the direction that the
eccentricity was imparted. Figures 4 and 5 show the magnitude of the efficiency
facor, S, and how it varies with rotor eccentricity. Again, the differnce between
the two figures is the direction that the eccentricity was imparted. The efficiency

262
factor 8 was calculated using Alford's equation which gives th aerodynamic foce as
proportional to a radial deflection which is normal to the force. The constant of
proportionality is called cross-coupled stiffness. Thus we have

(1)

1"8
The cross-coupled stiffness is seen to be x n-H'
_

Equation (1) implies that 8 remains constant for varying eccentricity. Figure
5 confirms this. Therefore it was felt that the point on Figure 4 corresponding to
20 mils eccentricity was not representative, and should not be included in the least
squares fit.

Figures 6 and 7 show how Alford's force and 8 vary with torque. Again 8 was
calculated using equation (1) . Figure 8 shows how the measured Alford force varied
with speed holding the rotor torque constant. Figure 9 was derived from figures 6
and 8. The curve corresponding to 5000 rpm was derived using a least squares fit to
the data from all 3 tests in Figure 6. The four remaining curves on Figure 9 were
derived from this one curve and Figure 8, using extrapolation techniques. Figures
10, 11, 12 and 13 ,show how the measured Alford force 'varied with rotor speed and
torque. These figures help support the empirical equation of Alford's force to be
presented below. They show how, the developed rotor torque varies with the speed.

AN IMPROVED PREDICTION MODEL

One of the remarkable features of Alford's theory is its simplicity. Figures

2, 3, and 6 verify the predicted linearity of Alford's force with rotor eccentricity
and stage torque. However, Figure 6 shows that Alford's force does not appear at all
until a certain level of stage torque is reached: then it suddenly appears and in
creases steeply with torque. Alford mentions, in reference [1], that "large power
inputs to the compressor rotor appear to increase the hazard of whirl. The vibration
problem was encountered only at the full 100 percent power rating of the engine."
Figure 6 is compatible with this observation.

Figures 7 and 8 also reveal certain limitations of equation (1) , principally

that the efficiency factor 8 actually varies with the aerodynamic load torque and
that the developed Alford's force is speed-dependent as well as torque-dependent.
Also, extrapolation of the curves for Alford's force versus torque (e.g. Figure 6)
shows that the y-intercept is not zero as ,equation (1) predicts.

To obtain an improved prediction model, the speed'must be included as a perti

nent variable. Dimensional analysis shows that if the speed is pertinent, then the
inlet velocity is also pertinent. In dimensionless groups, the prediction equation
takes the form

(2)

263
 where N =
speed, rpm
V =
inlet velocity, fps
and the other variables are as previously defined.

In the experiments reported here, measurements of the inlet velocity V were not
made. Therefore the improved prediction model at this stage must be purely empiri
cal, to fit the available test data.

Equation (1) is modified to include a y-intercept, which gives

(3)

where the constant C is evaluated as C =

-17.37 from Figure 9.

In addition, Figure 9 shows that the slope of the curve is affected by the rotor
speed. A relationship can be calculated between the rotor speed and the slope
x/(DpH), as follows:

Sx -
- -10.54 x 10 -3 (N) + 89.33 (4)
D H
p

Substituting equation (4) into equation (3) gives

F =
(-10.54 x 10-3 (N) + 89.33) T - 17.37 (5)
y

In a more general form; factoring out x/(DpH) where x 40 mils, Dp 4 875 in.,
and
H =
0.875 in. gives

x
F
Y
=
{(-1.l2(N) + 9526.42) T - l852.35 } D H
(6)
p

where
Fy =
Alford force, grams
T -
developed rotor torque, in-lb
N =
rotor speed, rpm
x =
eccentricity, in
Dp =
blade diameter, in
H =
bucket height, in

If the units of Alford force are taken to be pounds (lb), the equation becomes

3 1 X
t (-2.47 x lO- (rpm- ) x (N) + 21.0) T - 4.08(in-lb) l D H
(7)
P

Equation (6) reasonably approximates all of the curves presented in this paper
that show Alford's force as a function of developed rotor torque, rotor speed, and
eccentricity. It is a purely empirical equation which fits the measured data for
the described test rig.

264
 CONCLUSIONS AND SUGGESTIONS FOR FUTURE RESEARCH

Due to the size of the test rig and the lack of measurement capability for aero
dynamic velocities and pressures, the conclusions which can be made from this re
search are quite limited in scope. They are as follows:

1. The aerodynamic force postulated by Alford for axial flow turbomachinery

does exist, and can be measured experimentally.

2. Alford's force is speed-dependent as well as torque-dependent, and is

probably also a function of the gas velocity entering the stage.

3. Although the relationship of Alford's force to the stage torque is fairly

linear at a given speed, the y-intercept is not zero. That is, no aerodynamic force
is generated until a certain level of stage torque is reached.

The authors believe that these conclusions are significant, since rotordynamics
engineers have used Alford's theory for years with no experimental verification.

However, it is easy to see why rotordynamic stability thresholds cannot be

reliably predicted at present by computer analyses which use Alford's equation as a
model for the destabilizing coefficients. In order to do this, it will be necessary
to develop more accurate prediction models for the aerodynamic force, which will re
quire experimental measurements on full scale turbomachinery components, such as an
axial flow stage from the compressor in a turbojet aircraft engine, or an industrial
process compressor.

REFERENCES

1. Alford, J. S., "Protecting Turbomachinery from Self-Excited Rotor Whirl,"

Journal of Engineering for Power, Vol. 87, Series A, No. 4, October 1965,
pp. 333-334.

2. Doyle, H. E., "Field Experiences with Rotordynamic Instability in High-Perfor

mance Turbomachinery," Proceedings of a Workshop held at Texas A&M University,
May 1980, NASA Conference Publication 2133.

3. Murphy, B. T., "Computer Programs for Cr itical Speeds and Stability Analysis,"
Rotordynamics of Turbomachinery Short Course, Texas A&M University, May 18-20,
1981.

4. Vance, J. M. and Laudadio, F. J., "Rotordynamic Instabilities in Centrifugal

Compressors - Are all the Excitations Understood?", ASME Journal of Engineer
ing for Power, April 1981, pp. 288-293.

5. Vance, J. M., "Instabilities in TurbomachinerY", Proceedings of the 5th Annual

Vibration Institute Seminar on Machinery Vibration Analysis, April 9, 1981,
New Orleans, La.

265
6. Thomas, H. J., "Unstable Oscillations of Turbine Rotors Due to Steam Leakage in
the Clearances of the Sealing Glands and the Buckets," Bulletin Scientifique,
A. J. M. 71, 1958, pp. 1039-1063.

7. Winter, C. J., "Instabile Bewegungen von Turbinenlaeufern infolge lastabhaengiger

Erregerkraefte," Energie and Technik, January 1969, pp. 7-13.

266
i---
I
! I .-- "hat"

/
-@
/

'
Fi gure 1. - Dr a w l'ng of top 0 f t e s t r1'9 showlng inlet configura t'on
1 and
coordinate system.

267
 e
Cf"I"s:::T: +x
ROTCR S"E:ED seea RA'I

. TCRCl.E B.6S3 II'H..B

A
L. 6
F"
o
R
D

F"
o
R 4
C
e:

(
s
,.

2
a
)

9
B

Eo:::Dil'FlICIT'r' (m11al

Figure 2. - Alford force as function of rotor eccentricity. showing 2 standard

deviations (shroud offset: +X)

..
19
Cf"I"s:::T: -x
ROTCR S"E:ED sa;,a RA'I

TCRCl.E B.6S3 II'i-UI

A
L. e
F"
o
.R
D

F" 6
o
R
C
E

(
S 4
,.


a
)
2

e
e

Eo:::Di1'RICIT'r' (m1Is)

Figure 3. - Alford force as fnctton of rotor eccentricity, shwing 2 standard

deviations (shroud offset: -X).

268
 3.e
SH<CUD Cf'i'"'SE:T: +x
ROTCR S02e Rf'I1

TCRGLE e.6S:I Il'H..Jl

2.5

E
r
r
I
c
I
E 2. III
N
C
Y

r
F\
c
T 1.5
o
R

i.e

Figure 4. - Efficiency factor, S. as function of rotor eccentricity, showing

2 standard deviations (shroud offset: +X).

3.5
Cf'l"'SC:T: -x
. RaTCR S"ED) S02e Rf'I1

Ta<Q. e.6S:I. I/'H..B

.
J3 3.B

&: "r-
r
r -
I
C 2.5
I (.
&:
.
N (.
C
Y

r z.e
F\
c --
T
0
R -:..
1.5

1.9

EOCENTRICITY (mils)

Figure 5. - Efficiency factor, S. as function of rotor eccentricity. showing

2 standard deviations (shroud offset: -X).

269
 S
5I-R1II) Cf"'F'5<:T: +x
Eo::::ENT"RICIlY 40 MILS

ROTCA s=t::D saaa RF'I'\

A -TE:ST I'(). 23
S -TE:ST I'(). 24
A If> -TE:ST I'(). 25
L 6
F'
.0
R
))

F'
0
R 4
C
E

(
9
,.

2

)

a
a.4 B.S 13.6 a.7 a.s

Figure 6. - Al ford force as function of developed rotor torque (for three

dlfferent tests).

2.5
s-R:lJD Cf"'F'5<:T I +X
Ec:a:liTl'!IClTY 40 MIl.S

ROTCA 5F-D) saaa RF'I'\

" -TE:ST I'(). 23

z.a
m n:sT I'(). 24
-

-TE:ST I'(). 25
I:
F'
F'
I
C 1.S
I
I:
H
C
Y

F' 1.15
A
C
T
0
R
a.

a.B
a.4 a.s a.6 a.7 a.s.

TCf<Q.E (In-lb)

Figure 7. - Efficiency factor, e. as function of developed rotor torque (for

three different tests).

270
 15
9-R:UI) CF'F'5il:T: +x
Ec:a:HT'RICIiY 41
I1I1.S

18
A
L
F'
0
R
D S

F'
0
R
C a
E

(
S
,.
-5

(i) -TCAQ.Jt a.691 IH-UI
) El -TCRCLE a.szs IH-UI

-18

Figure 8. - Alford force as functton of rotor speed ( for two different values
of torque).

A
L 18
F'
o
R -
D

F'
o
R a
c
E

(
!
,.

-18

)
D=I5HJ) C1.R\I F'ROrI
j:l"f"ORI)'S E:cu:lTICti
WITHS2

-29
a.a a.2 8.4 9.6 a.a

TCRCLE Ctn-Ib)

Figure 9. - Alford force as function of developed rotor torque ( for different

values of speed ) .

271
 CJ"'F'SET: +X
ECCOmUCIlY 4EI MILS

A
L e
r
o
R
II

r ,
o
R
C
E

(
S 4
,.



)
2

5F'EEJ) (!"JIm)

9.474 9.ES:3 9.E1B4

TCA;LE (In-lb)

Figure 10. - Alford force as function of developed rotor torque and rotor
speed, showin 2 standard deviations (shroud offset: +X).
1Z
SH<O.JI) a=i"5tT: .;x
EcaJiTRICITY 4EI MILS

1B
A
L
r ,
,-
o ."
R ,
/
II a

r
o
R
C ,
E

(
s.
,.
4


)

9.474 9.E1B4

Ta<Q.E On-lb)
Figure 11. - Alford force as function of developed rotor torque and rotor
speed, showing 2 standard deviations (shroud offet: -X).

272
 4
9-RCllD t1JTIO'i: +X
E:o:::E:N'TRICITY 49 MIl.S

J)
, .:3

E:
F'
F'
I
C
I
E: Z
M
C
Y

F'
A
C
T 1
o
R

S?E:ED (rpm)

0.474 e.6S'3 S.9S4

TCR:LE (jn-lb 1
Fi gure 12. - Efficiency factor, S. as function of developed rotor torque
and rotor speed. showing 2 standard deviations (shroud offset: +X).

5H'<O.!D Cf'F"StT: -x
E:o:::ufTRICITY 49 MIl.S

J)
.

E:
F'
F'
I
C
I
E:
Ii
C
Y

F'
A
C
T 1
o
R

S?E:ED (rpm)

0.474 0.653 e.9S4

TCRCl Cin-lb)
Figure 13. - Efficiency factor, S. as function of developed rotor torque
and rotor speed. showing 2 standard deviations (shroud offset: -X).

273
 HEURISTIC EXPlANATION OF JOURNAL BEARINJ INSTABILI1Y

Stephen H. Crandall
Massachusetts Institute of Technology
cambridge, Massachusetts 02139

SUMMARY

A fluid-filled journal bearing is viewed as a powerful pump circulating fluid

around the annular space between the journal and the bearing. A small whirling
motion of the journal generates a wave of thickness variation progressing around the
channel. The hypothesis that the fluid flow drives the whirl whenever the mean of
the pumped fluid velocity is greater than the peripheral speed of the thickness
variation wave is discussed and compared with other simple explanations of journal
bearing instability. It is shown that for non-cavitating long bearings the
hypothesis predicts instability onset correctly for unloaded bearings but gradually
overpredicts the onset speed as the load is increased.

INTRODUCTION

One of the important causes for high speed rotor instability is the oil whip
phenomenon in hydrodynamically lubricated journal bearings. It was first reported
in 1925 by Newkirk and Taylor (ref. 1) who described several experiments and gave
a simple explanation of why a lightly loaded journal whirls at half the frequency of
rotation. In the present paper Newkirk and Taylor's argument is reexamined and
compare d with a heuristic hypothesis briefly suggested (ref. 2) in the first
Workshop on Rotordynamic Instability in High Performance Turbomachinery in 1980.
When compared with conventional dynamic stability analysis both arguments are
incomplete. Nevertheless, both arguments predict the instability onset speed
correctly for unloaded full circular bearings. The heuristic hypothesis can
additionally be applied to loaded bearings for which it makes useful approximate
predictions of instability onset speeds for moderate loads. For simplicity the
discussion is centered on the case of a full circular bearing with two-dimensional
non-cavitating flow. To set the stage, the classical Sommerfeld analysis (ref. 3)
for equilibrium under a constant load is reviewed. After this the problem of
whirling stability is discussed, first for an unloaded bearing, and then for a loaded
bearing.

SOMMERFELD'S ANALYSIS

The idealized case treated by Sommerfeld (ref. 3) in 1904 is sketched in Fig. 1.

A journal of radius R rotates at a fixed angular rate within a full circular
bearing of radius R + ho where the radial clearance ho is very small in comparison
to R. The annular space between journal and bearing is filled with an incompressible
fluid lubricant with uniform viscosity. The fluid flow is taken to be two
dimensional; i.e., axial flow is assumed to be negligible. In addition it is assumed

274
 BEARI NG

Figure 1. - Equilibrium configuration of journal rotating at rate n under

load w. Journal center J displaced distance e with respect
to bearing center B.

that cavitation does not occur. The width of the bearing normal to the plane of
the figure is b.

Sommerfeld's principal result is that in the equilibrium position under a

vertical load W the journal is eccentrically displaced by a distance e in the
horizontal direction. Because of the eccentricity e the film thickness h varies
with position e around the annular space. For ho R the approximate relation is

h = h + e cos e = h (1 + a cos e) (1)

o 0

where a = elho is the eccentricity ratio. Under the assumptions of laminar viscous
flow with no pressure variation across the thickness of the film, the only possible
flow patterns that satisfy the requirements of fluid mechanics in a uniform channel
of thickness h are combinations of the two basic patterns shown in Fig. 2. Here x
is distance along the channel and y is distance across the channel with 0 < y < h.
The fluid velocity (in the x-direction at position y) is denoted by u. The volume
flow rate across a section of the channel of width b normal to the plane of the
figure is denoted by Q and dp/dx is the pressure gradient along the channel. For
the linear profile at the left of Fig. 2 the pattern depends on the parameter U
which is the velocity of the upper channel wall (the velocity of the lower channel
wall is taken to be zero). For the parabolic profile at the right of Fig. 2 the
pattern depends on the parameter A which is a velocity whose magnitude is four
times the peak velocity in the profile or six times the average velocity. While the

275
 -u --

T h

I
1
2 2
u D y/h u A(y/h _ y /h )

Q D bh/2 Q A bh/6

le. Op 2
0 -2A]J/h
OX
=

dX
=

Figure 2. - Component laminar flows in narrow channel with uniform pressure

across channel. Either linear or parabolJc velocity profiles.

relations in Fig . 2 are strictly correct only for steady state flows with constant
values of the parameters h, D, and A, the Reynold's theory of lubrication extends
them to apply to slow variations, both in time t and space x, of these parameters.

In application to the bearing of Fig. 1, the component flows of Fig. 2 are

superposed, with x = R8 and D = RD, with h given by (1), and the undetermined
parameter A to be fixed by the requirements of continuity [oQ/o8 =
0] and uniqueness of
pressure [p(8) = p(8 + 2n)]. The value of A so determined is

Jl
1 3
cos 8
(2 )
1 + 2 -

and the total volume flow rate is

2
1 - a
RQbh (3)
o 2
2 + a

The pressure gradients of the component flows in Fig. 2 are superposed and integrated
to obtain the pressure distribution p(8) acting on the journal. The resultant of the
pressures acting on width b of the journal is a force, acting vertically upwards
through the journal center J, of magnitude

(4 )

276
The viscous shearing stresses acting on the boundary of the journal also produce a
resultant force on the journal but its magnitude is smaller than the pressure
resultant (4) by a factor of order ho l R, and thus may be neglected. The force (4)
must then be equal and opposite to the applied load W for the journal to be in
equilibrium.

The remarkable property of the Sommerfeld bearing model is that the equilibrium
displacement is at right angles to the applied load. This characteristic implies
that an unloaded bearing is unstable with respect to slow forward whirling motions
of the journal. To see this, imagine that an external agent moves the center of
the journal J of Fig. 1 in a circular path of radius e in the counterclockwise sense
about the bearing center B. When the journal passes through the position shown in
Fig. I if the motion is slow enough the lubricant flow and pressure distribution will
be very nearly the same as for the equilibrium configuration shown there, which means
that the fluid will be exerting a force very nearly equal and opposite to W on the
journal. This force, in the same direction as the velocity of the journal center J
in its circular path does positive work on the whirling motion. The Sommerfeld
bearing thus promotes whirling instability for very slow forward whirling rates. To
consider more rapid whirling rates it is necessary to extend the Sommerfeld analysis
to include motion of the journal center J.

WHIRLING STABILITY OF UNLOADED BEARING

According to (4) the equilibrium position for an unloaded bearing (W = 0) has

zero eccentricity (a = 0, e = 0). In this configuration the parameter Al of
Eq. (2) vanishes and the lubricant flow pattern is simply the linear profile shown
on the left of Fig. 2. The stability of this configuration is discussed, first by
describing Newkirk and Taylor's explanation of half-speed whirl (ref. 1) and the

Figure 3. - Traveling wave of lubricant film thickness variation circling

channel when center of rotating journal J whirls at frequency w around
circle of radius E: .

277
new heuristic hypothesis (ref. 2), and then by performing a conventional whirling
stability analysis which sheds light on the preceding arguments. The bearing
kinematics are displayed in Fig. 3 for the case where the journal center J whirls
at the steady angular rate w in a circle of radius about the equilibrium position
in which J and B coincide. Note that the diametrally opposite sections of maximum
and minimum film thickness advance around the bearing at the rate w. At the location
8 the film thickness is

h(8,t) = h + cos (8 - wt) (5)

o

Note that the dependence of h on space e and time t is that of a progressive

wave circling the annular channel with an angular phase velocity of w or a linear
phase velocity of Rw.

The Newkirk and Taylor explanation (ref. 1) is based on an application of the

continuity requirement to the flow in a channel whose thickness varies according to
(5) . The continuity requirement applied to a differential arc of length Rd8 is

h
+ b a
0
dt
.lSL
Rae
= (6)

Newkirk and Taylor assumed that the lubricant flow retains the linear profile of
Fig. 2 so that Q = RQ b h/2 which reduces (6) to

g dh dh == 0
at
(7)
2 d8
+

When h(6,t) from (5) is substituted in (7) the result is that continuity cannot be
satisfied unless

Q
w = (8)
2

Newkirk and Taylor took this to provide analytical verification of the half
frequency whirl phenomenon which they observed in a vertical shaft running in a
bearing with plentiful oil supply. It also appeared to explain the If Oil Resonance"
peak in response when the rotation rate was twice the natural frequency of the system.
The simple result (8) was less satisfactory in explaining the oil-whip phenonenon
where the whirling frequency remains at the natural frequency as the rotation rate
Q is increased, although it was noted that the onset of whipping always occured at
speeds equal to or greater than twice the natural frequency.

In the heuristic hypothesis (ref. 2) the rota.ting journal is considered to be

a pump impeller which maintains the lubricant flow pattern with linear profile when
the journal is. unloaded and centered. The fluid velocity varies linearly from u 0
at the bearing to u RQ at the journal. This flow can be decomposed into a mean
=

flow with uniform velocity u RQ/2 and no vorticity plus a residual flow with zero
mean velocity and large vorticity. It is assumed that the mean flow is available to
=

encourage (or discourage) small whirling perturbations of the centered configuration.

If a whirl involving a thickness variation wave of the form (5) is imposed the
heuristic hypothesis (ref. 2) postulates that energy will be pumped into the whirl
if the mean flow velocity RQ/2 is greater than the phase velocity Rw of the whirl
around the periphery of oil film. Conversely energy will be removed from the whirl
if the phase velocity of the whirl is greater than the mean velocity of the lubricant.
Neutral stability occurs when these velocities are equal; i.e., when w = Q/2. This

278
hypothesis thus "explains'! half .... frequency whirls and oil whip for a system which has
an unloaded bearing with plentiful oil supply. If the system provides little
restraint on the journal its motion will be primarily d etermined by the fluid film
forces acting on it. When a low frequency whirl (w < 0,/2) is accidently started
energy will be pumped into the whirl, accelerating the whirl until w gets
sufficiently close to 0,/2 th a t the energy pumped into the whirl just balances the
system energy losses in a steady "half.,..frequency" whirl. If the system provides
considerable restraint on the journal and only permits appreciable whirling motion
at a nat ural frequency the fluid film in the bearing will remove energy from
accidental whirls at a natural frequency w whenever 0, < 2w. However, when 0, > 2w
the fluid film will pump energy into any such whirl. Whether or not oil whip
occurs depends on whether the energy supplied by the fluid film is sufficient to
overcome the system losses.

Both of the previous explanations are incomplete in the sense that they do not
mak e use of all the physical requirements involved. Both arguments use the flow
pattern of the underlying stead y centered flow and both use the k inematics of a
small whirling perturbation. The Newkirk and Taylor argument also makes explicit
use of the continuity requirement for the perturbation but neither explanation
explicitly invokes any consideration of the pressures developed in the oil film.

Historically, the first complete dynamic analysis of the whirling stability of

an unloaded bearing was given by Robertson in 1933 (ref. 4). The development which
follows is essentially just a linearized version of Robertson's analysis. We
consider the whirl defined by the eccentricity ( ho) and the frequency w to be
a small perturbation on the underlying centered rotation in which the flow
distribution is simply

u = Rrt y/h , 0 < Y < h (9)

o 0

When the journal is whirling the flow pattern will be a superposition of the two
profiles of Fig. 2. To fit the conditions of Fig. 3 we take

2 2
u = Rrt y/h + A(y/ h .,.. y /h ) (10)

where h(8, t) is given by (5) and A is to be determined from the continuity require""
ment (6) and the requirement of single.,..valued pressure [pee)
pee + 21f). Using =

a linear perturbation analysis we neglect terms of order (/ho)2 in comparison

with terms of o rder /ho and find

0,
A -6R ( - w) cos(8 - wt) (11)
h 2
o

The corresponding total volume flow rate for the lubricant film is

Q "" Rrtbh/2 - R( - w)b cos(8 .,.. wt)

(12)
R0,bh /2 + Rbw cos(8 wt)
o

The pressure distribution is obtained by integrating the pressure gradient given in

Fig. 2. To first ord er in E/ho

279
 2
p (8, t) p l2R (g - w)sin(8 - wt) (13)
2
h3
.
o

The resultant force acting on the journal due to these pressures is directed at
right angles to the journal displacement and has the magnit ude

F 12TIEb R3 (14)
h
o

The sense is such that when 0,/2 > w, F has the same direction as the instantaneous
velocity VJ of the journal center. The rate at which the fluid film forces do work
on the journal (i.e., the power flow into the whirl) is

(15)

This analysis shows that the amplitude A of the component flow with parabolic profile.
the pressure in the fluid film, the resultant force on the journal, and the power
flow into the whirl all are proportional to the factor (0,/2 - w). Low speed whirls
are encouraged and high speed whirls are discouraged. The whirling frequency of
neutral stability is w = 0,/2.

These results can be compared with the two simplified arguments considered
previously. The assumption in the Newkirk and Taylor argument that the velocity
profile remains linear is equivalent to assuming that the parameter A in Fig. 2
vanishes, which according to (11) implies that w must equal 0,/2. Furthermore
the vanishing of A implies an absence of pressure gradient and consequently an
absence of resulting force so that at the particular whirl frequency w = 0,/2 the
fluid film neither retards or advances the whirl. This is the neutral stability
condition. The heuristic hypothesis (based simply on the flow pattern of the
underlying centered configuration and the kinematics of the whirl perturbation)
that the mean flow drives the whirl whenever the mean fluid velocity is greater
than the phase velocity of the whirl is essentially a qualitative statement
equivalent to the quantitative statement represented by eqn. (15). It happens
that for an unloaded bearing the frequency of neutral whirl is given correctly by
the heuristic hypothesis. For loaded bearings it is difficult to see how the
Newkirk and Taylor argument can be extended to predict any other frequency of
neutral whirl than W = 0,/2. The heuristic hypothesis is however easily extended.
It no longer predicts the exact frequency of neutral whirl but it provides useful
approximations for moderate loads.

WHIRLING STABILITY OF LOADED BEARINGS

In this section a linear perturbation analysis is made to determine the

whirling stability of the equilibrium configuration of Fig. 1. The heuristic
hypothesis is then applied to predict an approximate value of the whirling
frequency for neutral stability. We consider a small whirling motion of the
journal centered about the equilibrium position of Fig. 1 where the journal center
whirls with angular velocity w in a counter-clockwise sense about a circle of radius

280
E centered on the equilibrium position which lies a distance e to the right of the
bearing c.enter. In the equilibrium configuration the film thickness h, the parameter
AI, the volume flow rate Ql, and the load WI are given by eqns. (1), (2), ( ), and
3
(4) respectively. When the whirling motion is added the film thickness becomes

h{8,t) = h + e cos 8 + Ecos (8 - wt) (16)

o

With this value of h the velocity profile is taken to have the form of eqn. (10) with
the parameter A to be determined anew from the continuity requirements (6) and the
requirements of single-valued pressure [p(8) = pee + 27T)]. Using a linear
perturbation analysis with respect to E/ho (but including terms of all order with
respect to a'. = e/ho) we find

6R
[Q f (8, t) + w g (8,t)]
E

Al + h 1 + cos e
(17)
A =
o
a

where

2
1 cos (8 - wt)
f (8, t) = -
a

2 1 + cos 8
-

2 + a:
a

(18)

3a:
g(8,t) cos(8 - wt) + -"::";';"- cos (,;!t
2
2 +
=

Note that (17) reduces to (2) when E = 0 and to (11) when a = elho O. The
corresponding total volume flow rate is
=

Q Q - ERb r6Q ---=-- cos wt - wg(8,t)

2
l (19)
l J
a'._'

(2 + a, )
__

1
=

which reduces to ( ) when E = o and to (12) when a = O. The corresponding pressure

3
gradient from Fig. 2 when integrated gives the pressure distribution in the
lubricant film. The resultant force acting on the journal due to these pressures
has horizontal and vertical components given by

2
1 27T}.lbR3 a
Q
2 2
[1

2
W] E sinwt
h\1_a )3/
H -

2 +
-

(20)
0

2 2 2
1 27T}.lbR3 - a 2 a Q + (2+a )w
= + Q - w + a
] E cos wt
WI 2 2 2 3 2 2
V
h3(1_a )3/ [ + a (2 + a )
0

which reduce to (4) and (14) respectively when E = 0 and when a = O. The forces (20)
may be decomposed into three forces: the steady-state load WI W, a force
proportional to [(1 - a'.2)Q/(2 + a2) - w] which whirls in phase with the velocity of
=

the journal center, and an oscillating vertical force whose magnitude is small when

281
a is small. The total work done on the journal by these forces during one whirling
cycle; i.e., the energy per cycle E imparted to the whirl is

(21)

The lubricant film forces are once more destabilizing for slow whirls and stabilizing
for rapid whirls. The neutral stability whirl frequency

2 4 2 2
4 - 2a - a S6 3 (1_a ) S6
+
(22)
=

w
2 4 2 2
8 + 2a _a 9 _ (1_a )

varies from w = S6/2 at a = e/ho = 0 for an unloaded bearing to w -+ S6/3 when the load
approaches infinity and the eccentricity ratio a approaches unity. The variation of
w/S6 according to (22) is represented by the curve labeled A in Fig. 4.

0.6

0.4

.0.

0.2

1.0

a = e
ho

Figure 4. - Frequency w of neutral stability whirl as function of steady state

eccentricity ratio: A, according to complete stability analysis (22); B,
according to heuristic proposal (25).

Turning now to the heuristic hypothesis that the lubricant flow pumps energy
into the whirl whenever the mean flow velocity is greater than the phase velocity Rw
of the thickness variation wave, we recall that in the equilibrium configuration

282
of Fig. 1 the film thickness h is given by (1) and the volume flow rate Q is given
l
by (3). The mean velocity

Q
- 1 RQ 1 - a2
U + a cos e
(23)
bh 1 2
2 + a
= =

varies with location around the annulus. To obtain a single velocity to compare
with the phase velocity of the thickness variation wave, we propose to take the
root-mean-square average of U. The heuristic criterion for neutral stability then
becomes

Rw U
[ 1
J
r21T_
U
2
]
de
1/2
(24)
rms
= = -
-

21T
o

Evaluating the integral, we obtain

W (25)
=

as the neutral stability whirl frequency according to the proposed heuristic

hypothesis. The variation of this frequency with eccentricity ratio a is
represented by the curve B in Fig. 4. The heuristic prediction (25) is not
identical to (22) obtained from the dynamic stability analysis, but it provides a
useful approximation for values of the steady state eccentricity ratio that are less
than a = 0.5.

REFERENCES

1. B.L. Newkirk and H.D. Taylor, Shaft Whipping Due to Oil Action in Journal
Bearings, General Electric Review 28, 559-568 (1925).

2. S.H. Crandall, Physical Explanations of the Destabilizing Effect of Damping in

Rotating Parts, pp. 369-382 in Proc. Workshop on Rotordynamic Problems in
High Performance Turbomachinery, NASA CP 2133, 1980. See Appendix.

3. A. Sommerfeld, Zur hydrodynamischen Theorie der Schmiermittelreibung, Zeitschr.

fur Math. und Physik 50, 97-155 (1904).

4. D. Robertson, Whirling of a Journal in a Sleeve Bearing, Phil. Mag. [7] 15,

113-130 (1933).

283
 INFLUEr'CE OF PARAMEITER ClJAN]E S 'ID STABILITY BFRAVIOR OF ROKR.S

Claus P. Fritzen and Rainer Nordmann

Department of Mechanical Engineering
Uhiversity of Kaiserslautern
Kaiserslautern, Federal Republic of Germany

SUMMARY

The occurence of unstable vibrations in rotating machinery requires

corrective measures for improvements of the stability behavior. Before a measure
will be realized, different possibilities should be investigated. In this paper
a simple approximate method is represented, to find out the influence of para
meter changes to the stability behavior. The method is based on an expansion of
the eigenvalues in terms of systemparameters. Influence coefficients show in a
very intuitive way the effect of structural modifications. The method first of
all was applicated to simple nonconservative rotor models. Furthermore it was
approved for an unsymmetric rotor of a test rig.

INTRODUCTION

The occurence of instability in rotating machinery may be caused by

different effects, for example oil film forces in journal bearings, forces in
seals, unsymmetric shafts, internal damping etc. A machine designer wants to
know, whether a rotor will run stable during operation and what size the stabi
lity threshold speed will have. Furthermore he needs information about the pa
rameters influencing the stability behavior.

Important informations about the stability of a linear rotor system can be

obtained by calculations in the design stage. Because of uncertain input data
for calculation, the results have to be considered critically. It is possible,
that unstable vibrations may occur during operation, although the calculation
was predicting a stable machine. On the other hand parameters may change during
operation leading to increasing oscillations. In such cases suitable corrective
measures for improvements of the stability behavior are required. Before a
measure will be realized, different possibilities should be investigated finding
out the simplest and most effective one. For that it would be very useful, to
have approximate formulas, expressing the sensitivity of the dynamic behavior to
changes of system parameters.

In linear rotor systems with nonconservative effects the stability can be

valuated by means of the system eigenvalues. The real parts of the eigenvalues
determine, whether the natural motion is decreasing or increasing. If the
variations of eigenvalues caused by variations of systemparameters (mass-,
damping-, stiffness-coefficients) are known, an estimation of system modifi
cations to the stability behavior is possible. Such sensitivities, respectively
influence coefficients, expressing the change of eigenvalues to changes of
systemparameters are presented here.

284
 Lund (ref. I) nas developed a method to calculate sensitivities of the
critical speeds (eigenfrequencies) of a conservative rotor to changes in the
design. Dresig (ref. 2) gives a more general development for conservative mecha
nical structures. The basic idea in his method is an expansion of the eigen
values in terms of the system parameters. In Taylor's expansion derivatives of
the eigenvalues to the system parameters are needed. Such derivatives were
developed from Plaut and Huseyin (ref. 3) and from Fox and Kapoor (ref. 4).

Based on Taylor's expansion for complex eigenvalues in this paper an

improvement of the method is presented for rotor systems with nonconservative
mechanisms (oil film bearings, seals ets.) Linear, quadratic or higher order
formulas are obtained, depending on the order of the derivatives taken into
consideration in the expansion. With the linear formula very simple influence
coefficients can be defined, pointing out the influence of special parameters
to an eigenvalue. Superposition of different parameter changes is possible in
this special case. Using the formulas the eigenvalues, the left-hand and
right-hand eigenvectors of the original system (without parameter changes)
must be known.

The application of the method is demonstrated for simple nonconservative

rotor models, investigating the effects of stiffness and damping coefficients
to the stability behavior. Furthermore the influence of the mass variation to
the eigenvalues of an unsymmetric rotor is determined. The predicted vibration
behavior caused by parameter variations could be confirmed by measurements.

NATURAL VIBRATIONS OF LINEAR ROTORS

A turbomachine consisting of a high pressure turbine, a generator and an

exciter is shown in figure I. The power of the machine is 110 MW and the opera
ting speed 3000 rpm. Contrary to nonrotating structures the dynamic behavior of
such rotating machines is influenced by additional effects. Of great importance
are selfexciting and damping effects (nonconservative effects), caused by the
oil film forces of journal bearings, forces in seals etc.

In linear rotor dynamics the natural vibrations can be described by linear

equations of motion, usually derived by means of the finite element method.

(1)

M mass matrix (order NxN)

C damping matrix (order NxN)

K stiffness matrix (order NxN)

displacement vector

285
The equations express the equilibrium of inertia, damping, and stiffness forces.
Because of the nonconservative effects damping and stiffness matrix contain also
skewsymmetric and nonsymmetric terms besides the symmetric ones. Furthermore some
of the matrix elements depend on the operating conditions of the machine (speed,
power, pressure etc.).

Investigating the natural motion (stability) of a rotor equation (1) has

to be solved. The solution has the form

--J At
(t) = e (2)

Substitution yields the quadratic eigenvalue problem

(3)

with 2N eigenvalues A and corresponding eigenvectors (natural modes) q The

eigenvalues as well a the eigenvectors mainly occur in conjugate compl x

pairs

eigenvalues: A = 0'. + iw , A = 0'. - iw (4)

n n n n n n

eigenvectors: q = s + it = s it (5)
-n -n -n n -n -n

The part of the solution, which belongs to such a conjugate complex pair, can
be written as

0'. t
n
q (t)
-n
= B
n
e {s sin (w t + Y ) + t cos (w t + y )}
-n n n -n n n
(6)

w is the circular natural frequency and 0'. the damping constant (decay constant).
TRe damping constant, respectively the rea r part of the eigenvalue determines,
whether the solution q (t) decreases (0'. < 0) or increases (0'. > 0).
-n n n
The four lowest natural frequencies and the corresponding damping constants
of the turbomachine (figure 1) are plotted versus the running speed in figure 2.
The eigenvalues are changing with speed, especially the damping constants 0'.1 and
0'.3. The diagram shows that the rotor instability onset speed is 3400 rpm deter
mined by the zero value of 0'.1. The relative distance between the instability
onset speed and the operating speed is very low. Therefore unstable vibrations
may occur if additional destabilizing forces are acting on the rotor. If such
calculated results are known in a machines design stage, corrective measures
improving the stability behavior should be arranged. If instability occurs in

286
operating machines similar problems have to be solved. Approximate formulas,
respectively influence coefficients expressing the influence of parameter
changes to the eigenvalues (stability behavior) may be very useful for the
above mentioned requirements.

INFLUE NCE OF PARAMETER CHANGES TO THE

STABILITY BE HAVIOR

E igenvalue Problem and Modal Parameters

Our aim is to find formulas for an approximate calculation of eigenvalue

variations caused by system modifications (parameter changes). Besides the
designed parameter changes this formulas will contain modal parameters of the
initial dynamic system, that means eigenvalues A, left-hand eigenvectors t and
right-hand eigenvectors q of the system without parameter changes. Normally they
are known from calculations (ref. 5).

Complementary to the eigenvalue problem (eq. (3 the corresponding left

hand-eigenvalue problem with transposed matrices

(7)

has the same eigenvalues A, but the mentioned left-hand eigenvectors t. Working
with the procedure of inverse vector iteration for complex eigenvalues (ref. 5)
a desired number of the modal parameters A, q, t, can be calculated effectively.
For further derivations we suppose that the Interesting modal parameters of the
initial system are given.

Taylor's Expansion for the Eigenvalues

The basic idea of the approximate method is an expansion (Taylor's series)

of the eigenvalues in terms of the generalized system parameters Pk
(k 1,2,
= K)

I d d d r

I
A A + E -/a- Pi +"3 P2 + "3 PK } A(P ' P ' .P ) (8)
2 K
n no r. I

P P2 PK
r I
0 0 0

1n which the p may be mass, damping, or stiffness parameters.

A
is the chan ed eigenvalue after a parameter variation. A is the corres
g
p nding eigenvalue of the initial system. The derivatives o Othe eigenvalues
to the system parameters are expressed in operational notation. r is the order
of the derivative.

287
 Truncating equation (8) after a desired derivative, the changed eigenvalue
may be calculated approximately. Easy to handle is a linear formula with first
derivatives. A more exact expression can be obtained if an additional quadratic
term is taken into account.

Linear Approximate Formula

Taylor's expansion (eq. (8, truncated after the first derivatives leads
to the following linear approximate equation (9)

A (9)
no

For determination of A , respectively A

n A - A the first partial derivatives
n = n no
dA/dp are required.
Usng fhe standard subscript notation for partial derivatives

dA dq dM de dK
A
n,k = :
dP
q
n,k = :
dP 'k = dP
k
g 'k = dP
k
!'k = dP
k
( 10)

the differentiation of equation (3) with respect to P yields

k

{A + A
n
+ K} n,
k
+ {2AnAn,
k
+ A
n,kf
+
A '
k
+ A Q'
n k
+
' } n
k
=
0
(11)
T
We premultiply equation (11) by to obtain the scalar expression
n

(12)

The first term in this equation is zero, it represents the left-hand eigenvalue
problem, multiplied with q k' Furthermore it is assumed that q and & are
-n n
normalized in a way to satsIy the relation

T
{2 A M e (13)
-n n- -} n
+ =

It follows that

(14)

288
Substitution of this result into equation (9) yields

A = (A -A ) = - L (15)
n n no

with the influence coefficients g

nk.
Finally we introduce -matrices

K K
t.M L L
k=1 k=1

'k Pk

K K
C- = L : L (16)
k=I k k=l 'k Pk

K K
K
- = k=l
L =
g
k l
K
'k Pk

expressing the change of mass, damping, and stiffness matrices and we obtain
the linear expression

T 2
A = (A -A ) = - t {A t.M + A C + K} q (17)
n LIN n no LIN -n no - no- - _n

This approximate equation is a good tool for calculation in many cases, pointing
out the influence of parameter changes to the stability behavior of rotors.
As above mentioned for application of the formula the modal parameters of the
initial system are needed besids the parameter changes. Derivatives of the
eigenvectors do not appear in equation (17) .

Influence Coefficients

In equation (15) we have defined influence coefficients g

nk

OA
n _ l: 2
g -
- - tT {A M + A C ' + K } q (18)
nk oP _ n no- 'k no k
--
_ 'k _ n
k

This differential quotient, respectively differential sens1t1v1ty yields the

influence on the eigenvalue A of an infinitesimal change of a particular
parameter P . It can be used
k
R
s a meaningful approximation for finite parameter
changes, when the considered modifications are relatively small.

6.w
n
+ i (19)
P
k

289
Influence coefficients are complex numbers. They depend on the modal parameters
of the initial system.

k
Changing the stiffness k of a spring with one end fixed and the other end
.
free (see Table 1), we obtain wLth M'k = 0 and 'k = 0

- T K' q (20)
-n - k -n

The derivation 'k leads to a unity main diagonal element k = I with all
other elements zero and therefore the influence coefficient of the spring is

(21)

Table contains further influence coefficients for the most important

elements in rotor dynamics like springs, dampers, masses, oil film bearings
and seals.

Quadratic Approximate Formula

Taking into consideration the quadratic terms in Taylor's expansLon too,

we obtain the following expreSSLon
2
K dA K K d A
n n
A = A + E Pk + E E (22)
n no k l dPk 2 k l
= = =1
o o

in this improved formula the second derivatives A'k are needed besides the
first derivatives and the given parameter changes. The determination of the
second derivative is shown in the appendix. The expression is more complicated
than the first derivative, but all quantities can be determined from the initial
system.

PARAMETER CHANGES AT SIMPLE ROTOR MODELS

We applicate the method at simple rotor models to show the influence of

particular parameter changes to the stability behavior. Naturally these simple
examples can be solved also exactly, on the other hand they are suitable to
demonstrate the approximate formulas. A comparison of the results with exact
solutions is possible.

290
 Rigid Rotor with Flexible Supports and Clearance Excitation

The first example is an elastically supported rigid rotor with mass m

3). kI t k
22
(fig. The supports have the stiffnesses in horizontal direction and
in vertical direction. A clearance excitation force acts in the middle of the
rotor. For instance such excitation forces appear in steam turbines. They result
from the unsymmetrical fluid flow through the radial clearances at rotor and
blading which appears according to the eccentricity between rotor and casing.
The clearance excitation force acts rectangular to the displacement direction
(fi. 3).

(23)

The coefficient k depends on the power of a turbomachine, increasing with power.

Investigating the translatory natural motion we employ the coordinates

q l and q2 .

Equations of Motion, Eigenvalues and Eigenvectors

The equations of motion for the rigid rotor with elastic supports are

[:
or with the definitions
(24)

2
W 2kIl/m T = w t ( ) = d( ) I dT
0 o

= (25)
k2z1kII
Y stiffness ratio of support

= k/2 l
k I
S dimensionless clearance excitation coefficient

in dimensionless form

(26)

291
The corresponding eigenvalue problem s (V = A/W )
o

2
+ V
= 0 (27)
-13

with the characteristic equation

2
+ y+S o (28)

2
we obtain the solutions Z = V from

_ l+y (29)
2

respectively the four eigenvalues V as a function of y and S.

n

Figure 4 shows the real part and imaginary part of the essential eigenvalue
with positive damping constant. The system is unstable within the range of
(1-213) y (I + 213). This range increases with increasing values of the
clearance excitation coefficient S. Isotropy (y = 1) of the supports is the most
disadvantageous case. An increase of the anisotropy stabilizes the rotor system.
For (I - 213) 2 y 2 (1 + 213) the real part of the eig e nvalue is zero and the
natural frequencies are split.

The amplitude ratio of the right-hand eigenvector corresponding to an

eigenvalue V can be determined with equation (27), the left-hand eigenvector has
the opposite sign

2 2
I+V l +V
-- -- (30)
13 13

Parameter Variations

To demonstrate the approximate formulas we choose an initial system with

parameters y = 1.1 and 13 = 0.1 (fig. 5) and investigate variations of y, 13 and
an additional damping D. With normalized eigenvectors the particular expressions
for eigenvalue changes are

-variation of y with increment y

292
 -variation of S with increment .6.S

[:S
.6.D =
[DD tillJ [:J =0_
-additional modal damping c/2mw

211 (i)q) + i2q ) till

The linear approximate formula points out the stabilizing effect of anisotropic
supports (.6.y positive in fig. 5b). Figure 5a shows an increase of frequency
caused by stiffening the system from y to y + .6.y.

instance in the isotropic case (y =

For special cases the linear approximate formula has disadvantages, for

results will be obtained with the quadratic formula

I) with horizontal tangent line. Then better
(fig. 5b).

The destabilizing effect of increasing clearance excitation coefficient S

and the stabilizing effect of additional damping .6.D are presented in figure Sc.

Rigid Rotor in Journal Bearings

=
Figure 6 shows a rotor of a test rig. Compared to the bearings the stiff
ness of the shaft is very high, therefore it can be regarded as a rigid rotor
supported in two equal bearings (cylindrical bearings BID 0.8), rotating with
the rotational speed st. The rotor has a mass m = 72 kg, the distance between
the bearings is 660 mm . The statical bearing load F is equal to 353 N. Each
S
journal has a diameter of 50 mm, the values for the E& ial clearance and for the
oil viscosity are .6.r = 210 m, respectively 3.35 Ns/m2

The dynamic behavior of journal bearings can be characterized by four

stiffness - and four damping coefficients k and c ' respectively by the non
ik ik
dimensional quantities

.6.r .6.r st
(31)
F F
stat stat

They are functions of the Sommerfeld number. The pure translatory motion can be
described by means of the coordinates q and q '
j 2

Equations of Motion, Eigenvalues and Eigenvectors

The following equations of motion for a rigid rotor in two journal bearings
describe the equilibrium of forces in the case of pure translatory motion and
without external loads.

293
 -

1 l :] I ] [:IJ
l: :J [%]
0 O kl k
11 12 l 12
+2 +2 0
c c k k q
21 22 21 22 2

(32)

Defining

2
W g/b.r, T = w t, W IJ/w , ( ) d ( ) / dT
0 0 0

] [] [YI YI2 J
the equations of motion can be written l.n the nondimensional form

[ J [i}* [ J
q
+
6 S
11 12 l
= 0
6 8 q
21 22 'Y21 'Y22 2

(33)
The statement for the natural motions

'" At llT ,
q. q. e q. e II \/w (34)
1, 1. 1. 0

[YII
yields the eigenvalue problem

611 6
2 12

qJr:J
+ -- II + II + II
w 'Y12 w
--

= 0 (35)
6 6
21 22
+ II + II
'Y21 w 'Y22 w
--

The four eigenvalues II are 1.. g. conjugate-complex quantities. If the e1.gen

values are known, the gorresponding eigenvectors can be calculated from equation
(35) .

Figure 7 shows the eigenvalue that causes the instability in dependence

of the dimensionless rotational speed w. Also the eigenvalue that belongs to a
rotatory motion is plotted, but it has no meaning for the stability behavior of
the rotor. The real part of the eigenvalue will come to zero at w = 2.66.
At this rotational speed the stability threshold is reached. The Sommerfeld
number here has the value S o = 0.563. The corresponding circular natural fre
quency w/w is 1.435.
o

294
 Influence of Parameter Changes at the Stability Threshold

We start from the stability threshold and we investigate how particular

parameters effect the stability behavior. Therefore we take the linear approxi
mate formula (17). First we ask about the influence of the single coefficients
Y' and B. /w. It can be judged by the influence coefficients in Table I. Figure
1k k
S shows tEe influence coefficients in the complex plane. For example, one can see
that an increase of Y is abilizn ?r that an increae of the damping coeffi
12
cients B /w and B -/w 1S stab111z1ng the rotor mot1on.
11 22

If we change the rotational speed, the Sommerfeld number and the stiffness
and damping coefficients also will change. Figure 9 shows the variation of the
eigenvalue in dependence of the rotational speed calculated by the approximate
formula. The formula figures the tangent line to the eigenvalue curve at the
expansion point. An increase of the rotational speed causes instability.

Finally in figure 9 the influence of the oil viscosity is shown. If we

increase the oil temperature from 38C to 44C, the oil viscosity will decrease.
The real part of the eigenvalues becomes negative and there is a stabilizing
effect.

ELASTIC ROTOR WITH FLEXIBLE SUPPORTS AND SEAL FORCES

A more complicated rotor-system, an elastic turbopump rotor with flexible

support an impeller and two plain seals is shown in figure 10. Between the
bearings where the basic shaft diameter is 64 mm, the shaft carries the impeller
mass of 55 kg. Two plain seals are mounted besides the impeller with a seal
length of 40 mm and a radial clearance of 0.3 mm. The dynamic characteristics of
the seals are given by the stiffness, damping and inertia coefficients K, k, C,
c, nf (fig. 10). The shaft is supported in two identical bearings, each one having
7
a horizontal stiffness k 1.0 107 N/m and a vertical stiffness k =.75 10 N/m.
22
=

lJ
The distance between the Bearings is equal to 1.2 m.

First of all eigenvalues and eigenvectors were calculated for the described
original system in a speed range from 2000 rpm to SOOO rpm. In figure 11 two
damping constants (decay constants) are plotted versus rotational speed. One of
the damping constants crosses the zero axis, pointing out the instability onset
speed of 5200 rpm, which is above the operating speed.

In order to improve the stability behavior, some parameter changes were

investigated, starting from the initial system with a rotational speed of 5000 rpm.
Figure 11 shows the results, especially the influence of seal stiffness and dam
ping coefficients. All coefficients were changed 20 per cent. An increase of K
and C stabilizes; an 1ncrease of k has a destabilizing effect.

Furthermore it is shown that a positive change of k 1n direction to

22
isotropic bearings is disadvantageous.

295
 VARIATION OF THE MASS OF TEST

WITH UNSYMMETRIC FLEXIBLE SHAFT

The test rig rotor shown in a f Ie shaft with

rectangular cross section 8 x 12 two ball which
are comparatively stiff. In the a disk supported,
having a mass of 0.89 kg. The nonrotating shaft ifferent natural fre
quencies 24.6 Hz and 35.1 Hz concerning to the different tiffnesses in two
rectangular planes.

It is well known that in rotors with moments different

dynamic effects may occur, for instance weight
and unstable vibrations (ref. 6). this case
a variation of the disk mass was inves and
approximate formulas) and for control
tical results for the eigenvalues of the in figures
13a and 13b versus the rotational , The eigenvalues
defined for the rotating coordinate system are . 13a)
or pure positive real (fig. J3b). The last one rotational speed range
between the two natural frequencies of the nonrotat . That means that
the rotor motion is unstable in this range.

In order to show the util of the

the eigenvalues (rotating system) caused a pas of
the disk mass was investigated for the three different 1200 rpm,
1800 rpm and 2400 rpm. The three figures 14a, 14b 4c show the variation of
the natural frequency, respectively the of the
mass. The mass of the original system is 1. I kg.
The three diagrams contain the exact, the 1 approximate
solution. Furthermore the results from measurements In f 14a
and 14c. There is a good agreement of the results. rpm the mass change
decreases the frequency (destab effect , for the mass change
increases the frequency (stabilizing effect . effect is also
given at the running speed 1800 rpm, where the a is lowered by
the mass change.

REFERENCES

I. Lund, J. W. , "Sensitivity of the of a Rotor to Changes 1n the

Design," J. of Mech. Design, Vol. 5-1 I 1980,

2. Dresig, H., "Methode zur Berechnung des Einflusses

auf die Eigenfrequenzen von H""UU"'Sys
pp. 427-430, 1977.

296
3. Plaut, R. H. and Huseyin, K. , "Derivatives of Eigenvalues and Eigenvecto'rs
in Non-Self-Ad j oint Systems", AIAA-Journal, Vol. 11, No. 2, pp. 250-21,
1973.

4. Fox, R. L. and Kapoor, M. P. , "Rates of Change of Eigenvalues and Eigenvectors",

AIAA-Journal, Vol. 6, No. 12, pp. 2426-2429, 1968.

5. Nordmann, R., "Eigenschwingungen von Maschinenwellen mit Dampfungen und An

fachungen, " VDI-Bericht, Nr. 221, pp. 157-164, 1974.

6. Gasch, R. and Pfiitzner, H. , "Rotordynamik ", Springer Verlag Berlin,

Heidelberg, New York, 1975.

297
 APPENDIX

Calculation of the Second Derivatives of the Eigenvalues

First we have to express the right- and lefthand eigenvalue problem

(eq. 3), (eq. 7) in the following form:

(A
- - An-
B) x = 0 (A 1)
-n
T
y (A - A B) = 0 (A 2)
-n - n-

where the matrices !, B and the vectors x and v are defined as

41 "'-n

[: : J :J
o
A B (A 3)
- M

x
-n

- [] Y
n
(A 4)

The vectors and satisfy the biorthogonality relation

Zn

T
Y B x = 0 (A 5)
'-n -m
- nm

where 0 is the Kronecker delta. The first derivative of A with respect to the
nm n
parameter p . has the form
J

T T
A . (y B x ) = Y (A,. - A B,.) x (A 6)
n,J -n - -n -n - J n- J -n

For the calculation of the second derivatives of A , we need the derivatives of

n
the eigenvectors x and y The vectors x . and y . can be represented as a
. n -n,J -n,J

. --n0 f x -and y:
l1near comb1nat10n
.
-n -n
2N
x . =
-n,J .L: c . . x. d .. y. (A 7)
1=I nJ1 -1 nJ1 -1

Using the ,equations (AI), A2), (A5) and (A7), Plaut and Huseyin (ref. 3) obtain
the following expressions for the coefficients c .. and d ..
nJ1 nJ1

298
 T
c . + d . = - v B, x
nJn nJn ..Lon - ..;-n

T
y. (A, .-A B, .)x
-1 - J n- J -n (A 8)
c ., -
_

nJ1 A - A.
n 1
for t'.I=I: i

T
Y (A . -A B,. )x.
-n -,J n- J -1
d .. =
nJ1 A - A.
n 1

In order to get the second derivative of A with respect to p. and P we must

Ji,
differentiate eq. (A6) with respect to p .ith eq. (AI), J (A2), (AS) and
Ji,
(A7) we obtain for A ' 0 (ref. 3)
n,J,I(,

An,J,I(,
0

+ (c . + d . ) A + (c + d ) A
nJn nJn n,Ji, nJi,n nJi,n n,j

2N
+ 1: 1
(A - A.) [d .. c ' + d ' c .. ]
1= n 1 nJ1 nJi,1 nJi,1 nJ1 (A 9)
i+n

where c .. and d .. can be calculated by eq. (A8).

nJ1 nJ1
If we go back to our initial problem with the matrices, f and! and
the eigenvectors! and , we get by substituting , , and for the second
derivative

oT 2 + K,. o )
A = ,I(, (A M,.o + A C" o q
'n
n,J,I(, -n n - J,I(, n - J,I(, - J,I(, -n

T
- { Ji, (2A M,. + C,.) S } A 0
-n n- J - J n n,,I(,

- {

2N
+
1: 1
1=

(A - A.)
i:/:n n 1

where G . means (A 10)

-nJ

299
 2
G A. M,. + A. C,. + K,. (A II)
-nJ

n - J n- J - J

The second derivative with respect to a single parameter p. can be obtained

by replacing the subscript in eq. (AIO) by the subscriptJj.

Element Force-motion-relation Influence coefficient

gnk

q
k dA

:l
- O\ q )
-
F k

k qk
Spring k n
k gnk dk
= = =

, \,
Damper F c q
k gnk
I A (Q'
n o kq k) n
= = =
-

k k 0\
0

dA I

l
:k - A 2 ( q )
.) -
Mass F qk gn k d no k k n
= = =

k
o

t l;:-"] [ ::::::j[:]t::::l:J
Journal - A U. q )
gnCi no 1 k n
=

bearing = k
q 1. - (. qk)
? gnk n
=

ik
1
I! , - " -

l'" gnC
- A (. .
no 1 q 1 + kq k)n

[::If ' lP f : J:J

Seals Of - A (. q - . )
Pumps gnc no 1 k kq 1 n
qk qi C q -k
-
k ( qi+ q
- - - gnK i k K)n

- (. q - q. )
gn k 1 k k 1 n

(q -kq)
1 k 1 n

300
 High Pressure Generator Exciter
Turbine

Clearance
Excitation Forces

Figure 1. - Rotor of turbomachine.

_----4
_-------3
3
>- 100
u
z
W
:J
@
LL 50
a::

..J OPERATING SPEED

3000 rpm
a::
:J

I-

z 0
0 2000 4000 6000
ROTATIONAL SPEED rpm
t1
I-
Z

If) 0 -----r---r-+---+---+
z 3 6000
0
u
E -0.5
_____ -1/
____ 4
Il..
H

:E:
_k--r(
o -1.0

....--+-... unstable

Figure 2. - Eigenvalues of turbomachine rotor.

301
 1.0

0.5

y = k22/kl1 j--ISOTROPIC CASE

CLEARANCE l3=k /2kl1 i )'"=1
EXCITATIO N COEFFICIENT k
i
12 i I
.5 STIFFNESS RATIO Y 1.5

0.10
.9...

T 0.05

STIFFNESS RATIO y 1.5

Figure 3. - Rtgid rotor with flexible Figure 4. - Eigenvalues of rigid

supports and clearance excitation. rotor with flexible supports
and clearance excitation.

1.
wtwo

0.8

f-'-_--------'1!-0 --------1-5 --
. V
0.1 b
two

0.05

0.10 c

t w o

0.0

Figure 5. - Parameter variations at rigid rotor with flexible supports and

clearance excitation.

302


30-
,0 , 3
>-
u
Z
IoU
::)
CJ
r- -c-_-..:I

4----+-
u..
2,0
-'
oc(
ROTA TION
w= ,66
r-
oc(
z
ROTOR WITH MASS m
Vl
2,0 Vl
---=------+1,0 -
JOURNAL BEARING z
o
COEFFICIENTS kik (ik
t::C:P
IoU
MOTION / REACTIONFORCE RELATIONSHIP TRANSLAnON
a
FOR A JOURNAL BEARING

-0,8 - 0,6 - 0,4 - 0,2 0 0,2

DIMENSIONLESS DAMPING CONSTANT /w.

Figure 6. - Rigid rotor in journal Figure 7. - Eigenvalues of rigid

bearings. rotor in journal bearings.

0.2 I-

BEAR. JBEAR 9 BEAR. IgBEAR.

a::
+-------'r?-1,50
1312/w gY22' Yn Y12
9
>-
a::
...

z:
0.1
I!J

:I:
- CHANGE OF OIL

BEAR. VISCOSITY

Y21
9

BEAR.-0.1

-0.2 0.1
F
... 131 /w
9 REAL PAR

-0.1 +---/-'----f; 1,40 ----1---

...
BEAR. BEAR.

1322/w
9 w= 2,5
1321/w
... 9

I -0.2

-0,05 0,05

Figure 8. - Influence of bearing Figure 9. - Variation of rotational

coefficients to stability speed and oil viscosity.

behavior.

303
 IMPELLE R
MASS m

MOTION/REAC TION-FORCE
FOR A PL AIN S EAL
/ RELATIONSHIP

-[ I ... I 1 I Ie I [1 I I [1]
F2
=
m

m q2
+
-
( C
c
q2
+
K k

-k K

Fi gure 10. - Elastic turbopump rotor.

10

5
k221 k22 = 0,2
1/5 I

o r------ (/( =O,

I
k 1 k = 0,2 A-
I

Yo K/K= 0,2
I
./
V
C/C=O.2
I I

V
II

-5

-10

-15
o 2000 4000 . 6000 8000
10000
ROTATIOAL SPEED rpm

Figure 11. - Influence of seal stiffness and damping coefficients

to stability b ehavior of turbopump rotor.

304
 BALL
DISK WITH MASS m BEARING

BALL
BEARING
ROTATIONAL
SPEED

600

Figure 12. - Test rig rotor with unsymmetric shaft.

IMAGINARY PART
REAL PART
EIGENVALUE EIGENVALUE

ROTATIO
SPEED
ROTAT IONA L
SPEED

Figure 13. - Eigenvalues of rotor with unsymmetric shaft.

305
 <;

<; ROTATIONA L SPEED

w 1200 rpm

115 .;
.,

.;..

o. 0 0.8 1.6 2. 2 2.

.;
MASS [kg]

cO
.!!:

cO

ROTATIONAL SPEED
a.
.;., 1800 rpm
1/5

o. I.A 2. 2 2.

cO
M ASS [ kg]

EXACT

MEASUREME

2.p . !

MASS [kg]

Figure 14. - Variation of eigenvalues in dependence of mass change.

306
 STABILI'IY EVALUATICN OF RarCR/BFARINJ SYSTEVl

BY PERIURBATICN TESTS

Donald E. Bently and Agnieszka MUszynska

Bently Nevada COrp.
P.O. Box 157
Minden, Nevada 89423

SUMMARY

The paper presents the stability study of rotor/bearing systems. Even though it was
limited to study of a fully lubricated bearing subject to oil whirl, and further
limited to low eccentricity region for linearity and with only one type of lubricant,
it can be seen that the perturbation methodology, together with the sorting of the
impedance terms into direct and quadrature with respect to input force can be very
useful to the general study of stability. Further, the concept of active feedback
should assist to increase knowledge ; n rotor system stability. While there remains
a large amount of study to be accomplished, perhaps some more tools.re available
to assist this field of analysis.

INTRODUCTION

The problem of rotor vibration in plain fluid bearing becomes more and more
serious as operating speeds of turbomachinery continually increase. Two of the
most significant sources of vibrations are fluid lubricated bearings and seals.
Despite the high interest of field engineers and researchers which has followed
the widespread application of oil/gas bearings and seals, the problems created by
their hydrodynamic actions in turbomachines are still not solved completely
[1-18].

This paper shows that the dynamic behavior of a rotor and its stability can be model
ed as an active feedback mechanism, which should greatly help in clarifying the
stability problems of rotor system. It further shows testing and data techniques
for bearings and seals, and the evaluation of the performance of a perturbation
test bearing.

The principal observed dynamic phenomena due to the lubricant action during shaft ro
tation are known as oil whirl and oil whip. Oil whirl appears at rotational speeds
lower than twice the first bending critical speed of the rotor (corresponding to its
first natural frequency), as a forward circular motion of the journal with the
frequency nearly equal to half the speed of rotation. This motion is related to a
loss of stability of a pure rotational motion (about a central position of the
journal in the bearing, or more generally - an equilibrium position, determined by
the static load) and creation of a stable steady circular motion with an amplitude
affected by nonlinear factors of the oil film. In the performance of real machines,
oil whirl may cause serious machine damage since the level of the steady-state
whirling amplitude may exceed admitted tolerances of vibration.

Oil whip also has the character of forward motion, but unlike oil whirl, is locked
to the frequency of the first self balance resonance (critical speed) of the rotor.
The rotational speed must be of at least twice the resonance speed for oil whip
instability to occur. Sometimes the rotor goes unstable in oil whirl, then converts

307
to oil whip when higher rotative speeds are reached. Oil whirl is not known to
occur at higher speeds than oil whip.

In earlier years it was believed that for all of the forward circular instabilities
that the range of rotative speeds from 2 to 2.5 times the self-balance resonance
speed was required. However, more recent works [1] show that such instabilities
may occur at any rotative speed ratio with respect to balance resonances, depending
on the type of instability. Some machines in the field with severe gas whip problems
show that this instability cannot be cured with the machine rotational speed below
the first self balance resonance.

The literature on oil whirl/whip phenomena is quite rich [1-18]. The results con
cerning the loss of stability of the pure rotational rotor motion are usually pre
sented in the form of stability charts describing the behavior of the rotor in the
vicinity of the steady equilibrium position for different variable parameters. The
stability rules indicated by different authors rarely agree in detail. This is not
surprising - rotor response is very sensitive to slight variations in parameters
(e.g., a change of temperature by a few degrees, or minor geometric asymmetry in the
bearing may stabilize the rotor), and it has not been recognized that the active
feedback is involved.

Classical analysis of the rotor/bearing systems usually starts from the very gen
eral Euler-Navier-Stokes equations. The solution of these equations, which may
supply information concerning magnitude and functional relationship of hydrodynamic
forces is unfortunately not easy to obtain. Analytical solutions require a lot of
simplfying assumptions (such as idealized boundary conditions, neglecting thermal
effects, ideal characteristics of lubricant, assumed pattern of the pressure distri
bution, neglecting curvature and capillarity effects of the lubricant, laminar flow,
etc.). The resulting solution represents, then, only a rough approximation, as some
of the neglected factors may happen to be dominant.

Numerical solutions of the Euler-Navier-$tokes equations, requiring, just as much,

a number of simplifying assumptions, involve a lot of time-consuming computations
[6] and can solve only some particular problems. Progress in computer facilities
and numerical techniques, however, improves the possibility of achieving solutions,
which allow for rotor/bearing coupling of motion and further parametric analysis.
Parallel to the progress in computer sciences, a remarkable progress in experimental
techniques and measurement facilities has been observed. The experimental results
bring better understanding of dominating phenomena occurring during rotor/bearing
motion. This, in turn, will clarify eventually the simplifying assumptions intro
duced in theoretical analyses.

In the present paper an experimental method of identifying the hydrodynamic forces

in a plain cylindrical bearing during the oil whirl condition will be dicussed.
The results are applicable to rotor/bearing or rotor/seal systems. The approach,
based on analysis of forced response of the perturbed system, presents a continuation
of previous research [7,8,9,12,13]. The mathematical model of the system considered
is based on the linear theory of the rotor motion and fluid flow effects, as this
series of tests have been deliberately run at low eccentricity numbers (generally
less than 0.5).

The agreement of the experimental results and the predicted motion is very good.
The method of investigation is shown to be very effective and sensitive to any
parametric variation, so that it can be highly recommended in many applications.

308
 TEST ROTOR SYSTEM AND THE PHYSICAL MODEL

The experimental rig to

perform the perturbation test
consists of two shafts with
circular cross sections (Figs.
1,2). The main shaft rotates
at the constant speed wR
(being quite well below the
first critical speed) in a
rolling, pivoting bearing
and in a plain cylindrical oil
bearing with diameter 0=211
axial length L=1. 511 and
radial clearance C=0.01511
The second shaft rotating at
perturbing speed wp in a pi
voting bearing and a rolling
antifriction bearing, is
attached to the journal of
the main shaft at the oil
bearing. The trans-
parent bearing, fed by blue
dyed oil type ISO-VG-32 from
an outboard tank with a
con 2 tant pressure p=7100
N/m permits direct obser
vation of the distribution
of the lubricant within the
bearing .

Both shafts are driven by

separate electric motors
Figure 1. - Rotor test rig. 1 - Bearing, 2 - Main
(0.75 HP for the main rotor,
shaft, 3 Perturbing shaft, 4 Perturbing
0.1 HP for the perturbing
- -

speed controller, 5 - Additional springs, 6 -

one) through a flexible rub

Main shaft tachometer, 7 Thermometer, 8 -
For compen ber coupling.
-

P roximitors, 9 Perturbed Reservoir.

sating weight of two shafts,
-

a pulling vertical force is introduced in such a manner that during rotation of the
main shaft, its response is a pure rotating motion in the center of the bearing. For
this pur pose , the main shaft ;s also straightened and balanced.
At the same section
of the shaft the additional symmetric springs can be int rodu c ed The perturbing force .

is c reated by a controlled unbalance at the second shaft (calibrated unbalance

masses attached to a single light disk mounted on the shaft). The unbalance inertia
force is then proportional to the square of the perturbing speed wp .

The motion of the shaft journal ;s observed by two displacement motion, noncontacting
probes (vertically and horizontally located) plus a Keyphasor probe to provide ac
curate phase and speed signals. Additionally the constant rotational speed of the
main shaft and slowly variable rotation speed of the perturbing shaft are recorded.
A special device designed to control constant angular acceleration of the perturbing
shaft was used. The Keyphasor, horizontal and vertical probe signals, filtered to

309
the main (perturbing) frequency, were continuously observed on the oscilloscope mon
itor (orbits and time variable displays) and stored in the computer. The computer
then provides numerical results and graphical representation of the amplitude/phase/
frequency relationships, or their derivatives ("Direct" and "Quadraturell Impedances,
see expressions (9) and (10)). The physical model of the system is presented in
Fig 2.

MATHEMATICAL MODEL OF THE SYSTEM

y
Pivoting Bearing
The mathematical model is
based on the assumption of
Measuring Probe
sma11" deflections, allow
II
Main Shaft Pivoting Bearing
Rolling Bearing Perturbing Shaft ing for linerization. The
vlt\
angular motions of the
shafts are then approx
m,9
imated by their linear
I, I, radial deflections. The
hydrodynamic forces are
introduced in the oil
Figure 2. - Physical model of system.
bearing are introduced
in their linear approxi
mation. It was assumed that the system has only two degrees of freedom (horizontal
(x) and vertical (y) deflections of the main shaft, as seen by the probes). The
shafts are supposed rigid. External damping is neglected. With these assumptions,
the equation of motion of the system is as follows:

(1)
where:
(2)

and
z == x + iy, i = Pi (3)

combines in a complex variable the horizontal (x) and vertical (y) deflections
of the journal; 11, 12, I ' I are moments of inertia of the main (1) and
I ?
perturbing (2) shafts (abo b t tn x or y and z axis), 1 . '7' are cor . . .

responding lengths (Fig. is main shaft rotational speed, w

2); w - perturbing
R p
shaft rotation speed, K stiffness coefficient of the additional springs; m ,
vertical constant pullingP
-

r-mass and radius of the controlled unbalance, P -

force, m1, m2 - masses of the main and perturbing rotors, g- gravity acceleration.
F represents the hydrodynamic force in the bearing:

F = (m + m i)z + (d -d i)z + (k -w k i)z, (4)

t r t r R t

where m,m d ,d ,k and k are bearing coefficients.

t, r t r t

The full symmetric linear two degree of freedom model of hydrodynamic forces is
taken into consideration.

SOLUTION OF THE MATHEMATICAL MODEL

We simplify the mathematical model (1)+(4) neglecting gyroscopic terms (J2==J2=0)

and the coupling "fluid inertiall term (m =O).
t

310
 If the condition of stability is satisfied,

- I f2 1 <wRkt/dr < I fl l , (5)

where fl,f2 are natural frequencies of the free vibration (see Appendix for
details) , then the dominating solution of (1) will be the forced response:

i (w t+y)
p
z = A e + B, (6)

where amplitude A and phase angle y are

2
/(D2
(7)
A = U wp + Q2),
y =
arctan (- 0/0), (8)

with

Direct Impedance: 0 = - (M + m) w 2 + d w + k +K (9)

p t P r
and
Quadrature Impedance: Q= d w - k wR . (10)
r p t
Note that the phase angle y
corresponds to the attitude
angle of the bearing.
ORIGINAL SYSTEM (0)

WITH SPRINGS
& MASS (SM) The static displacement in the
solution (6) is
(i(kr+K)-kt) (11)
HORIZONTAL WITH
SPINGS (5) w, PI
-2 -1 2 3 B = -----........-
..- ..-
...
-25 PERTURBATION SPEED w,
Irad/s 100J
(K+k r')2+k2t w2R
VERTICAL
-50
It will serve for evaluation of
-75
(0)
(5)
the stiffness coefficients by a
-100 (0)
preload test with no perturbation
-125 (SM) (w =0) . During the perturbation
c( (0) tet (variable w ) the pre-
150
REVERSE w
Q (SM) FORWARD load was eliminaed (Pl=O) .
5 mils

'1 = 64.4c pOise (T = 67 OF) iD

:=0 TEST RESULTS. IDENTIFICATION
w, 200 radls c(... 100
OF THE IMPEDANCES AS FUNCTIONS
=

I}l.s
z OF ROTATIONAL SPEED. STABILITY

III
W
CHECK
50 10)
a::

(0)
The experimental tests produced a (5)
(SM)
w,
set of results for the response
o 1 2 3 amplitude A and phase angle y
PERTURBATION SPEED w,
Irad/s 100J versus (slowly varying) per
turbation speed. The para
Figure 3. Phase frequency and amplitude/frequency meters of the set were rota
-

responses of system. tional speed and temperature.

The tests were performed for
three cases: with additional
springs (System IIS II), with additional springs and mass (System IISMII) and without

311
them (original system "0"). Fig. 3 presents an example of the test results for
a rotational speed w =200 rad/s.
R
PERTURBATION SPEED w, For a perturbation speed slightly
[rad/s 1001
less than half of the rotational
-2 - HIO) 3 speed, a well pronounced resonance
-3
,_ . '
/
VIO)
was observed. This pattern repeated
""--",. , HIS)
-1 VIS) '" for all values of the rotational
\.." '" speed. In the resonance vicinity,
_1000,J!>
-2
Inch
HISM) / the phase angle sharply changes
VISM) / ' values between zero and -90.
This behavior corresponds well to
-3 the predicted model of the sys-
H HORIZONTAL
V - VERTICAL tem (see expressions (7), (8)).
REVERSE FORWARD
The system without additional
springs and mass shows the highest
resonance amplitude. An addition
5.0
_ _ ORIGINAL SYSTEM 10)
-- WITH ADDITIONAL VISM)
of springs lowers the amplitude
SPINGS IS)

---- WITH SPRINGS

considerably.
& MASS ISM)

For the reverse perturbation speed

-3 resonance has not been observed
although the system attempts
to whirl forward. The ampli-
tudes of vibration slowly in-
2.5
crease with negative frequency,
-

'1 64.4c poise IT 67 OF) the phase angle slightly changes

WR 200 radls
at the frequency corresponding
-5.0
to minus half w . Major differ
R
Figure 4. - Impedances versus perturbation speed ences in horizontal and vertical
for three cases. phase angles are due to the 90
difference in probe readings. We
observe however, some nonsymmetricity in the system, especially in the -1/2w
R
frequency region.

The experimental data stored in the computer were transformed into the Direct
and Quadrature Impedances:

2
2 (12)
D = U w cosy fA , Q =-U w siny fA
p p

An example of the computer generated relationship between Impedances and perturbation

speed for w = 200 rad/s is given in Fig. 4. The graphs correspond to the results
R
presented in Fig. 3. In all three cases the symmetry of the system for the forward
perturbation speed was maintained - the amplitudes and phases showed to be the same
for vertical and horizontal responses, the orbits were circular. At the reverse
perturbation speed, we observed differences in horizontal and vertical Impedances.

Figure 5 presents a set of results of the response of the original system 110" for
variable rotational speed. All other parameters of the system were kept constant
during this test. The Direct Impedances have a roughly parabolic shape, while the
Quadrature Impedances are nearly straight lines.

312
 The data taken from Fig. 5 served to
obtain the cross relation for the
PERTURBATION SPEED c.vp Iradls 1001
Direct Impedance versus rotational
o r- 3
2r- 5
4
__ ____ ____ speed with the perturbation speed
. as a parameter (Fig. 6). An analy
FORWARD
w tical approximation to the Direct
c.J -1
z Impedance as straight line function

c
w -1000,JlL
Inch
of the rotational speed has been
II.
!!i -2 adapted:
I
D = C 1 W -C 2
c.J
R
.

w
II:
is -3
The next step of identification was
an evaluation of the coefficients C
-4 ------ and C
2
t
as functions of the perturba ion
5.0 r------:,----___, speed w . For this purpose, the val

ues of he Direct Impedance at zero
perturbation speed and the slopes
.... 2.5
of the straight lines present in Fig.
E 6 were crossplotted as functions of
the perturbation speed.
w
c.J
Z

w
II.
3 4
5 For both of these curves the parabolic
PERTURBATION SPEED [radls 100] approximation has been adapted [15J:
!!i cop

c
_3
-2.5
CI
C1 = - 1.8,10 w 2 - 2.5 w -200 [kg/sJ
p p
ORIGINAL SYSTEM (0)
2 4
I) = 64.4c poise (T = 67 OF) C2 = + 0.25 w -950 w - 9.5-10 [N/mJ
5.0 -----' p p
Figure 5. Impedances versus perturbation
- Finally the Direct Impedance is describ
speed for variable rotational speed. ed by for the following analytical
_3 form:
0/0.77 = -(1.8'10 w + 0.25)w 2 - w (2.5w -950)-200 w + 9.5-104, (13)
R p p R R

where 0.77 was not yet taken into account the dimensional factor. Comparing this
expression with the expression (9) derived from the mathematical model of this
system and introducing M=0.4 kg, K=O, the identification of the parameters of
the rotor can be completed:
_3
m = m(w ) = (1.4'10 w - 0.21) [kgJ
R R

d = d (w ) =-(1.9 w - 730) [kg/sJ (14)

t t R R
4
k = k (W ) = (7.3-10 - 150 w ) [N/mJ.
r r R R

The Quadrature Impedances (Fig. 5) were identified as the straight lines; varying
with rotational speed.

Q = (970-0.69w )(w -Aw ), (15)

R p R
where
W
l
p Q=o
A = (16)
w
R

313
 is the ratio of the pertur
bation speed for which the
300
100========
Quadrature Impedance has a
1====*===-
zero value to the rotational
100
__

o-l:: speed w . The parameter A

...

ROTATION,\L

appeare H to be nearly constant

(, SPEED

j
Z for all performed tests:
;;; -1 A"= 0.48.
u
z
'"

Its value agrees with the oil

;! -2 whirl ratio observed in prac
tice. The value Aw corre
R
sponds to the average oil
velocity.

The comparison between the

-4
expressions (10) and (15)
leads to the identification
Figure 6. - Direct impedance versus rotational sped of the system parameters:
and variable perturbation speed.
d = d (w ) = (970-0.69 wR) [kg/s]
r r R
k = k (w ) = (470 - 0.33 w ) [kg/s] = Ad (w ) (17)
t t R R r R
and m = 0, as it has been already assumed.
t
As we notice, all parameters of the system are functions of the rotational speed w .
R
It was not possible to obtain test results for very low values of the perturbation
speed. For this reason a Idisp1acement" test was performed. The main shaft rotating
at constant speed was perturbed by a constant pulling force P. The probes measured
vertical and horizontal displacements. The force-response characteristics showed
to be linear in relatively wide range of displacements, however, their slopes varied
considerably with the rotational speed. The formula (11) served for the identi
fication the parameters k and k (K=O in this test*):
r t
k = 4000 N/m (for 0 < w 1 50 rad/s) (18)
r R
k = ( 490-0.27w ) [kg/s].
t R
The evaluation of the coefficient k was not very precise. However, its relatively
r

* The calculation of the coefficients k and k was performed"indirect1y. First

t
the slopes of the straight lines corres ponding to the pulling force P versus
I
horizontal (x) and vertical (y) displacements were evaluated for different
values of the rotational speed.
The slopes are equal to "directional" stiffnesses K =P /x, K =P /y, corresponding
Taking (6) and (11) into account the s fifTness oefficients
1 1
to x and y.

k = Ky / (1 + (Ky / Kx ) 2) , ktw = k K K /
r R r y x

were then calculated for each value of the rotational speed.

314
small value was noticed especially for higher w . The agreement with the previously
evaluated coefficient k R
(expression (17)) is very good.
t

Having evaluated the parameters of the system the stability check has been
possible.

Introducing the expressions (14) and (17) into the stability criterion (5) and
rearranging terms, we obtain the cubic polynomical inequality:

1.29 w 3 + 3030 w 2 - 6. 17'105w -2.29-108 <0 (19)

R R R
The polynominal (19) has one positive root w = 366 rad/s.
R
The stability
criterion, then, has the form: w < 366 ratl/s.
R During the experiment it
o ORIGINAL SYSTEM
- was observed that for rot
D WITH SPRINGS & MASS
-
ational speeds higher than 350
rad/s the rotor became un
200 x WITH SPRINGS
stable, which is consistant
-

with the above analysis.

PERTURBATION SPEED

3 150 FOR D 0

o We notice that the value

w
w
c.. of amplitude peaks is close
<IJ o
ly related to the difference
100
11 STABILITY MARGINS between perturbation speeds at
CD
-

II: which Direct Impedance and

:::> PERTURBATION SPEED
FORO 0 Quadrature Impedance equal
to zero (Figs. 3 and 4).
ffi 50
c..

This difference is referred to

as the stability safety margin
o 100 200 300
(see Appendix). In Fig. 7
ROTATIONAL SPEED WR [rad/sl the values of perturbation
speed for zero impedances
Figure 7. Perturbation speed at zero impedances and
have been plotted. The mar-
-

stability margins. gin of stability decreases

with the rotational speed. For the considered systems ("0","5", and "SM") the
zeros of the Quadrature Impedance remain the same. The additional springs and
mass have influence on the Direct Impedance only.

At the reverse perturbation speed, the Quadrature Impedance is negative for

all speeds (compare (10) for W negative, see also Figs. 4 and 5). It does
not cross zero; the system is e ery stable.

Figure 8 presents the impedances for the squeeze film test (w =O). We notice the
R
symmetric parabolic shape of the Direct Impedance and antisymmetric (straight line)
for the Quadrature Impedance. (Slight displacement of the lines from zero at
W =0 is due to very low input force level). As expected, tangential stiffness,
b ing related to the rotational speed, equals zero when w =O.
. R
In the squeeze film test the Quadrature Impedance contain then only the radial
damping term d (see (10), which appears to be positive. The condition of
stability (5) fs satisfied for all rotational speeds. The displacement of the

315
 Direct Impedance parabole from
the symmetric position for w =O
R
-3 -2 -1 2
into the region of positive
perturbation speeds for w >O
3
o

R
indicates a relationship of the
tangential damping d with rota
t
tional speed. d =0 for w =O
g
and d >0 for w >O within the
vertical

R
1000
t
inch
-2
inves igated range of speed.
I (See (14) for comparison).
(.)
W horizontal
a: -3

The Fig. 9 presents the Direct

is
'7 64.4c poise IT 67FI
WR =
and Quadrature Impedances for
0 ISqueeze Filml
-4
variable oil temperature. The
variations of the Quadrature
W
(.)0
Impedance are quite significant
z
while the Direct Impedance
;3-
Wz
horizontal
remains nearly invariable in
a. 2.5

REVERSE the wide temperature range, espe

c
<!:
::l
cially for lower values of the
-3 -2 -1 d 0 perturbation speed. Let us
PERTURBATION SPEED w,
evaluate the Quadrature Impe
[rad/s 100] dance as a function of tem
perature. From the graphs Fig. 9
-2.5
FORWARD the relation between the slope
of the Quadrature Impedance
5.0 versus temperature and the
speed ratio A can be obtained.
Figure 8. - Impedances versus rotational speed for Finally the Quadrature Impedance
squeeze film. is evaluated as follows [15J:
Q/0.77 = (1700 - lOT) [W -A(T)w J [N/m] ,
p R
_3
A(T) = 0.48 (1.1 - 1.8'10 T),

where T is temperature in Farenheit degrees. Further we derive the relationship

between temperature and oil viscosity (measured by Cannon-Fenske Viscosimeter 200),
so that the Quadrature Impedance can be expressed in terms of the oil viscosity.
If the Direct Impedance is known for any temperature, then a correcting factor
can be introduced for the value of Direct Impedance corresponding to any different
temperature [15J.

Summarizing the test results, we see that the predicted model well reflects the
main dynamic features of the system. The Direct and Quadrature Impedances can be
identified easily as a function of the system parameters and in particular, as a
function of the rotational speed.

ACTIVE SERVOMECHANISM CONCEPT

The ability of a lubricated bearing to support both steady-state and dynamic loads
by forming a springy cushion consisting of a fluid wedge is well known. It is also
well known that a 3600 lubricated bearing has a propensity for instability known as
oil whirl/whip. In fact, even before the outstanding work of Newkirk and Taylor
[10J or Hull [11-13], Harrison [14J predicted in 1919 that the full lubricated
bearing would be unstable.

316
 The study of rotor instability has
generated a vocabulary including
PERTURBATION SPEED wp [redia 1001
such expressions as cross springs,
5
cross damper, (and even cross masses),
0
... negative damping, etc. Such cross
E actions exist as effects, unlike gyro
-1 scopic action which is a genuine cross
w
()
2 -10'
coupling physical law. The reason
inch
C -2
w
why these cross coupling bearing and
11-
'1 = 64.4c poise seal effects occur is that such ele
45.7
I-
() 35.6
ments clearly have a behavior much
w 3
-
a: 28.4 more complex than any possible passive
is
670 systems. These machine elements
-4 contain active feedback, in which
exists a mechanism transferring rota
0
tional energy into lateral vibration
0
... energy. This is true for every identi
! fiable rotor instability of the forward
2.5
w circular whirl and whip category,
i n ch
() 10'
2

including shaft internal friction,
C
w
11-
aerodynamic cross coupling or steam
0
2 3 5 whip. In particular, the behavior
C

-STATIC TEST PERTURBATION SPEED wp [rad/s 100] of a 3600 lubricated bearing is
::J
d DATA analyzed herein.. The situation
-2.5
in any gas or liquid seal is also
WR = 200 radls generally in the 360 degree category
so that the results can then be ex
tended to any seal or bearing dynamic
Figure 9. - Impedances versus perturbation Ibehavior. To show in simple descrip
speed with variable viscosity (temperature) tion that such systems are actually
of the oil. closed loop servomechanisms, observe
the performance of the bearing in
response to a steady-state load.

Apply a downward pressure to a shaft rotating anticlockwise with no initial

dynamic or steady-state preload. The obvious and well known result is that the
shaft moves downward very slightly, and moves strongly to the right. This
motion, of course, provides the converging lubricant wedge to support the
applied downward load. Since lubricant has virtually no spring effect of its
own, the action generates the spring cushion synthetically. Applying a down
ward steady-state load as a step function, the first effect is for the shaft
to move downward in response to passive direct (radial) damping (d ) . The
result of this downward action is to create a restriction of oil f l ow at the
bottom. This in turn creates a converging wedge on the left, which forces the
shaft to the right. This process continues (usually with some damped whirling),
until the shaft reaches an equilibrium position in which forces are balanced.
This position to the right is sufficient to create the constriction which
provides the wedge to support the shaft against the input force because the
input to the active element is motion and the output is force, as previously
described.

The dimensions of the active elements are that of impedance. It then seems
reasonable to present the rotor/bearing system as a servomechanism with a feedback
loop containing active elements of the system (Fig. 10). The general transfer

317
function of the system (described by Eqs. (1) and (4)) is given in the Appendix. As
it has been shown in the previous section, the parameters related to the bearing
depend directly on rotational speed - they represent active elements. The term
lIactivell refers to the mechanism of transformation of rotational energy into vibra
tion energy, which provides the major reaction to input forces.

The input to the system is represented by the exciting periodic force as well as
static load ((w i) - see Appendix).
p

The output is the motion -

vertical and horizontal dis
PASSIVE ELEMENT placements of the main shaft
TRANSFER FUNCTION
L[<I>(iWpll +L 1 z(iWp) (z(iw ))'
p
EXTERNAL .. .../
K+k, MW + Wp(-id.. + d, mwp) MOTION
FORCE - -

The output motion with respect

w
u to the input force shows all of
a:
0
LL the active and passive terms in
W
> simple series (see (A8)). As
i=
u such, it could be optional

which ones are put in the
ACTIVE ELEMENT active category, with the
TRANSFER FUNCTION
remaining ones left in the
ktwRe- ia
passive category. However,
the static loading test pro-
Figure 10. - Rotor-bearing active
vides a perturbation speed of
feedback diagram.
w =0 so that all terms con-
taining w as a multiplier are deleted. These terms de eted entirely are the
passive t g rms. Further, the squeeze film tests, where rotative speed w =O deletes
R
all terms containing w as a multiplier. The terms which are deleted entirely can be
R
considered .as the active terms. These observations indicate the proper location of
the terms.
If a further check is desired, since it is
fundamental that springs with negative
stiffness cannot exist as real passive
impedances, any negative stiffness term
component necessarily represents the active
portion of the feedback loop. Since spring
force is motion times spring coefficient,
in order to supply the restoring force
against input force the elastic force must
-
be at nearly 90 degrees with respect to
motion, because the motion itself is nearly
90 degrees with respect to input force
(Fig. 11).

As a result, the so called licross springll

is actually the very real (but syn
thetically manufactured by feedback)
Figure 11. - Force diagram for restoring spring force. In review of the
rotational speed below active behavior of such a bearing it is at
resonance.
the same time extremely simple with only
one moving part, as well as extremely complex, because this one working part
performs several functions.

318
Summarizing these considerations we describe the transfer function of the active
feedback loop in the following form (Fig. 10):
- ;0:
:
k -;w k sino: =k w (:os 0: -i in 0: )=k w e
- r.J M ....

r R t t R t R (20)
where k =k COSO:, k =k + k , k =k sin 0: =Ad
r t r r r t t r
The active loop then contains a fraction of the w dependent coefficient k
R
(radial stiffness) and full tangential stiffness coefficient. The last on is
entirely the result of oil wedge formation. As it was identified in the previous
section, the radial stiffness contains a constant term and a term varying with the
rotational speed. However, a precise identification was difficult in view of its
relatively small value.

It is supposed that the terms located in the feedback loop are directly responsible
for the transfer of rotational energy into vibrations. The angle 0: represents the
angle between the observed motion and the active force (Fig. 11). This angle is
usually a little less than 90 degrees for 3600 fluid bearing. For other types of
bearings it is nearer 0 degrees (e. g. for sleeve bearings with stable void islands in
the thick film region, generally referenced to as 11180 Degree" or IIHalf Sommerfeld"
bearings).

In the present study it has been found that the radial spring coefficient has a low
value, for 0: 900 cos 0: is very small this is reflected in the expression (20).
However, at this time we have not found a means to measure directly the angle 0:.

All other terms of the rotor system are located in the IIpassiveli part of the
servomechanism model (Fig. 10). Being functions of the rotational speed, they may
also modify the rotor response in some l Iactivell way, however to a lower degree than
the ones related to stiffnesses.

A rotating shaft in a bearing does (1) have passive impedance, (2) transfers energy
from input torque by dragging lubricant around, (3) moves sidewise of the input
force (as best it can) to provide a restriction of flow to the lubricant to build
a converging wedge for support, and finally (4) synthetically builds a force to
counteract the input force.

Rotor systems, as currently configured, are rarely well stabilized because (1) their
basic algorithims are complex, nonlinear and nonsymmetric, and (2) there has been
lack of recognition and knowledge of this feedback nature of rotor behavior. The
present paper throws some light on these problems, however much more research is
needed to provide answers to all questions related to the rotor/bearing/seal
stability.

REFERENCES

1. Tondl A., il Some Problems of Rotor Dynamicsll , Czechoslovak Academy of Sciences,

Prague 1965. .
2. Vibrations in Rotating Machinery, Proc. of the Conference, University of
Cambridge, The Institution of Mechanical Engineers, Cambridge, U.K. , 1976.
3. Rieger, N.F. , "Rotor-Bearing Dynamics: State of the Art 197611, Shock and
Vibration Digest, v. 9, No. 5, 1977.
4. IIRotordynamic Instability Problems in High Performance Turbomachinery", Proc.
of a Workshop, Texas A & M University, College Station, Texas, 1980.
5. Tribology International, 0301-979X/80/050233, JPC Business Press, October 1980.

319
 6. Hankey, W.L., IISelf Excited Oscillation in Fluid Flow ll, University of Dayton
Research Institute Seminar, May 1981.
7. Bently, D.L, Bosmans, R. F., 1I0il Whirl Resonance ll, Trans ASME, April 1979.
8. Bently, D.L, liThe Parameters and Measurement of the Destabilizing Actions
of rotating Machines and the Assumptions of the 1950'sll, in [4].
9. Kanaki, H., Mitsubishi Heavy Industries, LTD, Takasayo, Private communication,
1981.
10. Newirk, B.L., Taylor, H.D., IIShaft Whipping Due to Oil Action in Journal
Bearings ll, G.l. Review, v. 28, 1925.
11. Hull, LH., 1I0il Whip Resonance ll, Trans ASME, October 1958, pp. 1490-1496.
12. Hull, E.H., IIJournal Bearing Behavior Under Periodic Loading ll, G.l. Research
Laboratory, Rep. No. 55-RL-1354, Schenectady, New York, 1955.
13. Hull, LH., Darrow K.A., IIHydrodynamic Oil Film Stiffnessll, G.l. Research
Laboratory, Rep. No. 59-RL 2217, Schenectady, New York, 1959.
14. Harrison, W. J., liThe Hydrodynamic Theory of the Lubrication of a Cylindrical
Bearing Under Variable Loadll, Trans Cambridge Philos. Soc., v. 22, 1919.
15. Bently, D.L, Muszynska, A., IIStability Analysis of Cylindrical 3600 Bearing
Oil Whirlll, Bently Nevada Report, 1982 (to appear).
16. Adams, M.L., Padovan J., IIInsights Into Linearized Rotor Dynamics ll, Journal
of Sound and Vibration, 76 (1), 1981.
17. Lund, J.W., Thomsen, K.K., IIA calculation Method and Data for the Dynamic
Coefficients of Oil Lubricated Journal Bearingsll, Topics in Fluid Film Bearing and
Rotor Bearing System Design and Optimization, ASME, 1978.
18. Childs, D.W., Dressman, J.B., IITesting of Turbulent Seals for Rotor Dynamic
Coefficients,II in [4].

APPENDIX

AI. EIGENVALUE PROBLEM FOR THE EQUATIONS () AND (4)

The homogeneous equation (1) with hydrodynamic forces (4) has the following
eiqenvalues:

(AI)

where

(A2)

1
2k:lk: .t;:
(dr (dt-J1wR -J2wp ) + 2(mtkr -(M+m)ktwR)) ) JI 2/(v8 ( +m))
-1- 4 2 M , j=1,2.

The condition of stability (negative real parts of the equinvalues (AI) is as

follows:
2
+
dr((M m) k/mtktwR ) -dr (dt -J w - J wp )(mtkr - U1 +m )ktwR ) -(mtkr -(M + m)ktwR ) o. (A3)
1 R 2 >

For J1 = J2 = m = 0 it reduces to
t

320
 (A4)

or after transformation to

(AS)

where

(A6)

are natural frequencies of the system at the limit of stability, i.e., when
instead of inequality (A4) we have the equation. As we can easily check f1 and f2
are also the roots of the Direct Impedance (9).

For k > 0 the quantity

t

A = f1 - w k /d > 0 (A7)
R t r

gives a margin of safety for the stability. As the value w k /d is the root
R t r

of the Quadrature Impedance (10), the stability of the perturbed system is

assured if the zero of the Quadrature Impedance appears at a lower perturbation
speed W than the zero of the Direct Impedance.
p

All. FOURIER TRANSFORMATION OF THE EQUATIONS (1) and (4).

The mathematical model of the system can be presented applying the Fourier
transformation and the control theory formalism. From (1) and (4) we get
then

(AB)

where z (iw ) is the Fourier transformation of the variable z(t) (2), iw is

p p
the Fourier operator, D(iw )' Q(iw ) are given by (9),(10), L [ 0 (iw )] is
p p p

the Fourier transformation of the input forces 0 = Uw e iW t

p - iP1. [D(iw )
p
+

+iQ (iw )] is the transfer function of the system.

p

NOTATION
Amplitude of the response _4
Bearing radial clearance (0.015// =3.8 10 m)
Radial and tangential bearing damping coefficients
Direct Impedance
Acceleration of gravity
Moments of inertia of the main (1) and perturbing (2) shafts
about x or y axis (24.B gm2, 6.7gm2; I =50 gm2 for the case of
1
additional mass)

321
Moments of inertia of the main and perturbing (2) shafts about
the axis of rotation (0.0650gm2,0.03gm2; I =0.0652gm2 for
IZ
case of additional mass)
Radial and tangential bearing stiffness coefficients
Equivalent stiffness coefficient of the additional springs
(33100 N/m=189 lb/inch)
Length of the perturbing shaft (9.511=0.241m)
Distance from the center of bearing to the probe (0.62511=0.0159m)
Distance from the bearing center to the pulling force section (2"=0.0508m )
Length of the main shaft (11.75"=0.298m)
Distance to the position of the perturbing force (7. 7"=0.195m)
Masses of the main and perturbing shafts with accessories (280g, 150g)
Perturbing unbalance mass (4g)
Radial and tangential bearing " fluid inertia" coefficients
Original system
Pulling force
Quadrature Impedance
Radius of the controlled unbalance (33132"=0.0262m)
Radi us of the shaft journal (1.00"=0.0254m)
System with additional springs
System with additional springs and mass
Phase of the response - Attitude angle
Stability margin
Dynamic viscosity of oil (poise)
Oil rotation speed ratio
Perturbation speed (perturbing shaft)
Rotation speed (main shaft)

322
 AEBOOYNAMIC STIFFNESS OF AN UNJUJND B:CFNlRIC WfIRLIl'U CENlRIFlIW..

IlVPElLER WIlli AN INFINITE NU\llBER OF BLADES

Paul E. Allaire
Department of Mechanical and Aerospace Engineering
university of Virginia
Charlottesville, Virginia 22901

Lyle A. Branagan
Pacific Gas and Electric
San Ramon, california 94583

John A. Kocur
Department of Mechanical and Aerospace Engineering
university of Virginia
Charlottesville, Virginia 22901

ABSTRACT

This paper considers an unbounded eccentric centrifugal impeller with an

infinite number of log spiral blades undergoing synchronous whirling in an incom
pres sible fluid. The forces acting on it due to coriolis forces, centripetal
forces, changes in linear momentum, changes in pressure due to rotating and
changes in pr essure due to changes in linear momentum are evaluated.

NOHENCLATURE

Symbol Definition
-_._-

-+
a Vector acceleration

a Impeller perturbati9n

b Thickness of two-dimensional impel ler

Vector pressure forces

Vector surface forces on control volume

Vector force of the shaft on the fluid

Change in force in i direction due to perturbation in j

direction

K Principal stiffness
xx
K Dimensionless principal stiffness, (K/pU R )
xx R1 1
K Cross-cou pled stiffness
_yx
K pimensionless cross-coupled stiffness
yx
n Number of blades

This wor k was su pporte d in part by NASA Lewis Research Center, Contract No.
NAG 3-180 and in part by the University of Virginia Rotating Machinery and Controls
Industrial Research P r o gram.

323
 o Center of flow field

0" Geometric center of impeller

Dimensionless radii ratio, (R2/R1)

Inner radius of impeller

Outer radius of impeller

Radius of impeller with respect to center of flow field

fixed
URI
Average radial inlet velocity in c o ordin a t e frame

Radial velocity at point B

UB

VB Relative velocity of rotating coordinate system at point B

-+-

V An alytic velocity expression for force equation

Velocity in rotating coordinate system at point B

WB
-+-
Vector velocity at any point in rotating coordinate system
WTOT
B Blade angle ( t an B = UR1/wR1)

Dimensionless eccentricity

8 Angle measured from x axis

Entrance angle

Exit angle

Blade sweep angle (8b = cotBn(R2/Rl))

Angle of f luid p a rtic le alon g blade

p Fluid density

w Angular speed of the impeller

1. INTRODUCTION

The analysis of turb o mac hinery requires a comb in at io n of solid mechanics and
fluid dynamics. Vibrations of a centrifugal pump or compressor are c o nt r o l l ed by
the bearings, by the shaft geometry, by fluid forces on the impeller , by seals,
and other factors. Often the nature and magnitude of the forces and stiffnesses
ge nerated by b ear ings and shaft geometry are fairly well u n der stood and can, in
ge ner al, be accurately modeled. Those fo rc es acting on the impeller can rarely
be adequately determined as yet. These forces arise from the inter a ct io n of the.
impeller with the driven fluid, and are often called " aerod ynamic forces,1I "aero
dynamic.stiffnesses" or "aerodynamic cross-coupling".

Much of the la rge body of work on incompressible flows in impellers assumes a

ce ntered impeller wit h varying degrees of complexity, A few examples include one

d imens ional velocity vectors [1], two-dimensional considerations [2,3], and full
thr ee -dimens ion a l anal ys is [4,5,6]. These are not discussed fu rther.

Calculation of aerodynamic forces was first reported by Alford [71. considering

circumferential variations of static pressure and efficiency in axial compressors.

324
Work by Black [ J
8 , Barrett and Gunter Lund [9J, [10J,
and Gunter, et ale ll . have [ J
demonstrated the importance of aerodynamic forces to the safe operation of turbo
machinery, but have not, in general, addressed the calculation of these forces, or
their. associated stiffnesses. Further, in the work by Barrett [9J and Gunter [llJ ,
these forces are modeled by their cross-coupled stiffness only; consideration is not
given to the principal stiffness of the fluid-impeller interaction.

Several recent papers have sought to more accurately quantify these forces.
Colding-Jorgensen [12J used a simple source and vortex flow model for the impeller
along with a series of vortex sheets representing the pump volute. He then super
imposed a uniform flow over this model, without clear physical explanation, to deter
mine the forces on the volute, as a function of the volute spiral angle. The flow
field emerging from the impeller is assumed to correspond with the typical one-dimen
sional analysis, except that its location is per turbed. The force calculated by this
method is the total force acting on the entire impeller/volute model for uniform flow
- not the force on the impeller.

The work by Shoji and Ohashi [13J considers an ideal impeller whirling about
its geometric center. Blades are modeled by vortex sheets using unsteady potential
theory with shockless entry to predict the flow field within the impeller. Forces on
the blades are summed from the pressures calculated by an unsteady Bernoulli equation.
The calculation is two-dimensional without volute or vaned diffuser effects but does
allow for a whirl frequency other than the rotational speed.

The papers by Thompsen [14,15J outline the assumptions and general theory con
tained in a proprietary program. The technique uses unsteady flow theory and deter
mines the stability of the rotor stage based on nonsynchronous fluid excitations.
The proprietary nature of the program limits the discussion of the details of the
procedure. An actuator disc method for calculating forces is currently being used by
Chamieh et ale [ 16J; however, the work is incomplete and detailed results are lacking.

A paper by Jenny and Wyssmann [ i7J suggests the aerodynamic excitation is

approximately two orders of magnitude smaller than other destabilizing mechanisms.
Their simplified analysis, however, considers only the variation in radial clearance,
while neglecting the change in momentum of the fluid. In a radial flow machine, the
change in force due to clearance variations should be negligible compared to changes
in force due to momentum variations since the former are perpendicular to the main
flow path.

A paper by Washel and von Nimitz [18J gives an empirical formula for aerodynamic
cross coupling forces in compressor impellers. It is used for comparison to the
theory developed in this work.

The purpose of this paper [1 9J is to consider the nature and magnitude of aero
dynamic cross-coupling forces 1:md stiffnesses as generated by the change in fluid
momentum in an eccentric impeller. The approach is to perform a control volume
analysis on a perturbed, infinite bladed impeller undergoing a steady orbit at the
rotational speed of the shaft. This assumes incompressible, ideal flow without
volute or diffuser effects. It further assumes that the impeller consists of an
infinite number of infinitely thin blades of simple geometry; the use of which allows
an analytic expression for the aerodynamic forces to be obtained. Boundary layer
development separation, and secondary flows, however, are not considered. The
forces calculated are generated by a perturbation of the impeller about the flow
center and thus are related to principal and cross-coupled stiffness terms. A
,

325
 sep arat e analys is usin g a perturba t ion velocity would have to be p e r formed in order
to d et e rm ine principal d amp ing terms for the impeller.

2. VELOCITIES IN AN ECCENTRIC IMPELLER

Fi gur e 1 shows a pert urbed centrifugal impeller whe r e O is the geometric center
of the imp eller and 0 is the center of the st eady flow field. It is assumed that the
flow en te rs through an inlet pipe cent e r ed at 0 and the radial ve loc it y varies in
versely with ra d ius away from this point. The eccen t ri c i t y a is small such t h a t the
dimensionless eccentricity

a
e: =
Rl

is much less t han unity. For this a n aly s is terms of order (2 and higher ,.,rill be
n eglect ed . The impeller consists of an infinite number of infinitely thin logarithm
ic spiral b l ad e s with angle 8. Th e blades are backward curved with a blade angle
8b, an inner radius Rl, and outer radius R2'

The analysis is to be performed in rotating at rotor speed

a coordinate system
w about the center of the impeller. about the flow c en t er
Only synchronous whirling
o w i ll be considered. The fluid is a ssume d to follow the shape of t he blades through
the imp eller passage and to exit pa r allel to the blades. The particle path in the
absolute reference f rame is shown in Figure 2. The impeller is purely t\.,ro-dimen
s iun al and ideal, incompressible flow is assumed.

The radial inflow to a centered impeller in the fixed (non-rotating) coordinate

system centered at 0 is

.& ( 1)
r

where the average inlet velo c ity is

I r 2u1T l,el)del
URI =
Z; .1 (R
0

Here 61 is the ang le of a f luid particle measured along r = R/. The impeller
ro t a tes with angul ar velocity w. The blade an gl e is given by

URI
tan 8
wRl
=

From Fig. 1, the length is

RB = Rl + a easel

Th u s, along an eccentric impeller centered at 0; the inlet velocity is

URIR1
I .
u = (2)
I + f: cosel

326
 From the binomial approximation

1
1 - e: cosSI
1 + e: cosSI
.::=

which leads to

(3)

This is the assumed form.of the flow entering the eccentric impeller at a p oin t B
around-the impeller.

From the geometry of the eccen t ri c impeller, this ve l o c i ty can be resolved in t o

components normal to and t angential to the impeller (shown in Fig. 3)

UBN = U coso .::= U (1 - e: cosS )

B R1 l

UB .-U sino .::= -U (1 -e: c os S l ) sinS! -UR1 sin81

T R

B
= =

for 0 < < 1. The relative velocity of the r otat i ng impeller is

along the inner radius of the impeller. Resolving this velocity into normal and
tangential components gives

VBN = VB sino .::= wR

l
sin SI = w a sinSI

VB = V coso .::= wR (1 + e: cos81)

I
T B

In the rotating coordinate system, the velocity of point B is W =

UB - VB or in
B
components

= URI (1- e: COSSI - e: cot B sinSI-) (4)

WB N

3 sinS )
WBT = -wRI (1 + e: cosSI + (5)
tan
l

Since the tangential velocity does not enter the control volume, it will be
neglect ed in this analysis.

From continuity,- the radial velocity must vary inversely w it h r in the

interior of the impeller. Let S' denote the angle of the fluid par ticle as it
travels along the log spiral blade relative to t he x axis in the rotating coordinate
system. The relative velocity of the f l uid is then

(1- cosS'" - cot B s inS ") (6)

r

Let 62 denote the angle of the fluid particle at the exit of the im pe l l er . Figure
2 shows that, in the rotating coordinate system, the fl u i d wh ich ent e r s t h e impeller
at angle S also leaves the impeller at angle S I - S ' Thus
l b

327
The relative velocity at the exit of the impeller is

Rl
, at e"'= 8z,

(1 -i B sina2)
WN2
=
URI Rz
cosa2 - E:cot

For flow along a blade; the relative magnitude Is g iven by

W OT
T
- [/ 1W +
w
T
12

It is a s sumed that the number of blades is infinite so the streamlines follow

the l og spir al b lade exa c tly .

tan B = =

Then the m agnitud e becomes

W O
T T
. -[ I + (tan S)
2J1f2 W
N
(7)

where W is obtained from Eq. (6). While an exact representation of the variation

of a'" w th radius would be logarithmic, the. approximation of a linear v a r ia t ion will
be u se d to facilitate the development of an analytic expression. It is

r - Rl

R2 - RI

The total relative velocity vector has the angle S tken from the t an gen t to a
ci r c l e of radius r. Thus the x and y c omponen ts of W are given by
TOT

-..
WTOT = WTOT
r in
L
(a + S) i
-..
- cos (9 + Bll ]
The general velocity expression becomes

URI _l [1 ] 'E
+ (tanBl V - E: cos
( - 2-_Rhl)
a (8)

- E: cot S sin (9 - z )I -_
R in (a + B) i
-+
cos (9 + J
Bll

328
 3. FO RCE ON THE mUR1.ING UIPELLER

The linear momentum equation for an accel e r a t ing control vo l ume [20] is

+
F
s
-
1+ + ;t
[a + 2w x v +
+
wx(w
+
x r)
+
+
+
w x
+
r) p dv
cv

lr
V pdv + V p(V
+ +
=

t cv + [
cs
dA)
-r

where

= net surface forces act ing on t he control volume

s

of the mo v ing reference frame with re sp ect to

"T
a = rectilinear acceleration
the fixed frame
angular velocity of the movi ng reference frame
+
w

V = velocity of a p art icl e in the m ov ing reference frame

w = angular acceleration of the moving r e f e re nc e frame

position of a in the moving reference frame

+
r =
partcle
For a reference frame rocating with constant angular velocity, this expression
reduces to

-+
F
s
=

jv [2w
-+
x V + w x (w x r ) ]
-+ -+ -+ -+
pdv +
jcs

-+ -+
p(v .. dA) (9)

The surface fo r c e s consist of two parts

+ -+
F
5 == F SH.A.FT + FP

where-+
FSHAFT fo r c e of shaft on control volume

F = press u re f o r c e a ct in g on control volume

p

Bernoulli's equation for a rotating reference f ra me is

CONST

The pressure force on the control volume is given by

Fp = - [ s
P dA

329
 from Bernoulli's equation

I
or,

Jrc s ' t 12
p
.F x -
l!
= - 2

The final cxpreo3ion for the force exerted by the shaft (inpeller) on the
fluid in the control volume 1s

FS HAFr =lri cv
2
+
w x V
+
+ PJlx x J
+
(w
+ +
r) dv

+
i p s
+
V V
+
dA +18 Ix 1lvldA 12

The desired f or c e here is the force e erted

negative of th e above ex p ress ion .
x by the flu.id on the shaft. It is the

. F::..." = Lp x
V : L x x dv
v p ( ) /

----
Coriolis Force Centripital Force

-1 cs
vd
. d A -
L .e.
2
liD x dA "t12 (10)

" ) .J
'-..r
\.. '-..r
Change in Linear Change in Pressure
Momentum Due to Rotation

{ P IVI2
lc s
+ dA
\., 2"
J
------_ ________J
Change in Pr essure Due
to Change in Linear
Momentum

t = (r + a cosS) (cosS .i
+
+ sine j)
+

+
w = +
wk
R
G ;Yl z G - E c o sS
'

J L
V
+
UR ri + ( tan B)
WTOT

J rt:
+ =
I

IJ
8)j1
=

- Ecot B s in S , in (S + 8)1 - cos (S +

dA = (cosS 1 + sinS 1) r d S dz

(V . dA) = WN - UR
I
:
R (1 - e: cose' - E cot B s inS ' )

330
 dv = r d 8 dr dz

Integration is taken over the control volume from Rl to R2, and from 0 to 2. On
the control surface, integration is performed both at r = R2 and at r = Rl over e
from 0 to 2n. The thickness of the impeller is assumed to he the constant h.

Fo r convenience in evaluation, the force acting on t he shaft (impeller)

due to the fluid is .evaluatedas five parts ..

(11)

After integration, they are

Coriolis Force

x ([- sin (6
b
+ B) + sinS - cotB cos (6
b
+ B) + cotB cosB)i
.....

+ [cos (Bb + B) - cosB - cotB sin (B

b
+ B) + cotB sinBl !
Centripetal Fo rce

in
Change Linear Momentum

x {- [: sin(8b - + B) i;
R-
cotB cos (8
b
+ B) + sinB

t [: cos
] !}
+ cotB cos +

- cosS + c o tB sinS

331
 Change in Pressure Due to Rotation

Change in Pressure Due to Ch a n Linear Momentum

G) = 11' awZpb R1Z (tanB)2 [1 + (tanB)2]

x
{[- Ri
8
b +
!L 8
b + i

cos cotS sin
R Rz
z

Rl
+ b +
Rl
sin 8 - cotS cos 8 cot
R2 b Rz
[-
The physical significance is indicated above each term.

The five p a r ts can be summed together yi e ld i n g the vector f o r c e acting

on the sh a f t . Dividing the for ce into components
.
g iv e s F and F A similar
xX yx
impeller eccentricity along the y axis would yield th e forces F and F What
xy yy
is of i n t e r est here is the stiffness o b ta i ned from
fF ..
--22
a

The pr inci pa l stiffness is

Kxx =

- 'ITwZpb R12 ( IT] + [] + 0 + G]+ [IJ ) (12)

where

[]J =
-2

x [-sin (8
b
+ B) + sinB + c otB cos (8b + B) + cotS cosB]

- 2 sinS ( tanS ) 2 [1 + (tanS)2)

1 1
sin (8b + B) - cotB cos (Ob + B) + sinB + cotB cos
R R

- (il2 1)
fI1 -

332
 1
=
R
cos 8
b
+ cot8 sin 8b +
The cross-coupled stiffness is

(13)

wher e

1 1/
= 2. (R - 1) tanS [1 + (tanS)2] 2
B
_

x [cos (8 + S) - cotS sin (8b + S)]

b

& = - 2 sinS (tan S)2 [1 + (tanS)2]

1
cotS sin (8 + S)]
b
_

1 1
x [- sin 8b - cotS cos8 + cotS]
R b
R

Note that both principal and cross-coupled stiffnesses are obtained. In a rotor
d y namics anal ysis , the other stiffnesses would b e given by

K K
xx
=

yy

K = -K
xy yx

For comparison with other works and onvenience in plotting results, the
stiffnesses have been made dimensionless in the following ma n ner

K
1J
.

K .
1J

The resulting ex p r es sions are

- b 1
K = -TI (& +& + +& +)
xx Rl tanZS
b 1
K
yx
= -TI
Rl t anZ S
<& +& +)
The dimenionless stiffnesses are f u n c t ions of the ratio b/Rl' the radius ra tio R,
the blade angle B, and angular sweep of the bl ade 8b , The terms R, S, and eb are

333
 related to one another so act all u y only two are independent. Also the term b/Rl
appears only as a multiplicative constant. Thus only the radius ratio R and the
blade angle a need be v aried for these stiffnesses.

4. RESULTS

Figure 4 shows a plot of the dimensionless stiffnesses K and K vs. b lade

angle B for a radius r a tio i = 1.5. p
For plotting pur ose s b/ I has b n set to
unity. It shows that the blade angle has a strong influence on both the magnitude
and sign of the stiffnesses. The principal stiffness is large and positive for
low b a d e angles.
l Pumps generally have a low blade angle so they fit in this
category. The cross-coupled stiffness is negative for this level of blade angle.
The large positive principal stiffness is generally a stabilizing effect while
the neg at ive K term is destabilizing. In the range of pump b lade angles, 10
Y
to 20 degrees, e principal stiffness term is larger than the cross-coupled
fh
t f f n es s probably leading to an overall stabilizing effect.
s i Compressors generally
have large blade angles (as defined in this paper), per h a in the range of 60 to ps
80 degrees. With a radius ratio i = 1. 5, both the p rin ipa l
c stiffness and cross
coupled stiffness are negative indicating destabilizing effects. Field results
indicate that pumps are usually stable while compressors are sometimes unstable.
v
Thus the theory de e loped here appears to ag ree, at le a s t qualitatively, with
r es u l ts from actual machines.

,A semi-empirical formulation for the cross-coupled stiffness was developed in

[18] of the form
6300 liP (Mol Wt P
o
= -
NDhp
s

where
liP = pump horsepower

Mol W = fluid m le c o ular weight

= discharge density
Po

Pg = suction density

N =
speed in RPM

D =
i p ell er diameter
m
h =
restrictive dimen sion in: flowpath

incom
While this expression was developed for compressible flow, it can be
reduced to

pressible flow. Non-dimens ionalizing this re d c ed expression gives u

K -0.6n (b/R ) n cotS Rei-I)
.yx
=

where

n = the number of blades

334
The curve for this expression was added to Figure 4 with a blade number of eight.
Another choice of number of blades will move the curve up o r down somewhat without
changing the overall shape of the curve. The plot of K from the t heo r y developed
in this paper agrees fairly well with the semi - emp i rica fX f or mula .

Figure 5 gives the results for a radius ratioR = 2.0. Again the princ ipal
stiffness is l arge for small blade angles. At larger blade a ngles, it d e c a y s
to zero but does not go negative as it did for R = 1. 5. The cross-coupled stif f
ness is negative and fairly large over the blade an gle range of 15 to 75 degrees.
The semi-empirical formula gives a somewhat larger negative c oe ff ic i ent than the
t heo ry.

One of the purposes of this study is to examine th e effects of various terms

on the stiffnesses. Table 1 gives the five terms involved in the pr inc i pal stiff
ness vs. blade angle for R = 1.5. At low blade angles, t he dominant effect is
the change in pressure due to rotation. This ter m ar ises in Bernoull i's e q u a t ion due
to the rotating coordinate system. At larger blade angle s , the dominant terms
become the linear momentum ones. Now the o the r term from Bernoulli's e q u a t i o n is
the largest but it is nearly balanced by t he c h ange in linear momentum. ! able 2
shows the three terms in the cross-coupled stiffness vs. blade an gl e for R = 1.5.
At low blade angles, the Cariolis force produces the cross-coupled stiffness e ffec t.
At higher blade angles, the liear momentum terms become dominant. Tables 3 and 4
give the numerical values for R = 2.0.

It is sometimes difficult to o b t a i n a p hys ic a l feel for the r e su l t s of nn

analysis when dimensionless parameters are used. In an effort to impart a better
idea of the numerical values devel o p e d from the theory here, some sample impellers
were chosen. T he results are g i ven in Table 5. Generally small pumps will have
very small stiffnesses while large pumps such as b o i le r feed p umps (,3n have [ai.rly
large stiffness acting on them. The stiffness value here K = 124,000 lbf/in is'
the same order of magnitude expected from the seals in a pu p . A small compressor.
even at high speed, produces a relatively small c r oss - c ou p lin g stiffness. Large
compressors have somewhat larger coefficients. Oft en suc h a com p re ss or will have
eight or ten stages with a fairly flexible shaft. The cumulative e ff e ct can produce
an instabilit y.

5. CONCLUSIONS

The theoretical solution for a simple impeller model shows the principal and
cross-coupled stiffnesses to be of about the same order of magnitude. For blade
angles (S) less than thirty degrees, common for water pumps. the flow seems to
provide tabilization for the shaft, while for the larger blade angles, common
for compressors, the impeller is generally destabilizing. Since the magnitude of
Kxx and Kyx are nearly the same, it would be important to include both terms in
any rotor dynamics analysis of the shaft, rather than incorporating just the cross
coupling terms

. Though this work does not fully resolve the nature of aerodynamic forces on
centrifugal machines, it does offer more understanding of these forces. An important
element of any future work, however, is the availability of experimental data for
verification. As of the present, no such data exists, although efforts are being
perused in this area.

335
 6. REFERENCES

1. Sh e pherd , D. G., Pr inciples of Turbomachinery, MacMillan Publ ishing Co., New

York, 1956.

2. Ac o s ta , A. J., "An Experimental and Theoretical I nv estigat ion of Two-Dimensional

Centrifugal Pump Imp elle rs, " Transactions of the ASME, July, 1954, p.; 749-763.

3. Balje, O. E., "A Flow Model for Centr if u gal Comp r e s sor Rot ors, " Transactions of
the ASME, Vo l . 100, Jan., 1978, p. 148-158.

4. Howard , J. H. G. and Lennemann, E.,"Measured and Predicted Secondary Flows in a

Centrifugal Impeller," Journal of Engineering f or P o wer, Trans. ASME, Jan.,

.
1971, p. 126-131.

5. Lennemann, E. and Howard, J. H. G., "Unsteady Flow Phenomena in Rotating Centrif

u gal Impeller Pas s a ges, " Journal of Engineering for Power, Trans. ASHE, Jan.
1970, p. 65-72.

6. Moore, J., "A Wa ke and an Eddy in a Rotating Radial Flow Passage," Journal of
.Engineering for Power , Trans. ASME, July 1973, p. 205-219.

7. Alford , J. S., " Protecting Turbo machinery from Self-Excited Rotor Whirl , " Journal
of E ng ineerin g for Power, Trans. ASHE, Oct . 1965, p. 3 3 3 - 34 4 .

8. Black, H. F., " La teral Stability and Vi b ra ti o n of Hi gh Spe ed Centrifugal Pump

Ro tor s , " IUTAM Dyn amics of Rotors Symposium, Lyngby, Denmark, Aug. 1974, p. 56-
74.

9. Barrett, L. E. and Gunter, E. J., "Stabilization of Aerodynamically Exicted T ur

bomachinery with Hydrodynamic Journal Beari ngs and Supports," P roceed in gs of

NASA/ARO Workshop on Ro tordyn ami c Instability Problems in H igh Performance

Turbomach inery , College Station, Texas, May 1980.

10. Lund, J. W., " Some Unstable Whirl Phenomena in Rotating Machinery," The Shock
and V ib r a t ion Digest, Vol. 7, No. 6, 'June 1 975, p. 5 - 1 2 .

11. Gunter, E. J., Barrett, L. E., and Al la ire , P. E., " S t ab i l iz ation of Turbomachin
ery with Squeeze Film Dampers - Theory and Application," 1. Mech. E., Pro
ceedings of Conference on Vibrations in Rotating Machinery, Cambridge, England,
C233/76, Sept. 1976.

12. "Effect of Fluid Forces on Rotor Stabi l i ty of Centrifugal

Co ld ing-Jor gens en, - J.,
Compressors and Pumps, " Proceedings of NASA/ARO Workshop on Rotordynamic Insta
bility Prob l ems in High-Performan ce Turboma chin ery, Col le ge St at io n, T e x a s ,
May 1980.

13. Shoji, H. and Ohashi, H., " Flu i d Forces on Rot at ing Centrif u gal Impeller with
Whirling Motion," Proceedings of NASA/ARO Workshop on Rotordynamic Instability
Problems in Hi g h- Perf orman ce Turbomachinery, Coll e g e Station, Texas, }my 1980.

336
14. Thompson, W. E., "Vibration Exciting Mechanisms Induced by Flow in Turbomachine
Stages." Proceedings of NASA / ARO Workshop on Rotordynamic Instability Problems
in High-Speed Turbomachinery, College Station, Texas, May 1980.

15. Thompson. W. E., "Fluid Dynamic Excitation of Centrifugal Compressor Rotor

Vibrations," Journal of Fluids Engineering, Vol. 100, March 1978, p. 73-78.

16. Chamieh, D. and Acosta, A. J., "Dynamic Forces on a Whirling Centrifugal Rotor,"
Proc. 6th Conference on Fluid Machinery, Akademiai Kiado, Budapest, Hungary,
1979.

17. Jenny, R. and Wyssmann, H. R., "Lateral Vibration Reduction in High Pressure
Centrifugal Compressors," Proceedings of the Ninth Turbomachinery Symposium,
Texas A & M University, Dec. 1980, p. 45-56.

18. Wachel, J. C. and von Nimitz, W. W., "Assuring the Reliability of Offshore Gas
and Compression Systems," Proceedings of the European Offshore Petroleum
Conference Vol. I, EUR205, October, 1981, pp. 559-568.

19. Branagan, L. A., "Aerodynamic Forces on an Unbound Centrifugal Impeller:

Effects of Change in Fluid Momentum," M.S. Thesis, University of Virginia,
January, 1982.

20. Fox, R. W. and McDonald, A. T., Introduction to Fluid Mechanics, John Wiley &
Sons, New York, 1978, p. 166.

337
 T ABLE 1. - EFFECT OF VARIOUS TERMS ON PRINCIPAL STIFFNESS FOR R 1.5

Blade Corio lis Centripetal Change In Change In Change In Principal

Angle Forc e Force Linear Pressure Due Pressure Due Stiffness
Momentum To Rotation To Change In
Linear Momentum

B [II IT] IT! m ill K

xx

10.00 -.75 .63 -.09 -1.25 .03 144.75

15.00 -.67 .63 -. 15 -1.25 .OS 59.6;
20.00 -.57 . 63 -.21 I -1.25 .15 29.76
I
25.00 -.50 .63 -.30 -1.25 .26 16.74
30.00 -.45 .63 -.44 -1.25 .44 10.0 5
35.00 -.42 . 63 -.6 5 -1.25 .73 6.15
40.00 -. 40 .63 -.98 -1.25 1.20 3.6 2
45 . 00 2.00 1.80
-.40 .63 -1.55 -1.25
50.00 -.41 .63 -2.55 -1.25 3.44 .33
55.00 -.43 .63 -4.46 -1.25 6.20 -1. 05
60 .00 -.47 .63 -S.44 -1. 25 12.00 -2.59
65.00 -.53 .6 3 -1778 -1 . 2 5 25.75 -4.66
I
70.00 - .63 .63 - 4 3 . 96 -1.25 64.53 -S.04
75.00 -.SI .63 -140.25 ! -1.25 207 . 9 2 -14.94
-34.20
SO. OO -LIS .63 -714.71 ! -1.25 1066.62
I

TABLE 2. - EFFECT OF VARIOUS TERMS ON CROSS-COUPLED STIFFNESS FOR R 1.5

Blade Coriolis Change In Change In Cross- Cross-

Angle Force Linear Pressure Due Coupled Coupled
Momentum To Change In Stiffness Stiffness
Linear Momentum

a & & &. K

xy
Wachel [18]

10.00 .33 .0 3 .06 -42.96 -6 4 . 1 4

15.00 .7 1 .10 .09 -39.32 -42.21
20.00 ;91 .18 .12 -28.73 -31. 07
25.00 1. 07 .27 .15 -21.52 -24.25
30.00 1. 23 .3 S .1S -16.S4 -19.59
35.00 1.41 .53 .19 -13.69 -16.15
40.00 1.64 .74 .19 -11.46 -13.4S
45.00 1 . 95 LOS .11 -9.7S -11.31
50.00 2.37 1.53 -.ll -S.40 -9.49
55.00 3.00 2.32 -. 7 0 -7.10 -7.92
60.00 3 .96 3.71 -2.3 1 -5.62 -6.53
65.00 5.57 6. 4 5 -6 . 9 3 -3.47 -5.27
70.00 8.52 12 . 65 -22.29 .47 -4.12
75.00 14 .9 0 30.06 -8 7.7 9 9.66 -3.03
80.00 33.13 101.59 -533.9S 39.00 -1. 99

338
 TABLE 3. - EFFECT OF VARIOUS TERMS ON PRINCIPAL STIFFNESS FOR R 2.0

Blade CorioUs Centripetal Change In Change In Change In Principal

Angle Force Force Linear Pressure Due Pressure Due Stiffness
Momentum To Rotating To Change in
Linear Momentum

a ill [l] ill ill I1J K

xx

10.00 -.89 1.50 -.09 -3.00 .03 247.42

15.00 -1.53 1.50 -.22 3.00 .08
-
138.95
20.00 -1.58 1. 50 -.35 -3.00 .15
, 77.74
25.00 -1.50 1.50 -.51 -3.00 .26 46.85
30.00 -1.42 1.50 -.73 -3.00 .44 30.17
35.00 -1.36 1.50 '-1.06 -3.00 .73 20.43
40.00 -1.33 1.50 -1.59 -3.00 1.20 14.35
45.00 -1.33 1.50 -2.46 -3.00 2.00 10.35
50.00 -1.37 1.50 -4.00 -3.00 3.44 7.60
55.00 -1.45 1.5 0 - 6.9 2 -3.00 6.20 5 .6 4
60.00 -1.5 8 1.50 -12.95 -1.00 1200 4.22
65 .00 -1.79 1. .5 0 -27.08 -3.00 25.75 3.16
70.00 -2.15 1.50 -66.57 -3.00 64.5 3 2.37
75.00 -2.76 1.50 -211.50 3.0 0
- 207.92 1.77
80.00 -4.05 1. 50 -1074.58 -3 . 00 1066:62 1.32

TABLE 4. - EFFECT OF VARIOUS TERMS ON CROSS-COUPLED STIFFNESS FOR R = 2.0

Blade CorioUs Change in Change In Cross- Cross-

Angle Force Linear Pressure Due Coupled Coupled
Momentum T o Change In Stiffness Stiffness
Linear Momentum

a & & & K

xy
Wache1 [18]

10.00 -.37 -.02 .09 30.89 -171.04

15.00 .44 .0 4 .14 -26.93 -112.56
20.00 1.12 . 14 .19 -34.58 -82.86
2 5 .00 1.63 .26 .25 -31.07 -64.68
30.00 2.07 .4 1 .33 -26.49 -52.24
35.00 2.52 .61 .41 -22.65 -43.07
40.00 3.03 .88 .50 -19.69 -35.94
45.00 3.69 1.28 .59 -17.44 -30.16
50.00 4.57 1.89 .64 -15.71 -25.31
55.00 5.84 2.89 .56 -14,31 -21.12
60.00 7.79 4.68 .00 -13.05 -17.41
65.00 11.00 8.18 -2.20 -11.60 -14.06
70.00 16.92 16.11 -10.85 -9.23 -10.98
75.00 29.68 38.40 -51.91 -3.65 -8.08
80.00 66.16 130.04 -353.46 15.36 -5.32

339
 TABLE 5. - NUMERICAL EXAMPLES FOR PUMP AND COMPRESSOR IMPELLERS

R K K
-E....- --!!!.... b B xx
Impeller
slug/ft3 rpm in in degrees 1bf/1n 1bf/in

Small Pump 1.94 600 3.0 0.5 15 16.5 -3

Large Pump 1.94 10,000 9:0 1.5 15 124,000 -24,000

Small Compressor 7.5 x 10-3 30 000 3.0 0.5 60 205 -702

,

Large Compressor 7.5 x 10-3 10,000 12.0 0.5 60 720 -2,200

Figure 1. - Geometry of perturbed impeller.

340
 y

I
I
I
I
I
,PATH
I
I
I
/" --
.
+-=:::.::"r" _ - -"9INITIAL _
..... -I----'I!l_
x

Figure 2. - Fluid path through impeller over passage time (T). Fixed coordinate
frame [ap = lilT - ab].

ECCENTRIC
/IMPELLER

CENTERED
IMPELLER

o A x

(a). Absolute velocity components at point B.

Figure 3. - Velocity components at impeller inlet.

341
 VSt
ECCENTRIC
IMPELLER

/
CENTERED
IMPELLER

o A

(b). Relative velocity between fluid and impeller

. (c). Relative velocity components in rotating coordinate system.

Figure'3. - Concluded.

342
 100

100
R = 2.0
80

80
R ,= 1.5
60

60

40

40 fII
fII
III
Z
fII Il. 20
fII
III !!:
Z I-
Il. 20 fII
!!:
I-
fII 0
BLADE
ANGLE
0 .. ;;0""
BLADE '" (DEGREES)
'"
ANGLE '"
-Kyx -20
.... (DEGREES) '"
,
.......
'"
-20 "/ ,
/ ,
/.: /.
-40
I -Kyx /-Kyx
- 40 WACHEL [18] I
(8 BLADES)
I WACH E L [18]
-60
I (8 BLADES)
I
,
-60

Figure 4. - Dimensionless stiffness Figure 5. - Dimensionless stiffness

versus blade angle for R = 1.5. versus b l ade angle for R = 2.0.

343
 UNSTFADY FLaV PfJFNJVlENA IN INIXJS1RIAL CENlRIFlDAL aI\1PRESSCR STPDE

L. Bonciani, L. Terrinoni, and A. Tesei

Nuovo Pignone
50127 Florence, Italy

SUMMARY

It is well known that rotating non-uniform'flow pattern has a strong influence on high
pressure centrifugal compressor vibrations. This paper shows the results of an

experimental investigation on a typical centrifugal compressor stage running on

an atmospheric pressure test rig. Unsteady flow was invariably observed at low

flow well before surge. In order to determine the influence of the statoric

components, the same impeller was repeatedly tested with the same vaneless diffuser,

but varying return channel geometry. Experimental results show the strong effect

exerted by the return channel, both on onset and on the be.hav ior of unsteady flow.

Observed phenomena have been found to confirm well the observed dynamic behavior

of full load tested machines when gas density is high enough to cause appreciable

mechanical vibrations. Therefore, testing of single stages at atmospheric pressure

may provide a fairly accurate prediction of this kind of aerodynamic excitation.

INTRODUCTION

Problems which have arisen in past years with high density centrifugal compressors

have considerably increased the demand for full load testing (ASME PTC 10, Class 1),
mainly aimed at checking the stability of machines under aero-induced excitation

forces.

Present experience leaves no doubts that aerodynamic excitation is always present

in centrifugal compressors and it is obvious that the forces involved become more

marked the greater gas density and machine speed. This is the reason why machines

subject to aero-induced vibrations have been found more frequently in natural gas

injection or urea synthesis plants than in ammonia synthesis or refinery compressors,

even when running at the same pressure levels (ref. 1). Although the action of

aerodynamic forces involved leads to subsynchronous vibrations in general, two

completely different behaviors have been encountered: self-excited vibrations and

forced vib;-ations. Vibrations belonging to the first category (self-excited

vibrations) have been noted in high performance turbo-machinery {ref. 2)including

high pressure compressors (ref. 3 and 4), apparently with greater frequency in

back-to-back versions of the latter. The vibration frequency observed is normally

the :fundamental bending natural frequency of the rotor. An explanation commonly

given for this phenomenon is based on destabilizing forces caused by the labyrinth

seals (ref. 5 and 6). On the other hand vibrations belonging to the second category

344
 ( forced vibrations ) have been recoraeu in all high density compressors. They have

the following typical behavior ( ref. 7 and 8):

They appear relatively near to surge and are very stable in amplitude.

Asynchronous frequency is very low ( order of magnitude 10% of RP M )

Asynchronous amplitude depends on tip speed and gas density.

In order to examine this phenomenon more closely a test program was set up with

the aim of investigating the amplitude and frequency of pressure oscillations

within a centrifugal compressor stage. A standardized low specific speed stage,

which is normally utilized in high pressure applications, was slated for testing.

SYMBOLS

R Radius

p static pressure

total pressure

C radial velocity
r

A pressure oscillations amplitude

f fundamental frequency of pressure oscillations

s

.n impeller angular velocity

amplitude of pressure oscillations normalized to stage total inlet

pressure
'"

f fundamental frequency of pressure oscillations normalized to impeller

s
speed (f = 2 'IT f /.n
s s

Mu tip speed Mach number

Reynolds number ( based on diffuser axial width )

inlet flow coefficient

'" '"

C normalized radial velocity (C = C /D R)

r r r

absolute flow angle ( referred to tangential direction )

pressure recovery coefficient

Subscripts:

10 = measured at section 10

10 =
" " "
10'

20 =
" " "
20

345
20' =
" " "
20'

30' 30'
" " "
=

40 40
" " "
=

40' =
" " "
40'

50' =
" " "
50'

60 " " "

60

60' " " "

60'

D diffuser

RC return channel

* = onset of unsteady flow

TEST FACILITIES AND _INSTRUMENTATION

Testshave been arranged on one of the three test rigs available for the individual
stages development (ref. 9). A cross section of the test rig is shown on Fig. 1;

the stage consisted of an impeller, a free vortex diffuser, a cross-over and a

return channel. The gas utilized was air at atmospheric pressure in an open loop

circuit. Speed was adjusted by an hydraulic coupling torque converter while an

electrically actuated discharge valve was operated to vary the pressure ratio.

Table 1 shows the conventional instrumentation used for industrial stage testing.
Table 2 shows the instrumentation utilized to detect pressure oscillations

connected with non-stationary flow conditions. The following should be mentioned

of the conventional instrumentation:

The data acquisition system is based on a Solartron system 35 with a PDP 11/03
control unit

All pressure readings are connected, through a scannivalve, to a single pressure

transducer. The transfer function of the measurement chain has been experimentally
tested, to check that the output is the time average of the pressure within the
frequency range of interest.

Regarding non-stationary reading we can note that:

static pressure probes are the Kulite XT-190-50 type

total pressure probes are the Kulite XB-093-50G type

probe signals have been recorded on an Ampex PR 2200 tape recorder and finally
analized through an Ono-Sokki CF-500 real time spectrum analizer. Data shown

for each tested point are the RMS averages of 256 spectra.

Two different stage configurations have been tested: configuration A and cohfigurat
ion B. Both configurations utilize the same impeller and diffuser but different

return channels and, of course, cross-overs.

346
 TEST RESULTS

Configuration A

It is necessary to point out that the following procedure was adopted to obtain

the data relating to configuration A.

1. Readings were taken of time "average" quantities (i.e. pressures, temperatures

etc. at the different measuring sections) with conventional instrumentation.

2. Next, readings were taken of pressure oscillations (remeasuring inlet pressure,

flow and RPM) independently of prior readings.

3. Finally some "points" were repeated in regions of particular interest.

Qualitative Description

When flow is reduced at constant speed the following behavior was observed:

Pressure oscillations started simultaneously on all pressure transducers with

very low amplitude and almost sinusoidal shape.

A small flow reduction resulted in a slight increase in frequency and a

considerable increase in amplitude while the signal shape remained sinusoidal.

If flow was further reduced the phenomena observed was dependent on the tip

speed Mach numbers (Mu) as follows:

At Mu = 0.45 and 0.60 we noted:

The shape of the signal suddenly changed from a sinusoid with a frequency of

f to a signal having frequencies f and f /2 with comparable amplitudes. The

s s s
onset of the second frequency
. (Le. f /2) is usually very sudden.
s

When flow was reduced the fs and fs/2 frequencies slowly increased and the f
s

amplitude component remained almost constant while the f s/2 one rose gradually.

On rethrottling, the fs/2 signal finally disappeared and the f signal had a
s

sudden increase in amplitude.

Further flow reductions caused a shifting of the f frequency while amplitude

s

remained almost constant till full surge.

On opening the valve and exploring the phenomenon starting from full surge we

found that frequencies and amplitudes were repetitive related to flow. A slight

hysteresis was noted in the onset region.

At Mu = 0.75 and 0.85 we observed that:

Pressure oscillations might have a sinusoidal shape with f frequency till full
s

surge.

347
 With the same inlet flow coefficient it was possibJe to shift from a single
frequency signal (fs) to a dual-frequency signal having the same fs frequency
and an additional component at fs/2 frequency. Each one of the amplitudes
(i.e. at fs and fs/2) with dual frequency was lower than the amplitude with
single frequency.

Although the factors governing the shift from one shape to the other were not
discovered, the following observations were made:

a. The shift from a single frequency shape to the other seemed easier when
increasing the flow from the surge.

b. Sometimes introduction of the conventional instrumentation probes into the

diffuser was very effective in triggering the shift to the dual-frequency shape.

c. Mantaining a constant opening of the discharge valve and raising the speed from
low Mu (i.e. Mu = 0.45) with dual frequency shape the signal might retain the
same shape even if Mu exceeds 0.85.

d. Cases were observed where shifting appeared several minutes after the last
positioning of the valve.

Summarizing we believe that the phenomenon can be described as follows:

The unsteady flow pattern may exhibit two distinct shapes (single frequency
and dual frequency) .

The dual frequency shape is stable within a range of flows-RPM (or better
- Mu) and metastable in the remaining range.

When dual frequency exists is may survive in the entire speed range explored.

It will later be seen that the two different signal shape correspond to two
distinct values of the return channel recovery coefficient, based on time average.

Quantitative Description

To keep the length of the paper within reasonable limits we will indicate only the
results obtained at section 20', behavior at all the other measuring sections
being very similar.

Fig. 2 shows the frequencies normalized to the RPM while fig. 3 and 4 show the
amplitudes of the pressure oscillations corresponding to fs and fs/2 normalized to
the suction pressure. The following can be remarked:

Amplitudes at section 10' (impeller suction) were always negligible till surging.

At the different measuring sections pressure oscillations amplitudes exhibited

shapes similar to those shown for section 20' . The following are the average
values found:

348
 at f (with both single and dual
s.
frequency shape)

A 80% A A 81 % A
30, 20, 30' 20'

A A 60% A
40'
60% A20' 40' 20'

A 60% A , A 17% A
50' 20 50' 20,

A 9% A A ::: 9% A
60' 20, 60' 20'

Irrespectively of the flow coefficient, the phase difference between static probes

at the same radius is 900 for the f /2 component and 1800 for the fs component
s
with dual frequency signal shape and is 1800 for the fs component when the signal

is single-frequency. Therefore it seems that f /2 is associated with a single

s
stall cel while fs is associated with two stall cells

The phase difference between static probes at the same angular position was

practically independent from the flow coefficient. Typical values measured (in

round figures) were: -100, between sec. 20' and sec. 30', and -200, between

sec. 20' and sec. 40', for the f component (with both single or dual frequency
s
shape) and always near 00 for the f /2 component.
s

Analysis of Time Averaged Data

Fig. 5, 6'and 7 show the pressure ratios (static to total and total to total)

versus flow at three different measuring sections. An appreciable variation in

the slope can be noted at the onset of the pressure oscillations followed by a

region with a positive slope. At Mu = 0.75 and 0.85 two distinct branches clearly

identify the working regions relevant to the two shapes of the signal previously

described.

The lack of connection between "average" measurements and instantaneous measurements

initially led to considerable confusion. Only later was it realized that the two

disturbance shapes have distinct "average" measurements and that the shift from

one shape to the other cannot always be triggered at will in the Mu = 0.75+0.85

area. As a result the graphs in fig. 7 and fig. 3 are slightly contradictory:

the shift point from one shape to the other at Mu = 0.75 and Mu = 0.85, shown in

fig. 3, does not coincide with those indicated in fig. 7. The general shape of

the amplitude curve at Mu = 0.75 looks similar to those at Mu = 0.45 and Mu = 0.60

but fig. 7 suggests that this was not always the case.

The flow coefficients corresponding to the onset of the pressure oscillations are

related to tip speed Mach numbers. However, if the inlet flow coefficient is
,.,

plotted versus any of the impeller exit variables, for example Cr20, as shown

349
on fig. 8, one can note that the onset starts at an almost constant value of such
....
variables. Fig. 9 shows both 0<2 and C(40 versus C . It can be noted that
0 r20
the unsteady flow condition invariably appears when 0< 20 .. 100 and 0(40 ... 130

The behavior of the diffuser pressure recovery coefficient, see fig. 10, shows

considerable scattering corresponding to the 0<

20* onset value. Fig. 11 shows

that the return channel pressure recovery coefficient exhibits considerable

discontinuity in the onset regions. It is of particular interest to note that

the CpRC vs. 0< curve is split into two different branches which correspond
40
to dual frequency and single frequency respectively.

Bearing in mind, that pressure oscillations at impeller suction were always

negligible and considering the behavior of pressure recovery coefficients, it seems

that the phenomenon had its origin from the statoric components. Three different

hypothesis can be formulated to explain the onset mechanism:

a. - the phenomenon pertains to the diffuser "in itself" and starts when a

"critical" inlet angle is reached.

b. - the phenomenon pertains to the return channel "in itself" and starts when a

"critical" incidence angle is reached.

c. - interaction between the vaneless diffuser and return c hannel is not negligible,
therefore a "critical value" of some exit variable exists which cannot be

clearly ascribed to each one individually.

It is clear that these hypothesis.as they have been formulated,are somewhat

simplified and that they ignore the fact that the inlet conditions of the stationary
'"
componeIi.ts for a constant value of Cr20, are not strictly similar at different Mu.

For instance:

- "average" velocity profiles are not exactly similar

- the diffuser inlet absolute Mach number increases from 0.25 at Mu = 0.45 to 0.43

at Mu = 0.85

the diffuser inlet Reynolds number varies from Re 0.45 to

Re = 160,000 at Mu = 0.85

However these variations are moderate and fig. 9, fig. 10 and fig. 11 suggest that

the influence of these factors is limited within the range explored throughout

testing.

Let us assume as a working hypothesis that the phenomenon is mainly influenced by

the return channel and starts when a "critical" incidence angle at the leading edge

of the return channel blades is reached. In the light o f this hypothesis a new

return channel was built with identically shaped blades, but a different axial

width, to move the onset close to the design flow of the stage.

350
 Configuration B

In order to avoid some confusion experienced when testing configuration A, both time

averages and instantaneous measurements were acquired contemporaneously, moreover

the tested points were concentrated in the unsteady flow area.

The phenomenon is easier to describe since it is characterized by a single-frequency

shape within the tested range of Mu. The onset as well as the growth of the pressure

oscillations are similar to those tested on configuration A when the single

frequency shape was present. Fig. 12 and fig. 13 show respectively the normalized

frequency and the pressure oscillations amplitudes at sec. 20' versus the inlet flow

coefficient. The following can be noted:

Amplitudes at sect. 10' are always negligible till full surge.

At the different measuring sections pressure oscillations amplitudes are

qualitatively similar to those of section 20'. The following are the average
values found:

A 78% A
30' 20'
A 60% A
40' 20'
A 56 % A
50' 20'
A 14% A
60' 20'

Irrespectively of the flow coefficient, the phase difference between static

probes at the same radius is 180 thus indicating a two stall cells configuration.

The phase difference between static probes at the same angular position was

slightly dependent on the flow coefficient. Starting from the onset and reaching

the full surge, the phase difference gradually changed ( in round figures ) from

0 to -10, between sec. 20', and 30', and from 0 to -20 between sec. 20' and

40', being the phase difference ratio almost constant and near to two. As a

consequence, loci of maximum and minimum pressure amplitudes are not radial

lines but curves shifted slightly backwards as referred to the direction of

rotation.

Analysis of the Time Averaged Data

Fig. 14, 15 and. 16 show the pressure ratios ( static to total and total to total)

versus flow at three different measuring sections. It can be noted that the onset

has been shifted to a considerably higher flow as indicated by the hypothesis

formulated. As already noted with configuration A, the onset of pressure oscillat

ions starts at an almost constant value of the impeller discharge parameters ( see
fig. 17 and fig. 18). In this case we had ()( 18 and 14
40", C(20. .

351
 It can be verified that the onset takes place when there is a practically constant

incidence angle at return channel blades while flow angles along the diffuser are

considerably different from those of the configuration A. Fig. 20 shows that the

pressure recovery coefficient of the return channel again exhibits a sudden drop at

the onset of unsteady flows. The CpD too curves downwards, fig. 19, however the

drop is less evident, whereas scattering is greatly reduced if compared with

configuration A.

TESTS SUMMARY AND CONCLUDING REMARKS

Two different stage configurations were tested to study the behavior of unsteady

flows. Configuration A and configuration B had identical impellers and diffusers

but different return channels and cross-overs. The following can be concluded on

the basis of tests performed:

1. The two configurations tested clearly showed that unsteady flow is caused by

the statoric components.

2. With configuration B unsteady flows begin at much higher. flow coefficient than

those of configuration A.

3. The onset of pressure oscillations takes place with considerably different

diffuser flow angles for the two configurations, while the incidence angle

at the return channel blades is almost constant.

Therefore it seems that the return channel blades played the most important role

in determining both the onset and the growth of the phenomenon.

To conclude, some additional remarks are necessary to give an indication of the

general validity of obtained results and some comparison with the behavior of

complete machines when gas density is high enough to reveal pressure oscillations

in the form of shaft vibrations.

Tests performed on completely different stages (i. e. several standard stages for

average specific speed and one 3-D type typical for pipeline applications)

exhibited unsteady flows with frequency-amplitude behavior similar to that of

the configurations described in this paper.

The frequencies of forced asynchronous vibrations, detected in full load testing

of high density compressors (ref. 7 and 8) showed a very good correspondence to

the frequencies of pressure oscillations of non stationary flows investigated

on a single stage test rig.

Reviewing current literature on this subject the following an be remarked:

a. The tests referred to with ref. 10, 11, 12 and 13 definitely show that a

vaneless radial diffuser may generate self-excited pressure oscillations.

The present paper suggests that such data should be used with some caution

for an industrial centrifugal stage, having a return channel.

352
 b. Test results of the present paper agree to a great extent with ref. 14

and with some of the data published in ref. 15, both based on testing of

industrial centrifugal stages with return channels.

REFERENCES

1. Ferrara P.L., Tesei A.: High Pressure Centrifugal Compressors. Inst. Mech. Eng.

Conference Publications 1978-3 March 1978

2. EK M.C.: Solution of the Sub synchronous Whirl Problem in the High Pressure

Hydrogen Turbomachinery of the Space Shuttle Main Engine. AIAA/SAE 14-th

Joint Propulsion Conference. Las Vegas,Nev.-July 2-27, 1978.

3. Geary C.H., Damratowsky L.P., Seyer C.: Design and Operation of the World's

Highest Pressure Gas Injection Centrifugal Compressor paper No. O.T.C. 2485

presented at the Offshore Technology Conference, Houston Texas, May 1976.

4. Coletti N.J., Crane M.E.: Centrifugal Compression on the Arun High Pressure

Injection Project. Inst. mech. Eng. Conference Publications 1981-3 March 1981.

5. Wright D.V.: Air Model Tests of Labyrinth Seal Forces on a Whirling Rotor.

ASME Book Editor W.G. Steltz 1977.

6. Benckert H., Wachter J.: Flow Induced Spring Coefficients of Labyrinth Seals

for Application in Rotor Dynamics. NASA Conference Publication 2133-Aug. 1980.

7. Bonciani L., Ferrara P.L., Timori A.: Aero-Induced Vibration in Centrifugal

Compressors. NASA Conference Publication 2133-August 1980.

8. Sabella D., Terrinoni L., Timori A.: Full Load Testing of Centrifugal Natural

Gas Injection Compressors. Inst. Mech. Eng. Conference Publication 1981-3 March

1981.

9. Benvenuti E.: Aerodynamic Development of Stages for Industrial Centrifugal

Compressors, Part. 1 & 2 - ASME Paper No. 78-GT4 & 5.

10. Jansen W.: Rotating Stall in a Radial Vaneless Diffuser. Transactions of the

ASME-Journal of Basic Engineering - Dec. 1964 p.p. 750+758.

11. Ab0elhamid A.N., Bertrand J.: Distinctions Between Two Types of Self-excited

Gas Oscillations in Vaneless Radial Diffusers - ASME Paper No. 79-GT-58.

12. Abdelhamid A.N.: Analysis of Rotating Stall in Vaneless Diffusers of Centrifugal

Compressors. ASME Paper No. 80-GT-184.

13. Abdelhamid A.N.: Effects of Vaneless Diffuser Geometry on Flow Instability in

Centrifugal Compression Systems. ASME Paper No. 81-GT-I0.

14. Abdelhamid A.N., Colwill W.H., Barrows J.F.: Experimental Investigation of

Unsteady Flow Phenomena in Vaneless Radial Diffusers. ASME Paper No. 78-GT-23.

15. Van Den Braembussche R.A., Frigne P., Roustan M.: Rotating non Uniform Flow in

Radial Compressors. AGARD C.P. 282-May 1980

353
 TABLE 1.

Total pressure Static pressure

Measuring sect. Thermoel.
probes probes

Sec. 10 4 Kiel 1 cobra 1 8 (circum. and

(impeller inlet) radially spaced

Sec. 20 1 cobra 1 2
(diffuser inlet)

Sec. 40 1 cobra 4
(diffuser exit)

Sec. 60 4 Kiel 8 8 (circum. and

(return channel (4+4 at inner and radially spaced
exit) over radius)

TABLE 2.

L
Total pressure Static pressure
Measuring sect.
probes probes

Sec. 10' 1 1
(Impeller inlet)

Sec. 20' 1 2
(Diffuser inlet) (90 spaced)

Sec. 30' 1 2
(Diffuser midspan) (90 spaced)

Sec. 40' 2
(Diffuser exit) (90 spaced)

Sec. 50' 2
(Return channel throat (90 spaced. Throat area

area) of two sections of the

return channel)

Sec. 60' 1
(Return channel exit) (At the exit of one of the
two sections of sec. 50)

354
 =-+-+GIII
@--

--.-- -+--- , ,

Figure 1. - Cross section of the test rig.

15.0

f.tol
130

11.0

rt; .45 .60 .75 .85

90

TtlROTTL ING .. CD I!I
THE VALVE

70 OPENING ..
THE VALVE

5.0
70 80 90 100 110 120 130 1 40 ISO fJJ 160

Figure 2. - Normalized observed f requency vs. inlet flow coefficient

(configuration A).

355
 520

AId.1(f
4'0 f'\J .45 .60 .75 .85
l'HROTTL 1110 6 " " III
THE VALVE
4'0
OPENINO A ..
THE .AL'E

400

I
/ 320
I
380
ald,10'
2.0

320
f'\J .45 .60 .75 .85
2'0 THROTTLING
THE YALVE
" '"
2'0

W
V1
I
I
OPENING
..
200 THE VALVE
0\
2.0 +
I
I
I
160

200 /l ...... ....

_

I ', \
f20 I/,,11--',
I "
,60 f ,

ih '
10 \
,

i
'20
I '-...\ "
I
I
I .0
t, ,
........" tI
'0 t

i/
I
I
I
,
40 I 70 80 to 100 110 '20 130 140 'so f/J 160
I
I
I
\
70 10 to ,00 110 '20 ,30 140 'so f/J "0

Figure 3. - Static pressure oscillation reduced Figure 4. - Static pressure oscillation reduced
amplitude (@ fs) at section 20 vs. inlet flow amplitude (@ fs/2) at section 20 vs. inlet flow
coefficient (configuration A). coefficient (configuration A).
 Itu I't,. "1Po.. Itu Pt.f'o.. Pif'o

'1
-----"'- .. .. .45 .. ..
,5
,60 .60

.75 .. ,., .75 . ..

.as II ......----'--- ,as II

"'-T----- ... ,
"
.... --.. -

_ ..
--

--lie --

. .....
.... -

"'--&--------

-A-.
w
VI
....,

,.05 1 ' .... 1

,.. ,..
i i i , , , i i i ' i i
10 10 100 120 140 110 110 200 220 240 cp ISO alo to '00 12. , .. 200 22. 2'. IJI 21 ...

Figure 5. - Static to total and total to total Figure 6. - Static to total and total to total
pressure ratios at section 20 vs. inlet flow pressure ratios at section 40 vs. inlet flow
coefficient at different Mach numbers coefficient at different Mach numbers
(configuration A). (configuration A).
 170 Nu Pelflolo P/POlO

.5
.60
.75 1!1 2410
. 85 1!1

"
26&

"'J
240
I
,
I
).51'!." ,
SINGLE J 220
FREQUENCY /
I
I
I
I 200
,

I
I
180
w I
I
V1 I
00
..
. .
.: '
.0
'
1 0
6
,
I
I
,
""


;{
;{ 120

//// r-
? Mu

i
;{
100 III 'l: I . .45
60
(!) .75
l!l .85
10

I.Dsn l , ' " 60

e,':"tO
, , , , , ,
I
60 80 100 120 140 160 180 200 220 240 rp 260 250 oo:so 0.0$0 0.070 o.oo O! 10 O 130 O ISO o 170

Figure 7. - Static to total and total to total Figure 8. - Flow coefficient versus nondimensional
pressure ratios at section 60 vs. inlet flow radial speed at section 20, at different Mach
coefficients at different Mach numbers numbers (configuration A).
(configuration A).
 31

a (0)
j ..
Mu
.45
.60
28
I!l .75
l!I .85

25

22

It

1&

13

10

0130 OISO 0170 e,':I90

F igure 9. - Absolute flow angles at section 20 and section 40 vs. nondimensional

radial speed at section 20 (configuration A)

. , ..
c". Mu
A .45
o '00 .60
.75
.85

Figure 10. - Vaneless diffuser pressure recovery coefficient vs. absolute flow

angle' at section 20 (configuration Al.

359
 SINGLE FREQUENCY SHAPE DUAL FREQUENCY SHAPE

FREQUENCY
Mu

.45
.60
(!) .75
I!I .85

0200

" " I' ao J2

Figure 11. - Return channel pressure recovery coefficient vs. absolute flow angle
at section 40 (con figuration A).

IS

IJ

"

!'lJ .45 .60 .75 .85

THROIILlNG I!I I!I

THE VALVE

OPENING
THE VALVE

10 80 .0 100 110 120 IJO lOa ISO 160 110 110 190 200
fJJ 210 220

Figure 12. - Normalized observed frequency vs . inlet flow coefficient

(configuration B).

360
 Mu PoIPo.o PH.

'" .45 .60 .75 .85

'_nllllC

-,.

( tiC. 'ALfE

&'.,.10 I
I
I
I
I
I
I
I
I
I
I
I
.
I
I
W I
0\ i
1
I-'

I I

I
I

I
I

I
I
I
I
I
I
I
I

,.,,01 '.
'00 '20 ,00 , ... , ... 200 22. 2'. 2'. III 2

Figure 13. - Static pressure oscillation reduced Figure 14. - Static to tot al and total to total

amplitude (@ f s) at section 20 vs. inlet flow pressure ratios at section 20 vs. inlet flow

coefficient (configuration B). coefficient at different Mach numbers

(configuration B).
 Mu PoA\IO P!Po:o Mu Po.,f).uo P/PolO

.45 .45
.60 .60

.75 . .. .75 II
. 85 . II .85 OJ

:
I

I
I

.. I
I
j
I
I f
/
W .,'

l
'"
0\
N l'I
I
I

I
I I

I : :
IF' '

:::::::::
=: : :: : : : . .

I
I
!
I
"'/:
:
/
;

- -
1.0,orl--T----r--r--r----r---'--T---.--T---' 1 ... 1
8. 100 120 14<) 160 180 200 220 240 260 fP 280 10 80 100 120 14<) '00 180 200 220 f/I 240 260 280

Figure 15. Static to total and total to total Figure 16. Static to total and total to total
pressure ratios at section 40 vs. inlet flow pressure ratios at section 60 vs. inlet flow
coefficient at different Mach numbers coefficient at different mach numbers
(configuration B). (configuration B).
 :ICC

.,
260

24C JI

j
Mu
a" .. .45
.60
220 28
I!) .75
I!I .85

200 25
1

180 22

160 19

W
0'1
W
1 40 16

120 u

Mu
100 .. 45 10
..
.
-

....
. .
.60
...
.

z. I!) .75
D. ...
I!I .85
.

%
.z.
80 .J JIf % o

% %

%
10 %
1 1 1 1 1
ONO 0.050 0.070 0.010 0.110 0130 0.150 0170 e 0.110 O.NO 0.050 0.070 0090 0110 O.laD 0-150 0-170 erZ:-I1O
r20

Figure 17. - Flow coefficient versus nondimensional Figure 18. - Absolute flow angles at section 20 and
speed at section 20, at different Mach numbers section 40 vs. nondimensional radial speed at
(configuration B). section 20 (configuration B).
 O OO

CPD Mu
'" .45
" .60
I!) .75
!!I .85

0100

0.00 -'-----r--r-r-----'r----r---'
r-
4 7 10 IJ 16 19 22

F igure 19. - Vaneless diffuser pressure recovery coefficient vs. absolute flow
angle at section 20 (configuration B).

0.500

Cp,,<:

0.400

'"
..

OJOO


0200


,1W
;
0.100 ,Ii;
Mu ,Ie;
,1tr.;
.. .45 / /
/1-;
... .60
/ /
0000 .75 ;

I!)

!!I .85

-O.IO
5 8 II 14 17
20 23 26 29 a 40(0) 32

Figure 20. - Return channel pressure recovery coefficient vs. absolute flow angle
at section 40 (configuration B).

364
 CXl\I1PARATlVE STUDY OF SUBSYN:lffi(N){JS ROfATIN3 FLCW PATIERNS

IN CENIRIFlDAL CUVlPRESSCRS WIlli VANELESS DIFFUSERS

P. Frigne*
State University of Gent
9000 - Gent, Belgium

R. Van Den Braembussche

von Karman Institute
1640 - Rhode Saint Genese, Belgium

SUMMARY

A comparative experimental investigation of the unstable operating

modes of a centrifugal compressor is made. Hot wire measurements, in
two different compressor configurations, allowed to distinguish between
the different unsteady flow pattern. Impeller and/or diffuser rotating
stall has been observed, depending on the flow conditions. The measu
red relative rotational speed of this perturbation is cross-checked
with other experimental data and it is shown that the rotational speed
is strongly dependent on the type of rotating stall. The diffuser
absolute inlet flow angle at the onset of diffuser rotating stall
agrees well with the value predicted by an existing stability criterion.

INTRODUCTION

Self excited vibrations of turbomachinery components, due to aero

dynamic forces, can be divided into two groups. A first one concerns
the vibrations where the destabilizing forces are function of the
whirling motion of the rotor. The initial displacement of the shaft
influences the flow conditions in the labyrinth seals or creates
an asymmetric blade loading on the impeller. The corresponding forces
sustain the initial perturbation because of the negative damping.

A second type concerns the vibrations where the main flow is de

stabilized by hydrodynamic or viscous forces which are independent of
the resulting mechanical vibration. They are known as rotating stall
and surge. The last ones have also an important influence on compres
sor performance.

At low pressure levels, the corresponding forces are often too

small to have any significant influence on the shaft vibrations, but
sometimes result in blade failures. However, at high pressure levels
the excitation becomes much larger and very often limits the_range of
operation because of vibrational problems.

This paper is limited to the study of the second type of desta

bilization in centrifugal compressors with vaneless diffusers.

*
Formerly Research Associate at von Karman Institut<.

365
 Hot wire measurements have been performed in two different com
pressor configurations in order to investigate the different unsteady
flow pattern.

SYMBOLS

diffuser width

static pressure rise coefficient

number of stall cells

static pressure

radius

Reynolds number, based on diffuser absolute inlet velocity

and diffuser width

u peripheral velocity

V absolute velocity

absolute flow angle, measured from meridional plane

s blade angle measured from meridional plane

mass flow coefficient

p static density

slip factor

absolute angular velocity of stall cells

angular velocity of the impeller

Subscripts

o settling chamber

1 inducer inlet channel

2 diffuser inlet

3 diffuser outlet

c critical

366
 DESCRIPTION OF TEST FACILITY AND INSTRUMENTATION

The experiments have been performed in the open loop test facility
(Fig. 1a) of the State University of Gent. Mass flow is fed into the
settling chamber by a normalized tube which allows for flow control
and mass flow measurement. Inlet guide vanes can be installed in the
suction pipe if prerotation is required. Flow is discharged at atmos
pheric pressure after a 180 turning at the diffuser exit.

All measurements have been performed with a vaneless diffuser.

Geometrical parameters of the unshrouded impeller (Fig. 2) are listed
in Table 1.

Besides the classical pressure and temperature probes, required

for overall performance measurements, four hot wire probes are also
installed. The hot wires are mounted perpendicular to the mean flow
direction. One is installed at impeller inlet (II), one at diffuser
inlet (10) and two at R/R2 = 1.3 inside the diffuser (MD1 and M02) at
42 circumferential distance (Fig. 1b). The linearized signals are
visualized on an oscilloscope and processed with a HP 3582A two channel
Fourier spectrum analyser. This allows to define the power spectrum,
coherence function and phase transfer function.

Perturbations rotating in the diffuser will result in periodic

signals at MOl and M02 with period T2 . Because of the 42 difference
in circumferential position there will be a time lag Tl which allows
to calculate the number of stall cells i, and the rotational speed ws:

i = 360/42 * Tl/T2 (i must be integer)

W = 2n/(iT2)'
s

TEST RESULTS WITHOUT PREROTATION VANES

The first series of measurements have been obtained with an axial

flow at compressor inlet. The overall performances are shown on
figure 3. The static pressure rise is not related to the mass flow
coefficient 2 but to the diffuser inlet flow angle a2' The last
parameter is used here because of its important influence on diffuser
stability and is defined by

tg a2 = /2'

The slip factor is calculated from correlations and is checked against

the measured impeller temperature rise. Increasing values of a2
correspond to decreasing mass flow.

The static pressure rise is non dimensionalized by

poU /2 to
become independent of RPM.

All curves show a maximum at a2 = 76 which coincides with the on

set of unstable flow. At higher RPM, these curves do not continue to
the left because of surge. At low RPM, surge occurs only at a2 = 87

367
which leaves a large range of flow conditions where unsteady flow can
be observed . The hot wire measurements are therefore made at 2000 RPM
for al values between 7 6 and 87.

The hot wire traces shown on figure 4 a are measured at rotor

inlet (II) and diffuser inlet (10) for a2 =7 6 . The relative ampli
tude of the oscillations, defined by (V -V )/V ' amounts to .14
max min RMS
at rotor inlet and .23 at diffuser inlet . The power spectrum and co
herence function, shown on figures 4 b and 4 c, clearly indicate an
instability at 27 .2 Hz. Inside the diffuser, at R/R2 = 1.3, the flow
is steady as shown on the power spectrum of probe MOl (Fig 4 d) . One
can therefore conclude that these instabilities are due to impeller
rotating stall, which can be explained by the high positive incidence
(12) at inducer tip for this operating point .

Measurements at a2 = 7 8 (Fig . 5a) do not only show an increase

in amplitude to .37 at impeller inlet (II), but also an extension of
these perturbations downstream inside the diffuser (MO). The hot wire
traces MOl and M02 (Fig. 5b) have a relative amplitude of .13. The
phase lag between both signals indicates the presence on one stall
cell rotating at 7 7 % of the rotor speed. The power spectrum and co
herence function of the II and MO signal (Figs. 5c, 5d) reveal a
decrease of the basic frequency to 25 .6 Hz and the presence of impor
tant harmonics at 52 and 77 Hz.

Figures 6a and '6b are measured at a2 81 and indicate a further

=

increase in amplitude to 200% at rotor inlet (II) and 18% inside the
diffuser (MO). The basic frequency is now 23 .2 Hz and ws/ .7 0.
= The
harmonics have almost disappeared at rotor inlet but are unchanged
inside the diffuser .

Low frequency oscillations completely suppress impeller rotating

stall at a2 = 87 . The hot wire traces of figure 7 a seem to indicate
that the frequency at rotor inlet (II) ;s twice the frequency of
oscillations inside the diffuser . However, hot wires read only the
velocity component perpendicular to the wire and are insensitive to a
180 change in flow direction. The real flow velocity at rotor inlet
changes sign (return flow) and is indicated by the dashed line . Both
signals are in phase at 6 .4 Hz and indicate compressor surge . The
power spectrum on figure 7b shows the presence of several harmonic
frequencies as well at rotor inlet as inside the diffuser.

This series of measurements show only one type of rotating stall,

originating from the rotor . The oscillations are gradually increasing
in amplituae and extending downstream inside the diffuser. They are
due to one stall cell rotating at high relative velocity .

TEST RESULTS WITH INLET PREROTATION VANES

According to a stability criterion, developed by the authors

(Ref . 1), vaneless diffuser rotating stall should occur at a2 77.
=

However, only impeller rotating stall has been observed in previous

experiments . A possible explanation for this disagreement could be

368
that diffuser rotating stall is completely suppressed by the impeller
rotating stall if the latter one starts first.

In order to verify this hypothesis, inlet guide vanes have been

installed in the suction pipe, in order to create 60 prerotation at
the impeller inlet. Preliminary calculations indicated that this
would delay impeller rotating stall at least to a2 82.
= In this way
a sufficiently broad margin for diffuser rotating stall would be
created.

The overall compressor performance map is shown on figure 8.

Compared to previous results, one can observe a 6 shift to the left
of impeller characteristics. The maximum compressor pressure rise
has slightly decreased because of the prerotation. The diffuser
pressure rise is almost unchanged.

First instabilities are observed in the diffuser at a2 77

=

(Fig. 9a). The frequency is 17.6 Hz (Fig. 9b) with a coherence of

55% (Fig. 9c). From the 132 phase shift between MD1 and M02, one
can calculate that three stall cells are rotating at ws/ Q =.176.
These instabilities are not observed at the inducer inlet (Fig. 9d)
and one can therefore conclude that they are due to diffuser rotating
stall.

These instabilities have a maximum amplitude of .11 at a2 78

=

but completely disappear at a2 =80. They suddenly reappear in the

diffuser at a2 = 805 with a relative amplitude of .14 and a cohe
rence of 98%.

With the achieved prerotation, the critical flow angle for impel
ler rotating stall is predicted at a2 82.
= This is experimentally
verified by the power spectra obtained at a2 = 826. The power peak
at 23.2 Hz ;s measured at rotor inlet and inside the diffuser. The
phase angle of 52 indicates that one stall cell rotates at ws/Q =

.673. These results are very similar to impeller rotating stall

observed with zero prerotation. The power peak at 19.2 Hz is observed
only inside the diffuser and the phase angle between MOl and M02 indi
cates that three stall cells are rotating at ws/Q = .192. I n analogy
to the measurements at lower a2 values, one can conclude that this is
diffuser rotating stall.

When closing the throttle valve further to a2 831, diffuser

=

rotating stall remains unchanged, but impeller rotating stall

(23.2 Hz) disappears completely (Fig. 11).

A new type of unstable flow, with a frequency of 8.8 Hz is

observed at a2 = 836 (Fig. 12a). These instabilities are observed
only in the diffuser and correspond to a second type of diffuser rota
ting stall. Two stall cells are rotating at w s/Q = 13.2%. These high
and low frequency diffuser rotating stall patterns do not occur simul
taneously but in a rather intermittent mode. This relative amplitude
increases further to .22 at a2 =847 (Fig. 12b).

369
 The low frequency oscillations (6. 4 Hz), observed at a2 86, =

are in phase and due to compressor surge. Rotating stall has disap
peared at this point.

COMPAR I SON OF T HE TWO TE S T SER IE S

Impeller rotating stall can be observed at impeller inlet and

inside the diffuser. In both test series, the onset of this type of
stall coincides with a 12 positive incidence at impeller tip. Pre
rotation therefore allows to delay this type of instability to lower
mass flow coefficients.

Diffuser rotating stall can only be observed in the diffuser and

starts at a2 = 77 . This value will be compared to other data in the
next chapter.

Both types of rotating stall can exist simultaneously if diffuser

rotating stall occurs first. This absolute frequency and amplitude
is of the same order of magnitude and measurements with one single
probe do not allow to distinguish between them. However, the relative
rotational speed is quite different as shown in figure 13.

Impeller rotating stall has a relative rotational speed of about

.7 and this velocity seems to be unaffected by prerotation. The rela
tive rotational speed of diffuser rotating stall ;s of the order of
.10 to .20 only. This difference is in agreement with experimental
observations of Lenneman & Howard (Ref. 2), Misuki et al. (Ref. 3)
and Abdelhamid (Ref. 4). This difference could therefore be used to
distinguish between impeller and diffuser rotating stall. However,
the definition of relative rotational speed not only requires to
measure the absolute frequency but also the phase difference between
two circumferential positions.

T HEORET ICA L PRE D IC T ION S OF VANE LE S S D IFFU SER

RO TAT ING S TA L L

Several stability criteria for vaneless diffusers have been

developed and published in the literature. They are based on dif
ferent assumptions and result in contradictory conclusions.

The stability criterion for vaneless diffusers of C. Rodgers

(Ref. 5) is based on theoretical considerations concerning variation
of Cp in function of a2' The critical value of a2C, separating the
zone of stable operation from the unstable zone, is function of the
non dimensionalized diffuser width b2/R2 and radius' ratio R3/R2.

This conclusion is similar to the one obtained by W. Jansen

(Ref. 6). His criterion is based on detailed calculations of the
viscous flow in vaneless diffusers. Besides diffuser width (b2/R2),
radius ratio (R3/Rz) and flow angle (a2)' stability is now also a func
tion of the Reynolds number. Both theories suggest that the critical
flow angle a2C is decreasing for decreasig diffuser width. This

370
would mean that for a given impeller, an unstable vaneless diffuser
cannot always be stabilized by decreasing the diffuser width.

A similar but more precise calculation by Senoo (Ref. 7), however,

results in a criterion which shows an opposite trend in terms of a2C
as a function of diffuser width. The critical flow angle is now
increasing for decreasing diffuser width. An unstable diffuser can
therefore be stabilized by decreasing diffuser width. This theory
also allows to draw important conclusions in terms of diffuser radius
ratio R3/R 2 and about the influence of non uniform flow at diffuser
inlet.

The stability criterion of A. Abdelhamid (Ref. 8) is based on

rotor-diffuser interaction and does not account for viscous effects.
This criterion relates stall operation to the pressure and tangential
velocity fluctuations non dimensionalized by radial velocity fluctua
tions. This criterion is not easily applicable and stability is inde
pendent of diffuser width.

In order to eliminate the contradiction between the different

criteria, the authors (Ref. 1) have performed some unsteady flow
measurements on different impeller-diffuser combinations. These
results and other experimental data from the literature allowed to
derive the following correlation. All values of a3C fall within a
relatively narrow band as shown on figure 14 after they have been
corrected for Reynolds number influence according to figure 15.
Points 1 to 5 are experimental data obtained by the authors. Points
6 to 16 are derived from published data.

These experimental data indicate an increase of a3C for decreasing

diffuser width and agree very well with the predictions of Senoo. His
theoretical results do not indicate an influence of Reynolds number on
diffuser stability for values of b2/R2 < .10. However, the experimen
tally observed variation of a2C in function of Reynolds number is in
agreement with the predictions of Jansen (a2 = 6 = between Reb
2
= 104
and Reb 105).
=

The flow angles 1 to 16, are calculated from measured temperature

rise and slip factor correlations. Therefore, some doubt could exist
about the precision by which these angles are defined. Some special
tests have been performed in order to evaluate this error. They are
made for one impeller-diffuser combination at two different Reynolds
numbers. Flow angles at diffuser inlet are measured by a crossed hot
wire. The experimental values shown by points 1 7 and 18 are already
corrected for Reynolds number influence.

The observed differences with the theoretical curve are within an

acceptable tolerance and therefore provide an additional verification
of this correlation.

Point 19 ;s the value obtained from the present test series and
confirms the stability limit for diffuser rotating stall.

371
 CONCLUSION

Measurements described in this paper allow to distinguish between

impeller or diffuser rotating stall. The onset of impeller rotating
stall can be related to inducer incidence and the relative rotational
speed is about .70. Diffuser rotating stall is a function of diffuser
inlet air angle and Reynolds number and has a relative rotational speed
of .1 to .2.

These experiments provide a criterion to distinguish between both

phenomena and indicate which part of the copressor should be modified
to achieve stable operation.

Flow conditions at the onset of diffuser rotating stall have been

compared with a criterion for diffuser stability and agree very well.

REFERENCES

1. VAN DEN BRAEMBUSSCHE, R.: FRIGNE, P.; ROUSTAN, M.: Rotating non
uniform flow in radial compressors.
AGARD CP 282, pp 12-1 to 12-14.

2. LENNEMAN, E. & HOWARD, J.H.G.: Unsteady flow phenomena in

rotating centrifugal impeller passages.
ASME Transact., Series A : J. Engineering for Power,
Vol. 92, No. 1, January 1970, PI> 65-72.

3. MISUKI, S.; KAWASHIMA, Y.; ARIGA, I.: Investigation concerning

rotating stall and surge phenomena within centrifugal compressor
channels.
ASME Paper 78 GT 9.

4. ABDELHAMID, A.N.: Effects of vaneless diffuser geometry on flow

instability in centrifugal compressor systems.
ASME Paper 81 GT 10.

5. RODGERS, C.: Impeller stalling as influenced by diffusion

limitations.
ASME Transact., Series I : J. Fluids Engineering,
Vol. 99, No. 1, March 1977, pp 84-97.

6. JANSEN, W.: Rotating stall in radial vaneless diffusers.

ASME Transact., Series D : J. Basic Engineering,
Vol. 86, No. 6, December 1964, pp 750-758.

7. SENOO, Y. & KINOSHITA, Y.: Limits of rotating stall and stall in

vaneless diffuser of centrifugal compressors.
ASME Paper 78 GT 19.

8. ABDELHAMID, A.N.: Effects of vaneless diffuser geometry on flow

instability in centrifugal compression systems.
ASME Paper 81 GT 10

372
9. LIGRANI, p.r.; VAN DEN BRAEMBUSSCHE, R.; ROUSTAN, M.: Rotating stall
measurements in the vaneless diffuser of a radial flow compressor.
ASME Paper 82 GT 257

TABLE 1. - GEOMETRICAL CHARACiERITICS OF tHE COMPRESSOR

tip
R1 = .095 b3 = .016 m

hub
R1 = .054 m N =
20

tip
R2 = .208 m 131
= 50

hub
R3 .400 m 32

131
= =

b2 = .016 m 132
= 0

373
 VANELESS DIFFUSER
+'+
+1+ DISPERSION IM P E LL E R
\
o

TUBE C O U PL ING

SETTLING
++ I
-

+
CHAMBER ;

+1
+' METER
TORQUE

D.C. MOTOR
j
+

Figure l ( a ) . - Open loop centrifugal compressor facility.

NOZZLE

MD1

NORMALIZED
TUBE

Fi gure l ( b ). - Hot wire locations.

INLET TUBE

rTHROTTLE
. VALVE

Figure 2. - Unshrou ded impeller.

374
1.3

1.2

P3-PO

1. 1 2
U
p 0 2
..!..
2

1.0

0.9

0.8

o 7 -

0,6

0.5

0.4

0.3 ---0--- RPM = 2000

--- +--- RPM = 4000

0 .2
- --x---"" RPM = 6 000

0.1

o
90 80 70 60

Figure 3. - Compressor performance with zero prerotation.

375
 L.,....,,H" _.

] 2m/s ID

] 'mIl

I.-___, 100 m 5
Figure 4(a). Figure 4(b).

COHERECEI 1.' FS eM " "ICRI - al.u.y

MlCill ,. CIt 81 Y FI

II+ID

MD

Figure 4(c). Figure 4(d).

MD1
]4m/s, \.,,;
. \. J{ r/'.
_
V \r V - \J"V-"
,y'\ I

D2

'w
- f\
;" V
tV'\\r
r\ (,
J . 1"-
4m 3
""" ,_.JIII,.'

'--___-' , lOOms '-- _ __ -', 100 m 5

Figure 5(a). Figure 5(b).

28.948Y 1114a.. ...... ..IY

CH A "ICRI - 1" COHERENCE I 1.' FS
CH 81 + 8d8Y FS "ICRI 91
II+MD

-,. -

8WI1"H"

Figure 5(c). Figure 5(d).

376
 IYII
I ]
A
/"

\
\j
1 \j'\/
,
\\;.J J'
4m/s
MO
/'

yo
; \ ,.) /\{\ ]
\1"" " . /
/'f\
\v(,.J \/ 4 m Is
\

\) 23.2 Hz

1..-___ --', 100 m s

Figure 6(a). Figure 6(b).
eM A MKR' - 17.9dBY
eM B' + IdBY FS

] lOm/s

] 10m/s
13.6 Hz

1.-.-...1 100 m s

Figure 7(a). Figure 7(b).

377
 1.4

1.3

1.2 I
1.1

0.9

P2-PO
0.8
POU
0.7

0.6

0.5

0.4

0.3

o RPM = 2000
0.2 -+- RPM = 4000
- x- RPM =6000

0.1 .

0
90 80
-

Figure 8. - Compressor performance with 600 prerotation at the inducer tip.

378
 CN R NKR. - 17.741Y
CN .41Y ,. 'I""'vV
I.'

"MD2

.... I.

L--J. 100m s
Figure 9(b).
Figure 9(a).
CONIRENCE' , ,. eN ...... - 44.7.v
... . .. .. eN .v PI Iiall

MD1 + M D2
LIIM"O'.
II

.... I."h ... ... " . . ...

Figure 9(c). Figure 9(d).

CN .. ... . - '1.1.v
eN .v PI-

\ "MD
MD2

Figure lO(a). Figure lO(b).

CH R NKR' - 26.64'Y
CH . + .... Y ,.

MD1
/

"MD2

8W' 111H"Hz
z '

Fi gure 11.

379
 eM III ... . - 2e.,aY
fllll eM lev I IIII

MD1
/
\
MD2
.1. Hz 241. I Hz I.. . . .

Figure 12(b).
F igure 12(a).

1. 00 r-------.,--.---I--;--,
WS/Q
-
.8,0 -

.......
70 - 0...... 6 0...0
-

...... """00- 60- -

60 - IMPELLER ROTATING ST All -

o NO PREROTATION
6 WITH PREROTATION

DIFFUSER ROTATING STALL

o WITH PREROTATION
.30 r
-

.20 r- -O.D-o-.Q. 00- -0.0-0

-

---0.
-0 000- __
"'0" -0
-
--
.10 r
-

I I I

90. 85. 80. 75.

Figure 13. - Relative rotational speed.

380
20

90 - a. 2C (DEGREES)

16

w--
--
12
-@
-- @)
- ..

@"
-@?S-
8 w--
-
4 /(D
o
o . .02 .04 .06 .08 .10 .14

Figure 14. - Critical ;nlet flow anqle complement versus nondimensional

diffuser width.

8
lla.2C (DEGREES)
SAME SYMBOLS AS FIG. 14
4

-.4

-8

Fi gure 15. - ritical inlet flow angle correction factor versus Reynolds
number.
381
 EXPERIlVlENfAL MEASURE.VJENfS OF HY.DROOYNAMIC STIFFNESS

J.Vlt\1RICES FOR A CEN1RI FU3AL PUVIP IlVlPEL:LER.-A<

D. S. Chrunieh, A. J. Acosta, C. E. Brennen, T. K. Caughey, and R. Franz

California Institute of Technology
Pasadena, California 91125

ABSTRACT

The objective of the Rotor Force Test Facility at the California Institute of
Technology is to artificially orbit the center of rotation of an impeller enclosed
within a volute over a range of frequencies from zero to synchronous and to measure
the resulting forces on the impeller. This paper reports preliminary data from the
first stage experiments in which the shaft is orbited at 'low frequency. Steady
volute forces along with stiffness matrices due to the change in position of the rotor
center are measured. Static pressure taps around the volute are used to obtain volute
pressure distributions for various fixed positions of the impeller center and for
various flow rates. Static pressure forces are calculated from these pressure distri
butions allowing a more complete analysis of the components of the impeller forces.
Comparison is made with various existing theoretical and experimental results.

INTRODUCTION

It has become evident in the last few years that the main flow through a pump or
compressor can contribute to the rotor dynamic inertia, damping and stiffness matrices
of the machine in a significant way (ref.l, 2). Consequently in many applications an
accurate rotor dynamic analysis of a turbomachine may require inclusion of the com
ponents of these matrices resulting from the flow through the impeller as well as the
contributions from seals and bearings. In the past the hydrodynamic forces on a
radial impeller enclosed in a volute have been primarily studied experimentally and
many of the earlier references are based on the semi-empirical relation introduced by
Stepanoff (Ref.3). Other experimental papers followed in the same spirit namely those
of Domm ano Hergt (Ref.4), Agostinelli et al (Ref.S) and Iversen et al (Ref.6). Later
as a first step towards studying the whirling problem, Hergt and Krieger (Ref. 7) mea
sured radial forces at various eccentricities of the impeller.

Theoretical studies directed toward evaluation of the impeller forces include

the work of Domm and Hergt (Ref.4) who modelled the impeller by a source vortex placed
withirl.the vblutp. Their analysis demonstrated the existence of a lateral force on the
impeller even in the absence of an orbiting motion of the shaft. Colding-Jorgensen
(Ref.8) extended the work of Domm and Hergt (Ref.4) by imposing an orbiting motion on
the soure/vortex model of the impeller. His results are discussed below. Other more
complex methods have also been developed. The actuator disk model, suggested by
Chamieh and Acosta (Refs. 9, 10) was used to obtain inertial terms that do not arise
in the source-vortex model. More recently both Imaichi, Tsujimoto and Yoshida (Ref.
11) and Shoji and Ohashi (Ref.12) repre$ent the impeller wakes by a finite number of
rotatng vortex sheets. Inaichi et al (Ref.ll) solved the unsteady problem of a
fixed impeller inside a volute and Ohashi et al (Ref.12) solved the unbounded whirl
ing problem. However neither were able to formulate the more complex probiem of an
orbiting impeller enclosed within a volute.

*
This work is supported by NASA George Marshall Space Flight Ctr.,Contract NAS 8-33108.

382
 NOMENCLATURE

A Impeller discharge area 27fR b

2 2 2
=

b Impeller discharge width

2

Cp Pressure coefficient, (p - P ) / ip U;
T2
e Instantaneous vector displacement of the external balance flexure system

[F } 0
Average volute force on impeller (vector)
2
[F*} 0
Non-dimensional average volute force on impeller (vector) =
[Fo}/ p A2U2
F* F* Components of (F*} in X,Y directions (see Fig.4)
X' Y o
F. , [F. }
--:l
Instantaneous hydrodynamic force on impeller (vector)
1.

[Fr.} Fundamental harmonic component of [F. }1.

[Ks j Stiffness matrix for external balance

[Kij} Hydrodynamic force matrix for volute/impeller

[K!j} Non-dimensional hydrodynamic force matrix

N,w Main shaft speed (rpm and radian/sec)

p Static pressure measurement in volute

Discharge total pressure

PT2

Q Volume flow rate

R2 Impeller discharge radius

U Impeller tip speed

2

t Time

X,Y Cartesian coordinates (Fig.4)

..9.., to} Instantaneous position of impeller center (vector, Fig.4)

[0 } 0
Mean position of impeller center

Fundamental harmonic component of (0)

(0f2}
, [E} Imposed eccentricity of impeller (lEI = ].26mm).

A Logarithmic spiral angle of volute

Flow coefficient, Q/A U

cp 2 2
=

383
 2
Head coefficient Total pressure rise across pump/pu
2
=

p Density of water

Radian frequency of whirl motion

e Angular position measured from the tongue (Fig.4)

w Radian frequency of impeller rotation

ROTOR FORCE TEST FACILITY

In an earlier paper (Ref.13) the facility was described in detail. We will

merely repeat that the objective of the facility is to impose known orbiting motions
of radian frequency Q on the basic rotary motion of a number of typical centrifugal
pump impellers (radian frequency, w, shaft speed, N). Two force balances are incorp
orated: (a) a simple balance external to the bearings and seals for scoping experi
ments and the measurement of steady volute asymmetry forces and stiffness matrices
at zero whirl frequency (Q=O) and (b) an internal balance upon which the impeller
is directly mounted for measurements at non-zero whirl frequencies (see Figs. 1, 2 and
Ref.13).

This paper reports measurements using the external balance and a l6.2cm diameter
five-bladed Byron-Jackson centrifugal impeller (designated Impeller X) with a specific
speed of 0.57. A matching logarithmic spiral volute (Volute A) with a spiral angle of 86
was employed with this impeller (other impeller/volute combinations will be tested shortly).
The volute was design.ed in accordance with the assumption of Ref. 3. During the meas
urements face seals on both sides of the impeller were backed off to prevent inter
ference with the force measurements. Various face seal clearances were investigated.
Figure 5 presents the performance characteristic for two seal clearances.

The external balance, consisting of three aluminum flexures (see Figs. 1, 2 and
4) was used with the entire impeller/eccentric drive system floating on soft spring
supports attached to the ceiling. Three stainless steel flexures parallel with the
shaft ma'intained the alignment i n this configuration. A small D.C. motor was also
mountedon the floating assembly to produce a shaft orbit 'speed of 3 rpm. The purpose
of this was to allow continuous sampling of the forces at all locations of the shaft
around a circular orbit with a radius of 1.26mm. By comparison with the forces at
certain fixed locations it was determined that the dynamic effects of this 3 rpm orbit
speed were negligible.

Two of the strain-gaged flexures are mounted horizontally (Figs.2 and 4) for mea
surement of horizontal force
. and torque; the other measures the vertical force. Cali
bration.s were performed by known forces applied by systems of wires, pulleys and
weights. The linearity of the calibrations was within 1% over the entire range of
calibration (-20 to +20kg) and the interactions were less than 1%.

Signal processing was accomplished by means of an eight channel digital signal

processor which was phase locked to an optical encoder attached to the shaft of the
3 rpm motor. In this way each of the 8 channels of data (three flexures plus pres
sure transducers, flow rates, etc) are sampled at 32 equally spaced positions of each
rotation of the eccentricity. Furthermore this sampling was continued for 15 or more
rotations in order to obtain averages of the forces for each of the 32 geometric
positions.

384
 The flexure signals were also recorded on magnetic tape and processed through
a digital spectrum analyser to investigate their content. A major peak was observed
at the 3 rpm whirl frequency. Much smaller peaks were observed at higher harmonics
of this whirl frequency. The largest observed magnitude of the second harmonic was
less than 10% of the fundamental indicating that the impeller forces at the small
eccentricity employed here vary quite sinusoidally around the circular eccentric orbit.
In other words the forces vary linearly with cartesian displacements X and Y and
this variation can be accurately represented by a stiffness matrix.' In addition to
these low frequency components the spectral analysis also showed peaks at the shaft
frequency and the higher blade passage frequencies. These represent dynamic forces
which are not accurately measured by the external balance and will be the subject of
a later paper based on the internal balance measurements.

The geometry and notation of the impeller shaft location, external balance
flexures and volute position are shown in Figure 4 (viewed from the pump inlet).

CALCULATION OF IMPELLER VOLUTE FORCES AND STIFFNESS MATRICES

The flexure system, though quite stiff, has some deflection due to the hydro
dynamic forces on the impeller; indeed this deflection is monitored by the external
balance elements. One consequence of this is that the actual position of the shaft
center is a combination of the 1.26mm eccentricity plus the deflection of the flexure
system. All of the data on the impeller forces and stiffness matrices were corrected
for this effect in the following way (see Fig.4). If [EJ is the instantaneous shaft
center due to the imposed eccentricity of [Ks]
1.26mm is the stiffness matrix of
and
the flexure system (measured by loading under conditions of no motion) then the actual
position of the shaft (see Fig.4) is

l
[o)=E]+[e } ; Le } = [Ks]- [ F.1 } (1)

where [ Fi } is the total force acting on the floating assembly due to the n and w

motions. T are forces were found to be negligible so tnat, with proper zeroing of the
balance, the forces, Fi , [ } are entirely hydrodynamic (see the no -impeller results of
Figure 9 for confirmation).

Fourier decomposition of the data yielded average hydrodynamic forces !o as

sociated with the mean position of the shaft center and first harmonics, [Fn J and
[onJ at the orbit frequency,
n (3 rpm). Then the hydrodynamic force matrix,
[Kij ] is defined as

[ Fn} = [Kij H 0rt} (2)

l
where tOrt}
is the fundamental of [Ks]- Fi [E+
and therefore [
[Kij]}}can be
l
calculated since [Ks]- E,
and Frt [
are all known. J
In the present experiments
the correction [Ks]-l Fn [ }
could be as large as 10% of on .'
The forces and force matrices are presented in non-dimensional form using
the normalizing factors defined in the nomenclature.

ADDITIONAL MEASUREMENTS

Two types of additional measurements were performed in order to investigate the

source of the impeller forces and stiffness matrices. Firstly, measurements of these

385
quantities were also obtained (i) with different seal clearances (ii) at zero flow
rate by total flow blocyage exterior to the pump (iii) with the impeller inlet and/or
exit blocked by sheet metal attachments (iv) with the impeller removed (v) with a
dummy solid impeller having the same exterior shape. Secondly, static pressure distri
butions within the volute were measured by means of the static pressure taps shown
in Fig.3 for four fixed positions of the eccentricity (the positions 'farthest,
'closest', 'right' and 'left' shown in Fig.4) and various flow coefficients. These
pressure distributions are pesented non-dimensionally using a pressure coefficient,
Cp, based on the downstream total pressure and
pU /2.

FORCE DATA; RESULTS AND DISCUSSION

The reduced data for the average force on Impeller X due to the Volute A when
the mean position of the former is coincident with the center of the logarithmic
spiral volute is presented in magnitude and direction form in Figs. 6 and 7 and in
terms of its cartesian components in Fig.8. The data presented is for seal
clearances of 0.14mm. However quite similar results were obtained for clearances of
0.79mm even though the performance characteristic in the latter case is significantly
degraded. Some results for both a forward 3 rpm sampling orbit speed (Q> 0) and
with the direction reversed (Q< 0) are shown to confirm that this orbit speed had
little effect on the results. The non-dimensionalized force data for different shaft
speeds (600 rpm up to 2000 rpm) is in substantial agreement though there would appear
to be a slight increase in the magnitude of the force coefficient at low flow coef
ficients when the speed is increased. It seems likely that this is a Reynolds number
effect. From a practical viewpoint it is clear that the Volute A and the Impeller X
are well matched at their design flow coefficient 'P = 0.092 since the force on the
impeller is virtually zero in this operating state.

Also shown in Figs. 6 and 7 are the experimental results of Agostinelli, Nobles
and MQckridge (Ref.S) and Iversen, Rolling and Carlson (Ref.6) for pumps with specific
speeds of 0.61 and 0.36 respectively. The former are in close agreement with the pre
sent results since the specific speed of the present Impeller X/Volute A combination
is 0.57. The results of Iversen et al for a lower specific speed also appear consis
tent in magnitude with the present results. For clarity a fourth set of experimental
results by Domm and Hergt (Ref.4) are presented in Fig.8. These results for a
similar volute (A= 86.3) appear to be in substantial disagreement (lower forces by a
factor of 2 or 3) with the results of Fig.6; the reason for this is not clear at the
present time.

Finally Fig.8 also presents the theoretical results of both Domm and Hergt (Ref.
4) and Colding-Jorgensen (Ref.8) for volutes with angles close to the 86 of the pre
sent set-up. These theories which are based on source/vortex models of the impeller
also underpredict the present results by a factor of about two.

Figure '9 present the force coefficient data for a number of tests performed with
the objective of assessing the source of the large forces present at conditions of
no flow ('P= 0 in Figs. 6, 7, 8). All of the data of Fig.9 appears to be virtually
independent of shaft speed ndicating that forces are proportional to the square of
the tip speed. The lack of any appreciable force in the absence of the impeller
(No impeller points) confirms that these shut-off forces are hydrodynamic in origin.
The data for the solid impeller, for Impeller X with the exit blocked off by a
sheet of metal and for Impeller X with both the inlet and exit similarly blocked are
all similar. Though data is shown for 0.14mm face seal clearances, the data for O. 79mm
face seal clearances is very similar in magnitude. We conclude that this force

386
coefficient component (about 0.04) is due to fluid frictional effects and the induced
pressure gradient acting on the entire exterior surfaces of the impeller. The data
with only the impeller inlet blocked is significantly higher, probably because the
discharge surface (blades and gaps) has greater friction. Finally the force data for
the flow blocked exterior to the pump is significantly higher still. The reason for
this is not entirely clear as yet but the associated observation of substantially
greater prerotation under shut-off conditions may be significant.

HYDRODYNAMIC FORCE MATRICES

Typical non-dimensional hydrodynamic force matrices, [Ktjl, are presented in

Fig.lO; these contain all the information on how the forces presented in Figs. 7, 8
and 9 vary as the position of the impeller center changes provided the change is with
in the linear region. The hydrodynamic stiffness matrix experienced by
the shaft is [-Kijl. The data shown in Fig. 10 is for seal clearances of 0.14mm,
various impeller speeds and flow coefficients. We observe that the variation with
N is consistent with that anticipated by the non-dimensionalization and that the
components vary only modestly with flow coefficient, cp, for values above about 0.03.
Data has also been obtained with 0.79mm seal clearances; it is qualitatively similar
to that of Fig'. 10 except that the magnitude of all of the components is roughly 0.5
larger for the larger clearances. The theoretical results of Colding-Jorgensen (Ref.
8) based on the source/vortex model of the impeller which are also depicted in Fig.lO
exhibit substantial disagreement with the measurements. The diagonal components
and y are different by a factor of 3. The off-diagonal or "cross-coupling"
terms are very different in magnitude ando, in the case of x , the experimental
x
K!
data do not exhibit the change of sign near the design flow coefficient.

Both the hydrodynamic forces presented in Figs. 6, 7 and .s and the hydrodynamic
force matrices in Fig.lO imply that the fluid forces will tend to excite a whirl
motion of the impeller. The force matrix could be presented approximately by the
form

2.0 -0.9
*
[KijJ =

0.9 2.0

at least over the range 0.04 < cp < 0.14. It is therefore a combination of a diagonal
and a skew-symmetric matrix. The former will simply reduce the structural stiffness
matrix and in many cases this hydrodynamic effect will be small. The skew-symmetric
or cross-coupling terms are much more important. Since y and x are of op
posite sign their effect is to tend to destabilize the rotor. The rotordynamic con

sequences of such excitation will, of course,
depend on the damping matrix as well
(see for example Ref.14). However since
and are proportional to
y N2
x
it follows that if the damping increases more slowly with N, then there will always

be a critical speed, N, above which the excitation will exceed the damping. Further
more it is clear that this whirl will not necessarily be synchronous. Future mea
surements using the internal balance will examine the hydrodynamic contribution to
the damping and mass matrices; this will allow more quantitative analysis of hydro
dynamically induced whirl.

Finally, we note thatthe force matrix data implies a "hydrodynamic center"

within the volute; when the impeller center coincides with this "hydrodynamic center"
the force [Fi is zero.
} The position of the "hydrodynamic center" for the Impeller
X, Volute A is presented as a function of flow coefficient in Fig.ll where the origin

387
is the center of the logarithmic spiral volute shape. It is seen that this data is
consistent with that of Domm and Hergt (Ref.3) for a similar volute.

STATIC PRESSURE MEASUREMENTS IN THE VOLUTE

Typical static pressure distributions in the impeller discharge flow (measured

as described earlier) are presented in Fig.12 for various flow coefficients and for
seal clearances of O.79mm. This data is for the impeller position labeled "closest"
in Fig.4. The form and magnitude of the pressure variations are similar to those
measured by Iversen et al (Ref.6) and to those predicted theoretically by Kurokawa
(Ref.lS). The pressure distributions were integrated numerically to obtain static
pressure forces on the impeller for each of the four impeller positions indicated in
Fig.14. Figure 13 presents the X and Y components of these static pressure forces
for one of the four positions and compares them with the total force components on
the impeller measured by the external balance for the same impeller location. Note
that the static pressure component is not sufficient in itself to explain the measured
total force. This is contrary to the result obtained by Iversen et al (Ref.6); they
observed significant agreement between the forces. On the other hand we conclude that
the discrepancy in the forces presented in Fig.13 is due to a difference in net flux
of momentum into and out of the volute. Some measure of the order of magnitude of
this momentum flux difference can be obtained by evaluation of the momentum flux dis
charging from the volute which, when non-dimensionalized in the same manner as the
forces leads to a momentum flux in the Y direction of 4.l42. This is of the same
order of magnitude as the forces. We conclude that the non-isotropy of the momentum
is an important contribution to the impeller forces.

Results similar to Fig.13 were obtained for each of the four positions in Fig.4.
Hence hydrodynamic force matrices or stiffness matrices due to the static pressure
forces could be evaluated. It transpires that this component represents only about
20% of the magnitude of the total force matrices presented in Fig.IO. We therefore
conclude that the non-isotropy of the momentum flux is the primary contributor to the
stiffness matrix. This fact emphasizes the need for direct measurement of the forces
on the impeller.

CONCLUDING REMARKS

This paper has presented measurements of the hydrodynamic forces on a centrif

ugal pump impeller as a function of position within a logarithmic spiral volute.
These correspond to the forces experienced by the impeller at zero whirl frequency.
The hydrodynamic force matrices (negative of the hydrodynamic stiffness matrices)
derived from these measurements exhibit both diagonal and off-diagonal terms of sub
stantial magnitude. The off-diagonal or cross-coupling terms are of a form which
would tend to excite a whirl motion in a rotordynamic analysis of the pump.

Static pressure measurements in the impeller discharge flow show that the hy
drodynamic force on the impeller contains a substantial component due to the non
isotropy of the net momentum flux leaving the impeller. Moreover a similar break
down of the contributions to the stiffness matrices reveals that the major component
of these matrices results from the non-isotropy of the momentum flux.

Future plans include measurement of these forces at non-zero whirl frequencies

in order to obtain a complete picture of the rotordynamic consequences of these
hydrodynamic forces.

388
 REFERENCES

1. Alford, J. S.: Protecting Turbmachinery from Self-excited Rotor Whirl.

J. Eng. Power, vol. 87 A, 1965, pp. 333-344.

2. Thompson, W. E.: Fluid Dynamic Excitation of Centrifugal Compressor Rotor

Vibrations. J. Fluid Engineering, vol. 100, no. 1, 1978, pp. 73-79.

3. $tepanoff, A. J.: Centrifugal and Axial Flow Pumps. Wiley, New York,
2nd Edition, 1957, pp. 116-123.

4. Domm, H.; and Hergt, P.: Radial Forces on Impeller of Volute Casing Pumps.
Flow Research on Blading (L. S. Dzung, ed.), Elsevier, 1970, pp. 305-321.

5. Agostinelli, A.; Nobles, D.; and Mockridge, C. R.: An Experimental In-

vestigation of Radial Thrust in Centrifugal Pumps. Trans, ASME, J. of
Eng. for Power, 1960, pp. 120-126.

e. Iversen, H. W.; Rolling, R. E.; and Carlson, J. J.: VolutePressureDistri-

bution, Radial Force on the Impeller and Volute Mixing Losses of a Radial
Flow Centrifugal Pump. Trans. ASME, J. of Eng. for Power, 1960,
pp. 136-144.

7. Hergt, P.; and Krieger, P.: Radial Forces in Centrifugal Pumps with Guide
Vanes. Proc. Inst. Mech. Engrs., vol. 184, Pt.3N, pp. 101-107 ,1969-70.

. Colding-Jorgensen, J.: The Effect of Fluid Forces on Rotor Stability of

Centrifugal Compressors and Pumps. Ing. Thesis, Technical University
of Denmark, 1979.

. Chamieh, D.; and Acosta, A. J.: Dynamic Forces on a Whirling Centrifugal

Rotor. Proc. 6th Conf. on Fluid Machinery, Aademiai Kiado, Budapest,
Hungary, 1979, pp. 210-219.

10. Chamieh, D.: Calculation of the Stiffness Matrix of an Impeller Eccentri-

cally Located within a Volute. Cavitation and Polyphase Flow Forum,
Joint ASME - ASCE Solid and Fluid Mechanics Conference, Boulder, Colorado,
pp. 51-53, 1981.

11. Imaichi, K.; Tsujimoto, Y.; and Yoshida, Y.: A Two Dimensional Analysis
of the Interaction Effects of Radial Impeller in Volute Casing. IAHR/
AIRH Symposium, Tokyo, Japan, 1980, pp. 635-647.

12. Shoji,H.; nd Ohashi, H.: Fluid Forces on Rotating Centrifugal Impeller

with Whirling Rotors. NASA Conf. Pub. 2133, Texas A & M University,
College Station, Texas, 1980, pp. 317-328.

13. Brennen, C. E.; Acosta, A. J.; and Caughey, T. K.: A Test Program to
Measure Fluid Mechanical Whirl-Excitation Forces in Centrifygal Pumps.
NASA Conference Publication 2133, 1980, pp. 317-328.

389
14. Adams, M. L.; and Padovan, J.: Insights into Linearized Rotor Dynamics.
Journal of Sound and Vibration, 1981, pp. 129-142.

15. Kurokawa, J.: Theoretical Determinations of the Flow Characteristics in

Volutes. IAHR / AIRH Symposium, Tokyo, Japan, 1980, pp. 623-634.

390
 8

o
10

"
W
\.0
I-'

PLAN VIEW ELEVATION VIEW

Figure 1. - Schematic of pump housing and force balance assembly of rotor test
force test facility. Pump housing (1), volute (2), inlet connection (3),
inlet bell (4), impeller (5), internal balance (6), double bearing system
(7,8,11), orbiting motion sprocket (9), main shaft (10), axial retaining
flexure (12), external balance flexure (13). retaining spring (14).
Figure 2. - Photographs of the items shown in Figure 1.

392
 FRONT PRESSURE BACK PRESSURE
TAP TAP

6mm
-.,,,r-----nIP=-+

162mm
DIA.

(oJ
INLET
-

80mm
DIA.

Fi gure 3. - Schematic showing main dimension and static pressure measurement

points within volute. ( Eight pressure taps more or less equally spaced
around volute circumference on front and back. )

t VERTICAL

'-o
'FARTHEST'
POSITION
WHIRL
ORBIT

Figure 4. - Schematic of forces and locations within impeller-volute system

as seen from inlet.

393
 . SEAL CLEARANCES SET

0.5
O
4> Lb 'O
<B. AT 0.14mm
AT 0.79m'"

..:
0. 4
A\
z 0
W
u
u: A'i 0
II.. 0
W
0
u
A
0 N (RPM)
.q:
w
0.3
'"
0
tt
:r: 600 0
A
0
0 1000
800
0 \
0 . 1200
'il 2000 A2-
0.2 0
'\l

o 0.02 0.04 0.06 O.OB 0.10 0.12 0.14

FLOW COEFF)CIENT.

Fi gure 5. - Performance characteristic of Impeller X inside Volute A for

front and back seal clearances of 0.14 and 0.79 mm. (Open and closed
symbols represent data for 0>0 and 0<0 respectively.)

O.IB

-- AGOSTINELLI. NOBLES. 8 MOCKRIOGE

(SPECIFIC SPEED' 0.611
* 0
II.. 0.16 --- IVERSEN. ROLLING. 8 CARLSON
(SPECIFIC SPEEO' 0.361
I-
Z
W N (RPM)
0.14
U
u: A 600
II..
W 0 BOO
0 0 1 000
U 0.1 2
0 . 1 200
w
U '\l 2000
0:
0.10

w
I-
:3 O.OB

w
C)
.q: 0.06
0:
W

0 0.04
W
N
::::i
.q:
::E 0.02
0:
0
Z

0 0.02 0.04 0.06 0.08 0.10 0.12 0.14

FLOW COEFFICIENT.

Fig ure 6. - Normalized average volute force for Impeller X, Volute A. and seal
clearances of 0.14 mm. (Open and closed symbols represent data for 0>0 and
0<0, respectively. Comparison is made with Iversen et al. (ref.6) bearing
reactions and Ag ostinelli et al. (ref.5) experimental data.)

394
 80

60" --X<>
\ v
C
\
40" \
\ t
CD 0
\
. 20 \
I&J V
..J \
(!)
z \
<l
O
I&J
U
0::
Ir -20"

I&J
I N(RPM)
:::I
..J -40
A 600
0 800
I&J C 1 000
-60"
<> 1200
0::
I&J V 2000

AGOSTINELLI. NOBLES. a

1oI0CKRIOGE (SPECIFIC SPEED' 0.61)
-BO'

-100

FLOW

Figure 7. - Direction of average volute force plotted in figure 6 expressed

in volute coordinate system. (6 is the angle between the direction of
average force and line joining center to tongue of volute.)

0.16 V-----"r----'--'-'---'---;

COLDING-JljRGENSEN ().-86.0"I

'0
0.14 TEORY } 00101101 a
H[RGT
()'&86.3'
c:
--
" -oj- Fy } EXPERIMENT
F.
0.1 2
....
II..

UJ 0.10
I-
Z
W
Z
0
lL
::IE
0
u
w
u
n::
II..
0
W
I-
:3
0 0
>
w
(!) * *
<l F F
-0.02
y
N(RPM)
n::
w A 600

-0.04
0 BOO
0 III C 1000
W
N -0.06 <> 1 200 (.!l > 01
:J
<l fII d I 2 00 (.!l <0I
::IE
0:: -O.OB V V 2000
0
Z

0.12
-0.10
0 0.08 0.10
COEFFICIENT. '"

Figure 8. - Comparison of normalized average volute force components of

figures 6 and 7 with work of Domm and Hergt (ref.4) and Colding-Jorgensen

(ref.8) (Impeller X, Volute A. and seal clearances of 0.14 mm).

395
 0.18
'"u."

..:
z 0.16
W
U
ii: "'"o-o,,
LL
W 0.14
0
U

W
U
Il: 0.12

W
I-
::>
..J
0
0.10
IMPELLER
>

W O.OB
C)

Il:
w IMPELLER EXIT BLOCKED
0.06

0
w
N
:J 0.04

::E
Il: SOLID IMPELLER
0
Z
0.02

SHAFT SPEED N, RPM

Figure 9. - Normalized average volute force as function of shaft speed for

various auxiliary experiments. ( Impeller X, Volute A. and seal clearances
at O.14 mm . )

2.0r

I
Z
!:!:! 1.5 r- -

u
ii:
LL
W
o
u
1.0
X
Il:
I

;:;;

0 0
0
L\ / L\
e

<h Kxy -

Q:] Q6 aJ;lo K

I
"0 L\ eo 0 0

- 1. 00--:0:-':. 0::-:::2:--::-
0.-':-
0"" 0-':. 0"'B --=-.0':-: 1 0:---:0..L. 1=-2 --
4--0""' .':-06:----: 0.l4

FLOW COEFFICIENT. t/>

Figure 10. - Force matrix coefficients as defined in text ( Impeller X, Volute

A. and seal clearances at 0.14 mm ), shaft speed = 600 rpm: 6 . ... ; 800 rpm
0,.; 1000 rpm: 0.111. ( Values of K x' K y' and K y x are represented
by open symbols; values of Ky by closed symbols. )
y

396
 0.03 Y -g
TONGUE If

0.02
X
}
W
Voq
0.01
N
a:
...
>- QUADRANT 2
0
W
.... QUADRANT 3 .V QUADRANT 4

6
<[
z
is -0.01
a:
0
0
t)
.O
>-

0
-0.02
.
UI
N
ECREASING
:::i -0.03 FLOW RATE
<[

a: N (RPM)
0
Z
t
-0.04 A 600
0
cV 0 BOO
C 1000
0 . 1200
-0.05
O
V 2000
V
c
-0.06

-0.04 -0.03 -0.02 -0.01 0 0.01 0.02

NORMALIZED X COORDINATE, X/R2

Figure 11. - Equilibrium positions of Impeller X within Volute A for seal

clearances of 0.14 mm for various flow rates. Solid line indicates
experimental data of Domm and Hergt ( ref.4 ) for A = 86. 30,

D.
t)

o
UI
N
:::i
<[
::;:
a:
o
z

....
Z
I1J
(3
ii:
Ia..
I1J
o
U

I1J
II::
::>
VI
VI
I1J
II::
Q.

00 3600
VOLUTE ANGLE STARTING FROM 8
Figure 12. - Pressure coefficient for various flow rates for main shaft
speed N = 600 rpm. Impeller X, Volute A, and seal clearances at 0.79 mm.
Shaft center position is at 'closest' point ( fig.4 ) . Solid lines represent
front pressure taps and dashed line back pressure taps of figure 3.

397
 0.12

IVERSEN. ROLLING. II CARLSON

STATIC PRESSURE VOLUTE FORCE

0.10 '-.a'. s;.

U)
I-
Z 0.08
'"
Z X
0
Q.
:E 0.06
0
U
0
I-
0
Z 0.04 <>
'"
<>
U
ii:: <>
I&.
'" 0.02
0
U

'"
U
a: 0
0
I&.

0
'" -0.02
N
:::i SHAFT SPEO (RPM I
C(
:E 600 800
..

a:
0
Z
-0.04
<> 0 X- DIRECTION } IMPELLER FORCE
Y-DIRECTION

-0.06
0 X- DIRECTION } STATIC PRESSURE
.t. Y -DIRECTION VOLUTE FORCE
\
0 0.02 0.04 0.06 0.08 0.10 0.12 0.14
FLOW COEFFICIENT. f/J

Flgure 13. - Normallzed lmpeller forces and statlc pressure forces of flgure 12
shown for maln shaft speed N = 600 rpm. Impeller X, Volute A, and seal
clearances set at 0.79 mm. Shaft center posltlon ls at Iclosest' polnt
(flg.4). Comparlson is made with Iversen et ale (ref.6) static pressure
volute forces.

398
 GAS 'IURBINE ROIOR/CASE S'IRlCIURAL RESPOirSE 'ID RafATIN3 STALL:

EXPERIMENTAL IXX.:UVIENA
I' TIOO AND ANALYfICAL APPROAQJ*

Philip J. Haley
Detroit Diesel Allison
Division of General Mbtors Cbrp.
Indianapolis, Indiana 46206

SUMMARY

This work presents experimental data describing the forcing functions and
structural responses characterizing gas turbine rotor/case system vibration
due to rotating stall in an axial-flow compressor. Two data sets with funda
mentally different response characteristics are presented; one is supersynch
ronous and the other subsynchronous. Conventional beam-element rotor dynamics
analysis is shown to be severely limited in its ability to predict these re
sponses. A new analytical approach, which significantly increases structural
response predictive capability for these phenomena, is briefly discussed.

INTRODUCTION AND BACKGROUND

The emphasis on aerothermodynamic performance in the high-performance gas

turbine engine of today and tomorrow has focused attention on increased per
formances from its components, one of which is the axial-flow compressor.
Compressor design trends, aimed toward high efficiencies and/ or high aerody
namic loadings, have led to machines that are very sensitive to running tip
clearances between blades/vanes and their respective case/ rotor structures.
The controlling criterion in the selection of these clearances is often the
necessity to accommodate rotor and case excursions associated with surge/stall
instabilities. Closely related to clearance selection is the rotor dynamic
design of a given machine. The mechanical designer typically has only empiri
cal guidelines based on severely limited data with which to design to accommo
date surge/stall deflections and to simultaneously provide minimum running
clearances. This paucity of knowledge of structural dynamic response during
surge/stall instabilities renders the optimization of tip clearances, rotor/
case flexibility, and bearing placement a difficult and abstract task.

Detroit Diesel Allison (DDA) is currently working under Air Fo ce con

tract to experimentally document compressor structural response to the rota
ting stall phenomenon and to improve state-of-the-art analytical techniques
for predicting such response. These efforts address some of the deficiencies
in knowledge noted above. This paper presents an abbreviation of some of the
results of this contractual effort with emphasis on the experimental data.

Rotating stall can be viewed as one or more regions or "cells" of low mo

mentum, low-pressure fluid, each covering some sector of the flow annulus of
the compressor . Although rotating stall typically occurs near the compressor
surge line, some machines exhibit it away from surge, even near the operating
line. The number of cells can vary from one to eight or more. Multiple cells
seem to be uniformly spaced circumferentially. In the fully developed rota
ting stall considered herein, the cells traverse the axial length of the com-

*This work is associated with U.S. Air Force Contract No. F336l5-79-C-2089.

399
pressor, so can be regarded as "tubes" of low-pressure fluid that are axially
twisted or "corkscrewed" through the machine. These stall cells result in net
forces exerted on both the compressor rotor and the case, assuming the cell
fills the flow passage span. These forces are radial and. if ramp angles are
present, can have axial components. The stall pattern rotates in the direc
tion of the compressor rotor but at partial rotor speed, generally between 25%
and 75% speed, as viewed from the inertial frame. Thus a single-cell rotating
stall at 50% rotor speed acts as a subsynchronous excitation mechanism. A
two-cell stall at 50% speed appears as a synchronous force, while a two-cell
stall at 60% speed appears supersynchronous.

In contrast to the surge phenomenon, rotating stall can produce quasi

steady, periodic forcing functions which can cause steady-state, periodic
rotor/case response manifested as structural dynamic instabilities. Such re
sponses can, based on experience, lead to high amplitude vibration and/or
radial excursions beyond the design clearances, causing rub and potential
structural failures. This paper presents a classic set of experimental data
describing the violent nature of rotating stall-induced rotor/case system vi
bration. The associated forcing functions are also presented, and Fourier
analysis is utilized to gain i nsight into the forcing function/response rela
tionships. The new analytical approach developed at DDA to calculate $tall
induced system response is briefly outlined.

EXPERIMENTAL DATA AND DISCUSSION

The data presented herein were generated at DDA on a large, multistage,

axial-flow compressor rig, which is fairly typical of the compressors current
ly found in high-performance, aircraft-type gas turbines. This rig is s ketch
ed in Figure 1. Though not shown in this sketch, the bearing/rotor arrange
ment of the rig is similar to that of the parent gas generator, of which this
compressor is a component. This rig was instrumented to measure the time-var
iant forcing functions and structural responses associated with rotating
stall. The elements of this high-response instrumentation scheme are shown 1n
Figure 2, which depicts the compressor case unwrapped into a plane. These
sensors are all mounted on the compressor case in the stationary frame. The
dynamic pressure sensors are high-response transducers flush-mounted in the
outer gas-path wall. They measure the fluctuating component of endwall pres
sure, which is assumed to characterize the stall cell at both its inner and
outer span. Vibration pickups measure case velocities, while rotor-to-case
relative motion is sensed with "Whip" pickups, which are proximity probes lo
cated over compressor blade tips. Rotor motion can be established by vector
ially adding case motion and rotor-to-case relative motion after transforming
these measured parameters to compatible units.

Running of this compressor rig produced several rotating stall and/or

surge sequences, w hich were recorded on magnetic tape via the described in
strumentation. The phenomenological sequence from which the data presented
herein were quantified consists of a high-speed surge followed by two-zone and
then single-zone rotating stall patterns. Figure 3 is a time record of the
rotor speed excursion showing the onset of surge and the subsequent initiation
of the stall patterns. As seen in this figure, the entire sequence of time
variant phenomena lasted approximately 5 seconds, during which time the rotor
accumulated over 900 revolutions under the influence of the related rotating
forces. The data presented in the subsequent discussion were defined over the
two time intervals shown in Figure 3.

400
 Fo llowing surge, a two-zone rotating stall pattern was spontaneously es
tablished, traveling (one-per-circumference spatial propagation rate) at 57%
rotor speed. Time traces of the dynamic signals over the defined analysis
interval are presented in Figures 4, 5, and 6. A common timing signal on
these plots is used to time-relate events. The responses shown in Figures 5
and 6 show high-amplitude compressor case structural vibration at the fre
quency of the individual stall cell passage from Figure 4, i.e. at two times
the two-zone rotating stall rotational frequency. The whip signals of Figure
6 also indicate rotor-to-case relative motion at the same frequency. These
responses and the associated forcing functions thus appear to be supersynch
ronous.

The two-zone rotating stall pattern evolved into a single-zone pattern

(Figure 3) with a propagation rate of 60% rotor speed. Figures 7, 8, and 9
present the traces of dynamic pressure, case vibration, and whip for this
stall occurrence. Again, high amplitude structural response is observed.
Both the compressor case response and the whip signals visually indicate major
components at the subsynchronous passing frequency of the Figure 7 forcing
functions. Clearly evident in some of the response signals are additional
higher frequency components.

Insight into the nature and relationships of the forcing functions and
structural responses characterizing these stall phenomena is afforded by Four
ier analysis. The two-zone stall data of Figures 4 through 6 were digitized
over the common interval shown on Figure 4, and the single-zone data of Fig
ures 7 through 9 were likewise digitized over the period shown on Figure 7.
Digitized data were harmonically analyzed to produce components for each sig
nal according to the cosine Fourier form:
00

Um cos (m fit - m()+ am )

m=O

where:

Um amplitude of mth harmonic (harmonic coefficient)

m harmonic index
n = fundamental frequency
() = circumferential location of sensor
a phase of mth harmonic
m
t = time

This procedure removes phase differences arising from the different circumfer
ential locations of the various sensors. Note that in Figures 4 through 9,
representing fully developed stall patterns, all these dynamic signals, con
sidered to represent transient phenomena in the sense of an operating steady
state turbomachine, are relatively periodic functions.

The Fourier content of the pressure traces is presented in Tables I and II

for the two-zone and one-zone phenomena, respectively. Because eight one-per
circumference spatial cycles were digitized (Figures 4 and 7), the first five
spatial harmonics of the waveforms correspond analytically to temporal harmon
ics 8, 16, 24, 32, and 40, as listed. A comparison of Table I with Figure 4
for two-zone stall confirms the predominance of second harmonic content (first

401
harmonic of a single stall cell times two cells per circumference} and indi
cates a strong fourth harmonic. Comparison of Table II with Figure 7 for one
zone stall shows a predominant first harmonic content with significant higher
harmonic content, notably second harmonic.

Table III presents the Fourier content of the measured responses to the
two-zone rotating stall pattern. The first five spatial harmonics are listed.
The amplitude predominance of second harmonic is seen, followed by fourth
(first harmonic of second). This pattern corresponds to that noted in the
pressure traces. The highest amplitude forcing functions (pressures) from
Table I are seen to be axially toward the rear of the compressor, whereas the
highest amplitude case responses from Table III are seen to be in the compres
sor inlet area, suggesting a proximity to a resonance in the inlet. Compari
son of these two tables does clearly show, however, direct structural response
to rotating stall. The presence and predominance of second and higher harmon
ic case response show the importance of localized distortion and "egging" be
havior in a flexible gas turbine case. Beam-like behavior, corresponding to
first harmonic force and response components, is a relatively minor contrib
utor to case response in these data. Figure 10 further clarifies these de
scriptions.

Table IV presents the Fourier content of the measured responses to single

zone stall. The amplitude predominance of first harmonic is apparent, along
with a strong second harmonic contribution, especially in the inlet. This
same pattern is seen in the rotor-tocase whip signal. The relative Fourier
contents of these response signals are comparable to those of the forcing
functions (Table II), again suggesting the approach of comparing forces and
responses on a harmonic basis. These data indicate the primary case response
to be beam-like, unlike that observed in the two-zone phenomenon.

As shown in Figure 2, axially-redundant signals were obtained for several

parameters. Examination of the Fourier content of these signals in Tables III
and IV reveals circumferentially asymmetric responses throughout this compres
sor in both first and second harmonic components for both case and rotor-to
case signals. The significance of these observations is that the responses of
this "axisymmetric" turbomachine to these stall-induced forcing functions are
not at all axisymmetric. The asymmetric case response in the first harmonic
indicates asymmetric mounting of the case to ground for the beam-like motion.
Asymmetric case response in the second and higher harmonics is due to asym
metry in the case itself, caused by holes, split lines, bleed manifolds, and
other deviations from pure axisymmetry.

A conclusion from these experimental data is that stall-induced response

of the compressor case, an important contributor to system dynamic response
and to rotor-to-case clearance response, cannot always be adequately described
by first harmonic, beam-like motion. Furthermore, significant circumferential
asymmetry is present in the different modes of case response. Obvious from
these observations is the fact that the beam-element analytical ap proach wide
ly used 'in rotor dynamic analyses is generally inadequate for modeling either
of the forced response problems described herein. Since the supersynchronous
two-zone stall is seen to have only minor components of forcing functions and
responses that can be described by the first spatial harmonic, the beam-ele
ment approach is inherently grossly inadequate. Although forcing functions
and responses of the subsynchronous single-zone stall have predominant first
harmonic content, the higher harmonic components are not negligible but are
ignored in beam-element analysis.

402
 NEW ANALYTICAL APPROACH

DDA has developed a capability for predicting rotor/case structural re

sponse to dynamic forcing functions. This analysis has been developed specif
ically for the rotating stall phenomenon in compressors. The analysis in
cludes the ability to predict localized motions of the static structure (i.e.,
compressor case), including "egging" modes. The approach entails using a
beam-element rotor analysis along with an axisymmetric finite-element case
analysis. Egging and localized motions are calculated for the case alone and
are assumed to be decoup1ed from the rotor. Beam-like motion is calculated
for the coupled rotor/case system using finite-element case modes modally
coupled to the conventional beam-element rotor model. Deviations from pure
axisymmetry are recognized: (1) in the beamre1ement calculation, through a1-
. 10wance for asymmetric support modeling, and (2) in the finite-element case
analysis, through availability of a discontinuous element for modeling struts,
vanes, holes in the case, bleed manifolds, and other variances from pure axi
symmetry. This technique has substantially addressed the shortcomings of con
ventional beam-element rotor dynamics analysis in the modeling of stall-in
duced response and has produced reasonable analytical simulations of both the
subsynchronous and supersynchronous preceding experimental data sets.

403
 T ABLE 1. - FOURIER DECOMPOSITON OF DYNAMIC PRESSURES FROM TWO-ZONE STALL

Stase

2
1: 1 1 2 1. .!! 2 10

Temporal harmonic - 8

Amplitude 0.2143 0.0346 0.0613 0.0670 0.0562 0.0475 0.1012 0.0924 0.2463
Phase -103.1 -75.9 9 7.0 -144.7 135.5 -149.5 -174.5 60.7 77.7

Temporal harmonic - 16

Amplitude 0.1050 0.1597 0.19 6 1 0.9671 1.784 2.711 2.622 3.878 2.340
Phase -1.5 -122.6 l8d -160.2 -176.8 -175.6 159.5 136.4 106.1

Temporal harmonic - 24

Amplitude 0.0498 0.0334 0.0260 0.0513 0.0620 0.0561 0.0688 0.0974 0.1809
Phase 47.3 18.4 115.3 -74.0 140.6 153.8 166.4 88.4 61.6

Temporal harmonic 32

Amplitude 0.1022 0.1339 0.1023 0.1286 0.3121 0.5181 0.4338 0.8993 0.1992
Phase -22.9 -24.8 -28.0 -164.8 -171.1 -154.2 141.5 102.6 117.0

Temporal harmonic - 40

Amplitude 0.0380 0.0026 0.0001 0.0311 0.0495 0.0188 0.0526 0.0914 0.0211
Phase 0.0 -180.0 -180.0 0.0 -180.0 0.0 0.0 0.0 0.0

Amplitude in psi.
Phase in degrees.

TABLE II. - FOURIER DECOMPOSITION OF DYNAMIC PRESSURES FROM SINGLE-lONE STALL

Stage

1: 2 1.

Temporal harmonic - 8

Amplitude 0.2323 0.2519 0.3808 1.041 1.576 2.222 2.122 2.139 0.7372
Phase 17.3 149.3 176.0 -157.3 -172.7 -173.5 177.1 155.6 128.1

Temporal harmonic - 16

Amplitude 0.1429 0.0926 0.1158 0.4098 0.7246 1.111 0.9731 0.9519 0.4192
Phase -7.6 38.3 -108.3 -134.4 -147.7 -155.3 179.7 166.4 108.8

Temporal harmonic - 24

Amplitude 0.1128 0.0888 0.1165 0.1704 0.3000 0.4062 0.3069 0.2796 0.0043
Phase -26.2 7.0 -47.2 -146.3 -145.2 -150.8 -164.6 -127.8 169.1

Temporal harmonic - 32

Amplitude 0.0399 0.0845 0.0883 0.0797 0.1446 0.1212 0.0718 0.1437 0.1235
Phase -32.8 32.1 -32.6 -124.5 -143.3 176.6 -40.8 -30.1 127.2

Temporal harmonic - 40

Amplitude 0.0085 0.0129 0.0245 0.0408 0.0491 0.0655 0.0659 0.1723 0.0220
Phase 0.0 0.0 0.0 -180.0 -180.0 -180.0 0.0 0.0 0.0

Amplitude in psi.
Phase in degrees.

404
 T ABLE I I 1. - FOURIER COMPONENTS OF EXPERIMENTAL RESPONSE FOR TWO-ZONE ROTATING STALL

Ref. Figures 5 & 6

Signal Description 1 1 ....!oJ2 2 ..E3 3 ..E.4 .!!:.4.. 5 5

Vibration Inlet hsg., 00 0.0441 26 2.286 -28 0.1 124 -226 0.1953 -170 0.0276 0
Vibration Inlet hsg., 2700 0.0706 -174 2.499 -182 0.0835 18 0.1082 -178 0.0165 0
Vibration Stg. 7 vane, 2200 0.0265 -200 0.6071 -269 0.0141 -105 0.0635 -243 0.0118 -180
Vibration Stg. 8 vane, 1500 0.0453 -187 0.2829 17 0.0471 -240 0.0682 -137 0.0206 0
Vibration Stg. 9 vane, 1730 0.0448 -231 0.7128 74 0.0060 -141 0.0275 2 0.0006 -179
Vibration Stg. 9 vane, 3550 0.Ol l8 2 0.6265 63 0.0129 -Ill 0.0171 - 45 0.0035 -180
Vibration Rear camp", 0 0.0447 6 0.5282 68 0.0171 16 0.0329 -193 0.0129 0
Vibration Rear comp., 1500 0.0241. -191 0.2541 -4 0.0065 49 0.0441 -268 0.0094 -180
Vibra tion Rear comp., 2400 0.0271 -148 0.8876 -209 0.0218 72 0.lll2 -96 0.0094 -180
Vibration Rear comp., 2700 0.0276 --63 0.4035 -108 0.0100 22 0.0429 62 0.0035 -180
Whip Stg. 9, 2400 0.0014 148 0.0023 201 0.0006 113 0.0002 88 0.0002 -180
Whip Stg. 9, 1500 0.0007 77 0.0018 46 0.0006 225 0.0011 191 0.0003 0

Um = amplitude (harmonic coefficient); vibration in in./sec., avg; whip in in., s ingle amplitude.
a'll = phase angle, in degrees.
m = harmonic index.

o
In
TABLE IV. - FOURIER COMPONENTS OF EXPERIMENTAL RESPONSE FOR SINGLE-lONE ROTATING STALL

Ref. Figures 8 & 9

Signal Description 1 l 2 2 ....!oJ3 3 4 .!!:.4.. ..E 5 5

Vibration Inlet hsg., 00 1.059 63 0.9482 45 0.0553 -257 .0.0676 -62 0.0276 0
Vibration Inlet hsg., 2700 0.6907 -41 0.9385 -133 0.1021 75 0.0388 -58 0.0137 -180
Vi bration Stg. 7 vane, 2200 0.8600 -77 0.3906 -221 0.2041 -150 0.0512 -200 0.0165 0
Vibrat ion Stg. 8 vane, 1500 0.6681 -104 0.3110 25 0.0269 -182 0.0525 -14 0.0006 -179
Vibration Stg. 9 vane, 1730 1.038 -107 0.2897 80 0.1545 -181 0.0333 -19 0.0200 0
Vibrat ion Stg. 9 vane, 3550 0.9588 66 0.4257 -232 0.ll52 -89 0.0281 -43 0.0072 -180
Vibration Rear camp", 0 0.9194 40 0.3806 -240 0.0576 -1l0 0.0194 -28 0.0109 -180
Vibration Rear comp., 1500 0.6627 -135 0.2436 -5 0.0579 -184 0.0657 -44 0.0161 -180
Vi bration Rear comp., 2400 1.485 -88 0.3059 -133 0.0376 -108 0.1682 -225 0.0171 -180
Vibrat ion Rear comp., 2700 0.5379 -30 0.2448 -48 0.1325 -23 0.0287 -107 0.0096 -180
Whip Stg. 9, 2400 0.OU2 -36 0.0029 -70 0.0011 -100 0.0005 244 0.0001 -180
Whip Stg. 9, 1500 0.0164 -157 0.0016 61 0.0003 137 0.0006 59 0.0003 0

U amplitude (harmonic coefficient); vibration in in./sec., avg; whip in in., single amplitude.
m
am phase angle, in degrees.
m harmonic index.
 Diffuser assembly

Plenum r Rear compressor

Pineapple
Inlet
llnlet bell
\:housing

0 (top)
OGV

I MJt-L---.i' L.r-rrrff
- - ,
I
-
" \
I I
I

r- ----,
I I
I
I ./
lOt I - --
1800
9th
Looking forward
8th
7th 3rd
6th 4th
5th
Compressor stages
Front mount Sealing
assembl y boot
TE-6657

Figure 1. - Compressor rig on test stand.

III Dynami c pressure

o Vibration
'\1 Whip

Angular position (clockwise)

UNCLASSIFIED
TE-6714

Figure 2. - High-response instrumentation scheme.

406
 13000

12000
Digitized interval

.....
E
c..

I
11000
2-zone stall
initiation
Digitized interval

.....
-0
Q)
Q)
c..
VI 10000
0 I-zone staII
15 initiation
0:::

9000

-1 r-
End stall zones
1S
8000

0:09 0:10
11:50 .
Time--...
TE82-2060

Figure 3. - Sequence of phenomena following surge.

407
 -

Period
-

Period
- -

initiation -encr
----
I - - -;/-
- 5- PSI-'- -
.2 Zone stall
- -
-

presu!:._5: f
_

1st -stage - - - .-
- - .
-- _ - _ HH_ _ . . H n_ __ _

K:Jr"',,;.,l .,'""'- . -

tage p ressure 268; ---- -- -- -- ----- - ---:-=:=:.--:---=:==::-:

_

e/<-9'.r>1"",.,.. f"-'" 'R"vv-.;,

---
.
--

,, "'",...
-

-- \'r:;::7'..;;:tv<V
-

3rd-stage pressure 268:: 5 PSI :._:-:::'._ =
__ .::: _ _ __ ' :. .::: -.:__ ==.::: __ =.::.:-.:=--===
-

==-===-
--

----- ==-:- ---====- - .::. e-===----- - -- -

-----=-.=-=---= .::.:--- - - -
-
-
- - -----
-
- - ------ --
--
-

- - -
-------- .--- -
-----_.----- - ---.-
- --- - .,---- -
- -- ----- -----
.. _-- - - ._ _... -
._. __
----------- --- --,-" ---_._ . - ._---- -------_.- -----_. ------_._----- - -
-- -
-
--------_.-------.-
--_._-- -_._---- ----.--- - - -
.--- .--- ------ -
----_. - -
--- ------------- ---
--- -
-------
------- - --

2640 r=---::-b
_

5 .:.---===.:-.:--==-=-=--=-=====--==
-

-.==: SI :=== ==.=-=

- -----

age pressu _
---

-
-
.

--

---

_
.

.p
o
00

::==-== lOth-stage pressure 258 : .::-:=

- - -.-
_ _ ___ _
-
--
--
----------' -_ .. . --- - - -- -

---
-- ---

259--------
-
-

-
=--f_6:(ff-s;:=rFoool sec-:---- -----p
.----.-.---.-- - _
----- .-.--
----_._
---.--
..

2.258-
----
-

11 sO;
..__ . _-----

-

- V pressure
-- OG
.-

..
--
-

. .
--
-
-.-

' !fJ.II ...

,., ,.. t,r,,f . ,.!; '\-"
---

... .... tf., " - '.: .'\.ft.

M #.f'
---

'" .
.. ..... .& ...

I
.
! . -- " .

Time code UNCLASSIFIED TE-6659A

Fig-yre 4. - Pressure oate durhg tW9-ZltJlle stall.

 .... - _.- _. -

o
\0

__ '!!. .. ,;. ___ _ "-' ___.:._."---L_.i._. "{y_!.:! ::... .::i-:!-_ __'".!_._ . _ ..

=--=-=::Rear bearing support 2830 '

,----_ .. _-- -----
.-._- ----
---_.-._--

-;;[-=mi;hy93-?-\,,12 .2 . , I

r

... U .

1'<--0.01 tn0001
,.
sec-:'1
i . . -

sec
. ... .... .

\
-

,
-

fi.r
I I ,
... .'" '\ '1 t. "
.
r' I: " A
IJIM\/\!' ..
! I fI: f': r ", I " :. t, " . .1 II ! 1\ Ii :'
.'\ .
,; " f.
!/I: \ ., :\1\:\,I\(IJ 1\ 1\1
':

,/\/'.,'11';(',( (\!.. f\.f,.!vlr ., !','\ ,. .
;.: \ A!':!\!'!
,.
,. i i I.! I. :.
.
. ,. !\ ,,\/\/\(, i./\!,.
. ," '" '\ (\/"/\/,, I, i' , \ .".1\/\ I\.I\AA;\ I. '" , .. . I .,
. '
,I
.

UNCLASSIFIED
Time code TE-6660

Figure 5. - Vibration induced by two-zone stall.

.J::'
I-'
o

=
fA
, I
=

/ '.'
9th-stage whi p 1500

JVoNV'J' J'v,J
A

.
.
, ;v1'\v'J,'NNv;
J J
.
_-_-_ ..L' _____---

'i J
-

( .'. I:V1'\VItV'J'V'v'v
. A ' ,' .'. '. :. ,\ ,
'i ', ' ":j VI)
, , v v
1I

::'
'0"

-\
" 'I':
,,
:
'"" '-'"
'

16" '.
. I
.OlC-.'ff' ..
.

"#"'....
! .
.
,

..
.. . . ....

"
.
t
V
l
\ ,
A
' I
, , , ,\
vv'\Nv"M'/ " ,,,
'
"
V v v v v
': .' ;\ '1\
v
'
\ I,
1
A.'."' t. 1\
VV" ' "
V V V v
.
'
'
, , 1\ " ,
Ii 1/\/\/\/\,' vv
v Ii

Time code .,',: , I. ' .

'

'.L., : ',f'
. '
UNCLASSIFIEI)
- .- ,. -'' . .

IE.-6661

Figure 6. - Vibration induced by two-zone stall.

 =---::..--=--==-=====-

Period Period
initi ation end
.
; _ __

i< - - l 1st-stage pressure 2350 t .

_ S r-Sl _
-- -

i';-'\\.--
-, -
'-'
-
,.
...

. - _
.
i-= _- +----- - . - ._-
_.. . -._ ._-_._------- -
. ... . . . _ - - ---- . .- . -.--- - ----_ . .

2nd-stage pressure 2680t--:-..:--:-:!':_:_ : --:.- -- . -,-;,4

:-- ==-3P5r===--=-===-= 169.49 Hz rotational freq.
i:=-: :------ ----- -- - -
.

:--'-- 3rd-stage pressure 268

o

-
-- - ------- _..:surge observed at 11:50:09.11 :
.:-=- - -
--:-: -=-::::
---
__ _____ _ ______. _----

.... -.:;;,----:: "'

._ - . _-------- _ _
- --

-
-;..
- -
-

--
--
-
-- - -
- _
--
--- .

;-
- . --.
-- -_

t====-- i-k ---';';\F'",M/II"I\

-- .
- -

-
-

-
-
- - .
---- -- ----

- II_"i;;;_-c_
'=:
-
- ... -.-.-----.-------
c..-
---- r_ -
-
AM_ - -- un-.
'"'\::
-

-"I;
--+ --
-
-1 ' -
- -
3
-
--
-

- "" '" :7""'.

===E=====-- -=- =-- -
-
=-- :l=
- - -



- -
- - - - --
--
--:-=--===- ---
- =::?\i
-
'7"" ":\;:::
:::-
.!:'
7th-stage p ressure
.
-- - - -
. _._.
_
-------_ .

2640 Et _i
-

l". - .51''---
--

.....
-
-
.....
- :S
_

:=F!' --
--.
. --:::-:. .t. 1Tq;J
___ - :'::4--::'=;
+----nYE. --=-
. - --.----=-
=::
- .

.
.--

.
----

--
-
.

-
- . Analysis period
-
-
-
-
---------.-.._--

-
- - . . , -,
.
- ----------- - --------- . _ --- "A --------
....- _... .--
.. ---
--'---

-:?\;/(>?2/2.
.
" -- - --
- . -
- .
------ -
. - ---

ioth-stage
-- ----
-

;-:
pr e ure 2 58_- BI51
3:75
.
OGV press;;; -259 o-i
.

t; . ls
. II:S": --
- . .
.. . -

, -.AI?: .>__ A J. . -
......;... j,.
\ .\....
." .l'
.

. " " . ., . , .tI,

1 , -'/ .. f "!rr.1iM\F:.-f'"
. ' , \,,'i'f..'
, \!\I\'1 mil7'i:1iM" \i\7\'W'
.. . . .

,,\ I
--
UNCLASSIFIED
TE-6662A

Figure 7. - Pressure data during single-zone stall.

- - -

---
 -=--= Inlet .::

.--::E;.3ifrir
'. iJ!tLfJ&19,! --------
___ -=== PiA vfbratioat 2650
-::-::::::::=== :::::-7'""c__--::=.==_2
=b

9 . 49 ,otot'onol f'e

.

16 Hz =

t bell 90 -==-- - ---

- ""- -
t f . ==-- In'e
0-=-==-- - - _ , -

;;;
=
j
;

+:
,n="
-
: ''&;
$ '
'71
I-'

-
: i
"'
'

l
N
. " , = ,,
, ""
". " .,,",'v ,

; i {
. S" ge obse,ved:"t j F
; 9
=- =="
Inl:c o u s
= :z =c- _ ".Z\- cc ;=--=
T-! 1 , ' lfI 50

-==: . "

J
. , _ = , C' -<
.u<nt
; : . rt =- J
.7 -
:8 "\ i M
...

-=A
:. " - -
_
__ = -; ; ' e
_ i n S" ppo
72 9 = .
- : .'
28 3
'" - --- - '
-
-n 'C' ; "
_-l=
.

=_=__-TI_ 'Ii'- -'; ..-... -o.:-:..-====-- .

l
..
...
-1' f'JJf!,/\f,'
i"
,,!
I,
J\/I(\ f/\/\/i" :
,'f>('.(\(1.!\!,f' i "
A!" '/A
1\ (\/\(I/\I\J' f, , . " :,
L, . ,
,'\i\/\!\!\,A! i', ,'\
.
/\f\!\/\/\./\!i:
''' A "A A"
,. d",\I\:\(\,,,/i ; I ; I \! ,1\1\1\,,,1\/, J\ " i \
!
/\,1\"1\,":'\(\
" .,.:" ,j
, :,. " I

Time code
75 :"/ ,1'/
,71, _ .77 , 7t

UNCLASSIFIED
'
TE-6663

Figure 8. - Vibration induced by single-zone stall.

 -':-_-- _
.:.:.:.:-------
-j-N;'---8th-stage .-
vane
-------- --.----.--:::----=. =-=-==-, -....
- -- ----:-.i ... :w1iS S!AI.l. ::--- -.--

=,;:fl:? __ =-- \tY:':f;;iil.=-==:l--:: -======l=1;:/!1=---i-rt!Etd__1-=-==: --:

.f:'
I-'
-
:;::::r===
Rear compressor 2700 --- --== _

;:-:
W
----
.

;
=;,;;; s""" s. E"'/
-

'W. '"'V\J'iT ==:;:: '
,9th-stage _:vane 3550
,-- ,'. ___- ___
------ -

--

'II
11'
-: ----- -
S,=:=' -,,,-- -,!I'-":-- (,' !,,', )1,, "";/./
-=
9h h' 2400 - . 1.
.;-J' I'\.
i j
_

:::=-:=,"'===-=cc-.c_-c-.-e=-=----==
::=:cc

:;-::::::-= =:--;.--:
-=:;-: :: -9;ht- hip--l50-:' _I;"':"";';""" ''''':'':'.:''' ' EO,Ol.ec-tF O. OOlsecc=,'C:
--'(1' '-",>
Time code !I\'>ii :", !'-.. ,, UNCLAS$ I F I EO
: ,.t :. .
_

>
._; _ . r
TE-6664
__ ...

J
'

-, ".> '. ' -

--- ...
- .

, -

Figure 9. - Vibration induced by single-zone stall.

Beam-like behavior A rrrms indicate
induced motion

Egging motion

local distortion

UNCLASS I F I ED

TE -6745

Figure 10. - Case motion descriptions.

414
 INVESTIGATICN OF OEEZE-FIIM LWV.lPEHS

IN FLEXIBLE SUPPCRf STRlCTURES

R. Ho lITes and Iv1. Dogan

University of Sussex
Brighton, England HNL 9QT

SUMMARY

Squeeze-film dampers are a means of curing instabilities in rotating-shaft assem

blies. Their efficiency depends very much on the condition of the oil, which in turn
depends on inlet and outlet arrangements, on damper geometry and on the flexibility
of the rotor and .surrounding structure. The work discussed in this paper concerns
rig investiga tions in which structural flexibility is included experimentally and
comparisons are made between measured and predicted results.

INTRODUCTION

In two previous papers (refs. 1 and 2), the performance of a s queeze-film damper
unassisted by any retainer spring was examined when interposed between a rigid rotor
and it s rig id bearing pedestals. In many gas turbine applications, however, both
rotor and pedestal s are flexible, the latter to the extent that the first two rotor
pedestal critical speeds are essentially bounce modes, in which the rotor does not
bend to any significant degree. Instead, on increase in speed the rotor often
vibrates first in a symmetric and then in an anti-symmetric mode, while the pedestals
are the only elements to show any appreciable degree of dynamic de flection. For such
systems a squeeze-film damper interposed between one or more of the rolling-element
bearings of the rotor and the pedestals can contribute a degree of damping which may
enable the passage of the system through such critical speeds without vibration
becoming excessive, and inhibit possible rotational instability.

It is not uncommon in gas turbines for the antisymmetric mode of vibration to ex

hibit a node near one of th e rolling-element bearings. With such an application in
mind, a test rig (fig.1) was designed and built, which afforded the investigation of
the squeeze-film damper 1 at one of its two roll i ng element bearings 2;
- the other
bearing 3 being of the self-aligning variety. This constituted a pivot about which
an antisymmetric (i.e. conical) mode of vibration would occur, when the rotor 4 was
acted upon by a force arising from rotation of the unbalance mass 5 . The test rig
was provided with flexible bars 6, which simulated the pedestal flexibility of an
actual engine. Alternati vey, the ]mousing 7 could be clamped by the special clamps
8 tv afford investigation of the squeeze-film damper alone. A heavy foundation
block 9 was provided which represented ground and into which the flexible bars were
screwed. Oil of 21 cp viscosity was supplied to the damper via t hree supply hole s
10 and a central circumferential groove (fig.2). End plates were attached at the

* Performed under contract to Rolls Royce Limited, England

415


c:n

.....
III
OJ
.....

OJ
Co
E
'"
o

416
 EN D PLATE

UNBALANCE MASS

Figure 2. - Detail of damper.

ends of the outer element of the damper to afford s ome sealing, the extent of whic h
could be varied b y the insertion o f spacing shims. The squeeze-film dimensions were
set by the outer diameter of the rolling bearing (136 mm). the damper land length
(9 rom) and its radial clearance (.216 mm).

Proximity vibration pickups were used to measure the vibration of the shaft
relative to the pedestal and relative to ground. A pressure t ransducer was placed
at a mid-land position at the base of the squeeze film to measure dynamic pressure in
the oil under operating conditions and a thermocouple was also provided there to
record oil-film temperature. The presence of the flexible bar s allowed conven ient
recording of transmitted f orce by the provision of strain gauges.

This paper describes investigations into the performance of the dampers and a
comparison of experimental findings with numerical predictions. For the latter, the
numerical method developed in reference (2) was used. T his was based on t he short
bearing approximation appl ied to the Reynolds equation to describe the pressure field
in the squeeze film and required the clearance between the end plates and the inner
member of the damper to be w ide enough to enable the assumption of atmospheric press
ure at the ends of the damper to be used.

In reference (2) it was shown that the independent parameters of the squeeze-film
damper could be presented in terms of three non-dimensional groups, namely

2
Q = P/mcw

2
Q = P /mcW
c c

417
and

in which P is the static load, usually provided by gravity, which for one land of the
damper was 147 N, m is the effective mass of the rotor at each land (12.25 k g), c is
th e radial clearance in the damper (0.216 mm), R is the radius of the inner member of
the damper (68 mm), i the damper land length (9 mm) and n the oil viscosity (21 cp).
w is the rotational speed of the s haft carrying the unbalance mass, which provides
the required dynamic force, P , causing vibration. An essential requirement in the
c
numerical computations is the specification of a cavitation pressure below which the
pressure field is curtailed. T his pressure was found experimentally.

TESTS WITH CLAMPED HOUSING

Firstly a series of tests was carried out with the housing clamped and details of
these are given in Table 1.

TABLE 1. - TESTS WITH HOUSING CLAMPED

Q Rotor speed rev/min

c

0.229 4000 4500 5 000

0.642 3000 3400 3750

A s the value of Q was increased the top speed of the rig was successively re

duced in order to avo d excessive vibration of the rolling-element bearing in the
damper clearance space. The value of speed dictated the corresponding values of the
non-dimensional groups Q and S and these are given with the appropriate experimental
recordings and numerical pred ictions shown later. An oil supply pressure of
2 (5 Ibf/in2
34.48 KN/m ) was used th roughout these tests to avoid excessive outflow
from the ends of the damper through the rather wide end-plate clearances of 1.25 mm .

A set of experimental vibration orbits and pressure recording s is given in figure

3a for Qc = 0.229. Fig. 3b shows the corresponding numerical predictions. Each
numerical hydro-dynamic pressure d ist ribution was curtailed at its experime ntally
observed,neKative pressure li.mi.t before integrati.on to obtain the squeeze-film forces
which acted on the rotor to produce the orbit of vibration. Figures 4a and b show
similar comparisons for a higher value of Q of 0.642, and are typical of many others.
c

Som e significant general observations can be made from these comparisons. In

figure 3a, the general sizes and disposition s o f the vibration orbits are well pre
dicted, while both the experimen tal and numerical recording s show a reversal in
position of the s harp 'tail' between 4500 and 5000 rev/min. The pressure waveforms
also show good agreement in respect of general shape and height. The orbits of
figure 4a are quite well predicted in general shape, size and disposition (even to
the extent of showing a sudden in crease in size between 3700 and 3750 rev/min).
Whilst the experimental pressure waveform at 3000 rev/min is fairly well predicted,
the waveform at 3400 rev/min does not show such good agreement in that numerical com
putations indicate a sharp positive pressure spike, which is not reproduced experi
mentally. It seems probable that the slightly s hallower slopes of the experimental
vibration orbits do not give rise to such pressures, which are very sensitive to
orbit shape. An inspection of the second e xperimen tal pre ssure peak in a given cycle
will, however, indi cate a good comparison with the numerical prediction.

418
 4000 rev/mln
4500 rev/min 5000 rev/min
Q 0.316
Q 0.25
Q 0.2
i3 = 0.02

f-'
i3 = 0.018 B 0.016
<,,0

r (a - E pe imen a a d es u ec d 9 m 2 psi 876 kl mL). &22.:0

 X10-1 Xle-1
1

" I;" ', !l"1 '" e oIl, '" ! ,0

!
I I' 0- 0

.l::
N
X10'
1 4000 rev /min X101 4500 rev/ min X102 . 5000 rev/min
o
25_ ps ig 30 Psig 4 psig
1 ;t
25
20
3
20
15.
15 2_
'0
10

0_ "
0-1 .J .. , ..... i ,. '. 0' I .
22 26 30 32 34 36 22 24 26 28 30 32 34 36 22 24 26 28 30 32 31
wt wt wt

orbits and pressures. Qc 0.229.

Figure 3(b). - Numerical
=
 3000 rev/min 3400 rev/min reduced scaling
Q 0.562 Q 0.44
=
P
=
3750 rev/min
I'.) S, =
0.027 8 0.024 Q .. 0.36
=
I-'

S = 0.0215

Figure 4(a). - Experimental orbits and pressure recordings (1 em = 127 psi = 876 kN/m2 ). Qe = 0.642.
 Xt0-t
t
8

1-- - 1'l1 - 1 k-, :;-6 , '; t, I ? 9, 1 113

I"
-.
1'1'
I , ':
.. ' ,
, - - 'E

-8
/
3000 rev / min 3400 rev/min X102
5
+:-
3750 rev/min
N
X:82 X102 10, psig
N
psig 10 psig
1 8

-I 8

I II 6.
J.., 6.

2J II I\ II I\ 41 II II 4-

-1 _
t),u.. \ lV, L
<2 24 26 28 30 32 34
w
,
t36.
:' ,,im=r ," "", i,=,
-2.
: 2 24 26 28 30 3 34
w
t 36
2
0. , - -
22 24 26 28 j0 32 31
uJt

Figure 4(b). - Numerical orbits and pressures. Qc 0.642.

 3500 rev/min 3800 rev/min 4200 rev /min 4500 rev/min

Q 0'413
= Q O 35
= Q 0- 287
= Q = O 25
N 13 o 023 B 0 - 0 21 2 S'''' 0 - 0192 13=0-018
=
=

Figure 5(a). - Experimental orbits and pressure recordings (1 cm 181 psi 1251 kN/m2 ). Qc 0.229.
 -
xle-I
I"
e
XIS-
l X19-
1,

X10-
1
e
1

5 6.

2.1
1 1
I:J-I
,0
2.
0 : ? '!? Ie
I - -5 - I
.0 2. S. l!l " -. ,-, , Il
I
; r , .IlI; ,

-2
-1
-6
ruI
-G !
- I

3,500 rev !mi n 3,800 rev /min 4,200 rev/min 4,500 rev/ mi n
Xf02 XHl2 Xle2
5 psig Xle2 5_psig 5,psig
' 1 6"psig
N
+1
I n\
3! I
t
21

I
I
0-1 z-: ./ \ wi; 1317""'. , , ..".;' , ', "*
:?
-1. .., 1 .: wt
-2.

Figure 5(b). - Numerical orbits and pressures. Qc = 0.229.

 3500rev 1m in 3800 rev Imin 4200 rev/min 4500rev/min
.p.
N
orbi ts
I..n

loads

Figure 6(a). - Experimental orbits relative to ground and polar load diagrams. 0.229.
Qe
(1 em = 0.5 mm and 1 em = 1070 N).
 XiO-1 X Hrl Xle-1 Xl1r1
In 10 10. Ie
8 8 8 8
G. 6. 6: 6.
4 4 4: 4
Xl0-1 , X10-1. X10-1
13 5 -: , t, 0 -:,!,,:-;,E; .... -:ffi 1"""''''m' 13 1
- ".:-;.E; .. r
Ie -6 .,l=?
,
-10 -1 .. 6 10
-l
1A
mm mm " mm'
--fY

--8G . l -6. - 6 .

-1 0 1
-8 -8 ..
-10. -10.:
.j::'
3500 rev/min 3800 rev /min 4200 rev/min 4500 rev/min
orbits
N
;<: 02 X102
0\
xle2

l'
' -'
213
15
20
15
10. 10.
I 10.
5: Xl02
x 1(2,2
-:lr:' '\ 10"
5.1
D'J0 -:?0, .
I
,.10
N
10 23
-'---'--'

-
-15J -15 -15
I
-20.1 213
- .
- 2C .

loads

Figure 6(b). - Numerical orbits and loads. Qc 0.229.

 TESTS WITH FLEXIBLY MOUNTED HOUSING

The housing was now unclamped, giving an undamped natural frequency of the system
of about 4500 c/m. A series of tests was carried out using a Q value of 0.229 at
speeds of 3500, 3800, 4200 and 4500 rev/min.
In addition to ob aining vibration
orbits of the rotor relative to the housing,orbits of the rotor relative to ground
and polar diagrams of transmi tted force were obtained. For such a con figuration the
mathematical model for the structure included the same damper model as previously,
together with the stiffness of the bearing pedestal, but for speed of computation,
neglected the relatively small housing mass which was onl y 14% of the effective rotor
mass.

Comparisons of experimental and numerically predicted orbits are shown in figures

5 a,b and indicate good agreemen t. For the case of 3500 rev/min, a distorted figure
of eight pattern is observed in both experimental and numerical recordings. The
double loops visible in the other experimental recordings persisted over the entire
periods of experimen tation and those in the numerical predictions persi sted up to a
non-dimen sional time wt of 96, that is about 15 revolutions, when the length of
computing time became prohib itive. The general sizes, shapes and dispositions show
good agreement, and t he size at 4200 rev/min is observed to be larger than at the
undamped naturai" frequ ency of 4500 rev/min. The pressure recordings of figure 5b
also show fair agreement in shap e and size. Negative 'spikes' are evident in figure
5a, indicating that the oil can temporarily support tensile forces. Such ten sile
forces have been discussed in reference 2.

Turning now to the vibration orbits of th e rotor relative to ground (figs. 6a,b),
very good agreement prevails both in size, shape and disposition and a peak vibration
between 4200 and 4500 rev/min is again observed. This suggests that the squeeze-film
damper has had some effect in reducing the critical speed of the system. The polar
diagrams of total dynamic load transmitted from both lands show striking agreement in
many respects, such as size, shape and di sposition. In particular the k ink in each
experimental polar diagram for 3800, 4200 and 4500 rev/min is predicted by numerical
computation. These latter resul ts give extra confidence in accepting the short bear
ing model of reference 2 for the squeeze film, albeit with a negative pressure cur
tailed at an experimen tally recorded limit.

Also shown in figure 6b are circles representing total displacements and trans
mitted forces when no damper is present. The e ffect of gravity on the flexible bars
increased di splacements by about 0.04 mm and all forces by about 274 N. It can be
seen that, at the lower speeds the damper is not really beneficial. However, at
speeds approaching the critical speed, it s benefits become apparent, especially at
the undamped critical speed of 4500 rev/min where, without the damper the ampli tudes
o'f vibration and transmitted force would theoretically be infinite.

CONCLUSIONS

Con tinuing on from experience gained .on test rigs involving open-ended squeeze
film dampers interposed between rigid rotors and rigid bearing pedestals (refs.1,2),
this paper has described experimental and numerical work on weakly sealed dampers
with flexible pedestals. Good agreement has been observed between experimental ob
servations and numerical predictions over a Wide range of operating parameters. These
observations included vibration of rotor relative to bearing housing and relative to
groud, dynamic pressure in the squeeze-film damper and transmitted force to ground.
The work shows that, given the correct in terpretation of boundary condition s in the

427
damper, reliance c an be placed on a mathematical model inc orporating the short
bearing approximation. The benefits of such a damper can also be readily seen.

This work will shortly be extended to include the effect on the damper of
tighter sealing and the use of higher supply pressures and very low oil viscos ities,
in keeping with present-day operational practice with gas tur bines.

REFERENCES

1. Humes ,B. and Holmes ,R.: The role of sub-atmospheric film pressure in the
vibration performance of squeeze-film bearings. Jnl Mech.Eng.Sci., vol.20, no.5,
pp.283-289.

2. Holmes,R. and Dede,M.: Dynamic pressure determ ination in a squeeze-film damper.

Proc.I.Mech.E. Paper C260/80. Second International Conference on Vibrations in
rotating machinery. Cambridge, Sept. 1980.

428
 ADDITIONAL MATERIAL PRESENTED AT THE WORKSHOP

Linear theory as a qualitative predictor

Although the squeeze-film bearing is a highly non-linear element, it is worth

examining its contribution to the damping of a system against a linear assessment.

0-1 JL
1-0
(j(jjj
X t: 0-2
6 u
u/C
:.mu 0-8 0-4
5

frame 10 : .......
4 0-6
housing
Y
3
0-4 E&.-x-y

2
0-2
4-0
6-0

o 0-4 08 12 16 20
24 0-4 08 12 1-6 20 2-4
Q/Q"tI.. Q/G.l
Figure 7 ( a) . - Rotor vibration amplitude Figure 7 (b) . - Rotor vibration ampl?lude

frame. relative to the bearing housing.

relative to the engine

6
T
5
F= mu
4
ip= TF

3
Figure 7 ( c ) . - Transmissibility.

b/v'kiil = 6
2
'lr-----4

-'-.- i , --,
24 w/w
o 04 08 12 16 20 n

If we consider a linear structure consisting of a series combination of mass,

damper and spring, the graphs shown in Fig.7 are readily obtained_ The mos t notable
features of these graphs are, in Fig. 7a, the cross-over point (1/ 12, I), and the
increase in vibration amplitude with damping at a frequency ratio of 1; in Fig.7b,
the independence of v ibration with damping at w/wj'l. = 1; in Fig. '7c, the increase in
transmissibility with damping at w/ = 1.

Figs. Sa, b, c show results obtained from the test rig when running at the
critical speed of 4500 rev/min i.e. w/w", = 1. In these figures, damping is increased
in three separate ways:

429
 i) by tightening the end sealing (Fig. Sa)
ii) by increasing the lubricant viscosity from 6 to 21 cp (Fig.8b)
and iii) by increasing the lubricant supply pressure from 5 to 45 psi (Fig.8c).

In each of these figures the photographs m ay be identified as follows:

a) Vibration orbit relative to b earing housing

b) Vibration orbit relative to ground
c) Pressure recording
d) Polar loa d diagram of transmitted force.

It may be seen that in all cases the results support the qualitative conclusion of
the linear treatment, namely that at the critical speed an increase in damping by
whatever means results in an increase in rotor vibration relative to ground, and
an i ncrease in transmissibility, b ut little change in the rotor vibration rela tive
to the squeeze-film container.

430
 (a)

(c) (d)

(a) (b)

(c) (d)
Figure 8{a}. - Effect of sealing.

Upper four figures loose sealing

Lower four figures tight sealing

431
 (a) (b)

(c) (d)

(a) (b)

(c) (d)

Figure 8(b). - Effect of lubricant viscosity.

Upper four figures 6 cp

Lower four figures 21 cp

432
 (a) (b)

(c) (d)

(b)
(a)

(c) (d)

Figure 8 (c) . - Effect of lubricant supply pressure.

Upper four figures 5 psi

Lower four figures 45 psi

433
 .

VIBRATION LIMITINJ OF ROfORS BY FEEDBAQ( CONTROL

David W. Lewis, James W. Mbore, Philip L. Bradley, and Paul E. Allaire

Department of Mechanical and Aerospace Engineering
University of Virginia
Charlottesville, Virginia 22901

SUMMARY

This work reports some experimental findings of a three-mass rotor with four
channels of feedback control. The channels are independently controllable with
force being proportional to the velocity and/or instantaneous displacement from
equilibrium of the shaft at the non-contacting probe locations (arranged in the
vertical and horizontal attitudes near the support bearings).

The findings suggest that automatic feedback control of rotors is feasible for
limiting certain vibration levels. Control of one end of a rotor does afford some
predictable vibration limiting of the rotor at the other end. The "ext step in the
work is to move from acoustic drivers to electro-hydraulics so that larger force
levels may be experimented with.

INTRODUCTION

High speed turbomachines are subject to lateral vibrations, which may limit their
performance level [1, 2]. Machines are often required to operate at least 20% away
from a rotor critical speed. They must also be designed so that, for machines oper
ating above one critical speed, all clearances in the machine are sufficient to accom
modate high amplitudes of vibrations near a critical speed. Lastly, machines are
sometimes subject to subsynchronous vibrations, which can lead to machine failures
[1]. One type of neans for reducing these vibrations is that of active vibration
control.

A number of analytical studies of active feedback controls for rotating machinery

have been presented in the literature [3-6]. These have been primarily developed by
researchers in the automatic controls area and the results are normally presented in
terms associated with automatic control theory. A very limited number of experimental
works have reported the use of active control for laboratory test rigs [7-9]. With
the possible exception of the electromagnetic bearing, there has not been an applica
tion of active feedback control to an industrial rotor at this time.

This work considers unbalances in rotating machinery which might be compensated

by use of active feedback control circuits. With proper algorithms, such closed loop
automatic control devices could permit machinery to be run longer even though the
unbalanced forces grew to significant values. Such an approach would have little
impact for small and noncritical rotors. However, for very important systems for
which an unscheduled shut-down would be costly, an analog or digital closed-loop
feedback controlled force actuator(s) could be justified.

434
 Also, for some very expensive turbines, pumps, and compressors, active feedback
f
controlled force actuators would allow machinery to pass through the criticals with
smaller clearances ( because of controlled vibration levels ) resulting in more effi
cient running. With .such an application, the proposed devices would be functioning
only during runup or rundown of the machinery permitting certain advantages for the
designers of the machinery.

Values in the text are given in both SI and U.S. Customary Units. The measure
ments and calculations were made in U.S. Customary Units.

EXPERIMENTAL UNIT

A small test rotor has been constructed and is generally described by the sche
matic drawing of Fig. 1. It consists of a 12.7 mm (0.5 inch ) diameter shaft supported
on ball bearings which are 406 mm (16 inches ) apart. For these tests, three masses,
uniformly distributed between the bearings were attached to the shaft by simple set
screws. These disks weighed 8.9 N (2.0 lbs ) , 13.3 N (3.0 lbs ) , and 8.9 N (2.0 lbs )
with the larger one at the center. The bearing carriers with the associated hardware
each weighed 4.4 N (1.0 pound ) . Shaft weight is 4.0 N (0.90 pound ) .

Each bearing has mechanical springs at 11.7 N / mm (67 pounds per inch ) in the ver
tical and horizontal directions plus loud speaker motors ( acoustic drivers ) attached
to the bearings directed in the vertical and horizontal attitudes. This test rotor
is unique in that forces may be applied to the bearings, through acoustic drivers
that are independent from one another in the vertical and horizontal directions.
These four acoustic drivers provide four independent routes for applying forces to
the bearings.

The vertical and horizontal displacements near the bearings were measured rela
tive to the static equilibrium position. A force was generated through an analog
feedback loop in the same attitude but of the directlon opposing the motion. At the
same time, an additional force was generated that was proportional to the instantan
eous velocity ( and also oppositely directed ) . These forces were then applied to the
shaft, through the ball bearings, by means of the acoustic drivers. These four acous
tic drivers are independently operable. The controls for setting the constants for
generating the forces, as a function of displacement or velocity or some combination
of displacement and velocity, are also independent one from another.

EXPERIMENTAL RESULTS

A number of experimental runs were made. The ones reported here are for runs
with the rotor fairly well balanced and with no control, a low level of control, and
a high level of control. Control. was limited in these cases to the velocity feed
back loops so that, effectively, only damping was being controlled. In addition,
the system was modelled and theoretical response was obtained on a digital computer.

The results of the first run are shown in Fig. 2. The lines with data points
shown represents the experimental results and the continuous line is for theoretical
results. There is an excellent match of critical speeds at 700 and 1200 rpm and a
reasonable agreement of results for a critical near 6000 rpm. In the midrange ampli-

435
tudes are not in close agreement. It is noted that the displacement monitoring system
had a limit at about 0.56 mm (22 mils).

For the second run the control loop gains were set to give an effective damping
constant at each bearing of 0,047 N sec/mm (0.27 lb sec/in). This was for all four
control loops. However, just as the run was about to start it was found that the
current amplifier in the free-end vertical control loop was not functioning. The run
was continued with data being taken at the motor end. The results here were very
similar to those of the uncontrolled system except that the critical at 700 rpm is
eliminated. The theoretical response did not agree well except for the location of
the criticals.

A large damping factor was applied in the last experimental and theoretical
runs. The results are shown in Fig. 4. The damping constant was estimated to be
about 0.37 N sec/mm (2.13 lb sec/in), about eight times greater than the previous run.
Results are dramatic over the range of all three criticals.

DISCUSSION AND CONCLUSIONS

This limited set of experiments has shown some of the effects that may be obtained
with the use of feedback control. Here, the analog control signals were applied in
both directions at the motor end and only to the vertical direction at the other end.
For Fig. 4, the motor end resulted in a small circular orbit and simultaneously a very
flat ellipse at the other end. The effects obtained here are through the use of velo
city feedback only with the velocity gain setting at about one fourth of its full
scale values.

These runs were made using analog control only. The experimental rig has a com
plete digital control system, also. The full potential for the analog system has yet
to be explored. However, it is known that it would use some adaptive ability, either
speed sensitive or logical, in order to fully develop its potential.

The digital control system is potentially far more flexible. In this experimental
system it can be brought into play by simply turning it on and flipping a switch. The
digital unit in the control loop is an Intel 8086 single board computer. This offers
a sixteen bit manipulator. We can anticipate some very interesting control algorithms
with this system which at the present time has been used for a simple look-up algorithm
only. There will be a possibility of even employing an algorithm that "learns" what
combination of displacement and velocity (when multiplied by appropriate constants)
will produce the minimum response for the system at each speed for that particular
rotor condition. Then as the rotor condition changes (through erosion, accumulation
of dirt onto the disks, etc.) this "learning" algorithm could accommodate the chang
ing disturbance.

The acoustic drivers are small, yielding forces that are small. Typically, the
output for the drivers is approximately 15 watts maximum but with the proper phasing
does produce significant results. It is felt that electro-hydraulics will be the
preferred means of applying forces in an industrial atmosphere. Preliminary calcu
lations indicate that electro-hydraulics will be satisfactory directly for speeds up
through 10,000 rpm. For speeds above that, it will be necessary to employ one or
more revolutions of the shaft to make the necessary calculations as part of the
control algorithm. However, this time delay (or more specifically the rotation delay)

436
will not likely be significant as rotors must go through many cycles before amplitudes
change an appreciable level.

REFERENCES

1. Allaire. P. E., "Design of Journal Bearings for High Speed Rotating Machinery,"
Fundamentals of the Design of Fluid Film Bearings, Fluid Film Bearings Committee
of Lubrication Division (ASME, 1979), pp. 45-84.

2. Barrett, L. E., Gunter, E. J., and Allaire, P. E., "Optimum Bearing and Support
Damping for Unbalance Response and Stability of Rotating Machinery," Journal
of Engineering for Power, Transactions of the ASME, Vol. 100, No. 1 (January
1978), pp. 89-94.

3. Moore, J. W., Lewis, D. W., and Heinzmann, J. D., "Feasibility of Active Feed
back Control of Rotor Dynamic Instability," Presented at Workschop on Rotor
Dynamic Instability Problems in High Performance Turbomachinery, NASA/ARO
Conference, Texas A&M University, May 1980, pp. 1-9.

4. Heinzmann, J. D., "The Implementation of Automatic Vibration Control in a High

Speed Rotating Test Facility," M.S. Thesis, University of Virginia, January
1980.

5. Stanway, R. and Burrows, C. R., "Active Vibration Control of a Flexible Rotor on

Flexibly-Mounted Journal Bearings," Joint Automatic Controls Conference, Char
lottesville, Virginia, June 1981.

6. Allaire, P. E., Lewis, D. W., and Jain, V. K., "Feedback Control of a Single Mass
Rotor on Rigid Supports," Journal of the Franklin Institute, Vol. 312, No. 1,
pp. 1-11.

7. Nikolajsen, J. L. and Holmes, R., "Investigation of Squeeze-Film Isolators for

the Vibration Control of a Flexible Rotor," Journal of Mechanical Engineering
Science, Vol. 21, 1979, pp. 247-252.

8. Schweitzer, G. and Lange, R., "Characteristics of a Magnetic Rotor Bearing for

Active Vibration Control," Institution of Mechanical Engineers Conference on
Vibrations in Rotating Machinery, Cambridge, 1976, paper C239/76.

9. Gondhaleker, V. M., Nikolajsen, J. L., and Holmes, R., "Electromagnetic Control

of Flexible Transmission Shaft Vibrations," Proc. IEEE, Vol. 126, 1979, pp.
1008-1010.

437
Fi gure 1. - Schematic diagram showing three-mass rotor 1, lou d speaker motors
2, mechanical support springs 3, non-contacting probe signals for measuring
the position of the shaft at the outboard bearing 4, similar signals for
the motor-end bearing 5, key phasor signal for speed measurement 6, and a
variable soeed motQr 7.

50

40

theoretical
30
00
H


.
.
.

20
H
H

10

O-I-
o 1,000 2,000 3,000 4,000 5,000 6,000 ---'---r--,------,---.--.r--r
SPEED in RPM

Figure 2. - No feedback control.

438
 20

theoretical

O-I-
o 1,000 2,000 3,000 4,000 5,000
------r---
6,000

SPEED in RPM

Figure 3. - Small velocity feedback control.

til

20
):!

experimental

H 10

theoretical

O-I-
o 1,000 2,000 3,000 4,000 5,000 6,000
-;=-r------T--r---r--r
SPEED in RPM

Figure 4. - Large velocity feedback control.

439
 EVALUATIOO OF SHFAR lVDUNTED ElASTO.VIERIC DM'lPER

E. Zorzi and J. Walton

Mechanical Technology, Inc.
Lathrun, New York 12110

SUMMARY

The safe and reliable operation of high speed rotating machinery often requires
the use of devices that dissipate undesirable rotor vibrations. As an alternative
to the more conventional squeeze film bearing damper designs, a Viton-70 elastomeric
shear mounted damper was built and tested on a T-55 power turbine spool in the
rotor's high speed balancing rig. This application of a shear mounted elastomeric
damper demonstrated for the first time, the feasibility of using elastomers as the
primary rotor damping source in production turbine engine hardware. The shear
damper design was selected because it was compatible with actual gas turbine engine
radial space coristraints, could accommodate both the radial and axial thrust loads
present in gas turbine engines, and was capable of controlled axial preload. The
shear damper was interchangeable with the production T-55 power turbine roller
bearing support so that a direct comparison between the shear damper and the produc
tion support structure could be made. Test results show that the Viton-70 elastomer
damper operated successfully and provided excellent control of both synchronous and
non-synchronous vibrations through all phases of testing up to the maximum rotor
speed of 16,000 rpm. Excellent correlation between the predicted and experienced
critical speeds, mode shapes and log decrements for the power turbine .rotor and
elastomer damper assembly was also achieved.

INTRODUCTION

With the advent of higher power density turbomachinery and the use of long
unsupported shaft segments, rotor systems are operating near or above one or more
critical speeds and are therefore susceptible to any number of destabilizing
mechanisms. To insure the safe and reliable operation of this machinery, consider
able effort is often expended in an attempt to control both synchronous and non
synchronous vibrations. Since balancing technology has matured to the point that
the control of synchronous vibrations is not a limiting factor, emphasis has been
directed at methods of dissipating undesirable vibrations and stabilizing rotor
bearing systems. Squeeze film dampers provide one method of control but, close
tolerance machining, oil supply requirements, and the associated hydraulic plumbing
make these systems expensive and vulnerable. As an alternative to the squeeze film
bearing damper a convenient, compact, self sufficient and inexpensive dry damper
with a wide range of stiffness and damping characteristics that will provide
comparable vibration control is desired. Elastomer dampers have many of the
desirable features but, their use has been limited due to the lack of adequate
design information and demonstrations of capability. With the research reported
on in References I thru 5,_ considerable progress has been made in establishing the
design guidelines for elastomer dampers. As a demonstration of elastomer damper

440
capability in the control of rotor vibration an experimental investigation was
performed at MTI. In this test program a Viton-70 shear mounted elastomeric damper
was designed specifically to control the vibrations of a T-55 power turbine rotor
mounted in a high-speed balancing rig. The high-speed balancing rig used as the
test facility in this program was constructed by MTI under a NASA contract
(NAS 3-20609) managed by Robert Cunningham for the U. S. Army's Corpus Chris'ti Army
Depot (CCAD) and was used with the Army's permission. The CCAD high speed balancing
facility incorporates production engine bearings, support structure and is capable
of achieving operational speeds. It therefore provided an excellent opportunity to
test a production gas turbine engine hardware compatible, elastomer damper design
for the first time.

Since many gas turbines are multi-shafted and radial design flexibility is
inhibited, the damper was designed to be active in shear. This design configuration
(fig. 1) minimizes the required radial envelope and accommodates axial thrust
loading which may be required in typical gas turbine applications. Once an analyti
cal model of the production CCAD balancing rig was established and calibrated against
typical response characteristics for the rig, damper optimization studies were
performed. The results of the optimization studies provided the required ranges of
damper stiffness and loss coefficiens for this application.

As documented, the shear elastomer damper operated successfully to the maximum

operating speed of 16,000 RPM. Orbit control was reliable with no severe or large
nonsynchronous components of vibration observed. Thermocouple monitoring showed
that the temperature of the elastomer never exceeded 650C during testing. Further,
there was excellent agreement between the analytical predictions and measured data
for critical speeds, mode shapes and logarithmic decrement which clearly demon
strated the ability of the T-55 power turbine elastomer damper to predictably control
rotor vibration for safe and reliable operation.

ROTOR OPTIMIZATION STUDIES

A typical response profile of the CCAD rig with production hardware and a T-55
power turbine installed is shown in figure 2. This figure is a trace of the
synchronous response of a slow acceleration pattern for a 900 pair of probes
(vertical and horizontal) at approximately midshaft location. The two dominant
features are the peaks at approximately 4,000 and 6,000 RPM. This highly elliptical
horizontal orbit occurring at 4,000 RPM, rapidly changing to a dominant vertical
ellipse at 6,000 RPM, is typical of characteristics of a retrograde and forward
precession encouraged by asymmetric support characteristics. The response of the
turbine end at these two speeds is dominant whereas the roller bearing end (cold
end) showd little activity, indicating a strong precession of the turbine such as
produced by a rigid body conical mode shape. A subsequent set of peaks occurred in
the 8,000 to 9,000 RPM speed range, with all probe sets along the shaft indicating
motion. The activity at this speed was not predicted analytically, but was general
ly subordinate to the two main peaks at 4,000 and 6,000 RPM and was most likely due
to the structure.

To further assess possible rotor modes and determine the acceptability of the
proposed rotordynamic model, a small exciter was attached to the CCAD rig to excite
the nonrotating T-55 power turbine. Swept sine-wave excitation indicated three
possible structural modes:

441
 115 Hz bounce mode of turbine

210 Hz bending mode of shaft

600 Hz bending mode of shaft

Figure 3 indicates the location of these structural frequencies when compared

to a tuned analytical model of the T-55 (whirl speed marked by "X" at zero spin
speed axis). This analytic model also predicted a first retrograde (backward)
precession turbine conical mode at 4,500 RPM, a first forward mode at 6,000 RPM
and a first bending (2nd critical) at 21,000 RPM. The production engine support
rigidity needed to calibrate this model was reasonable, with a roller bearing
stiffness of 8.75 x 107 N/m (500,000 lb/in.) and a ball bearing (turbine) pedestal
stiffness of 1.40 x 107 N/m (80,000 lb/in.).

Further evidence of the models validity was obtained by observing the super
synchronous excitation of the small 2/rev component of motion throughout a normal
acceleration pattern to maximum speed. The 2/rev peaked while operating at 7,100
RPM indicating the presence of a mode as shown in figure 3 at the intersection of
the 2/rev excitati?n line and the second critical speed line.

Therefore, from observation of rotor synchronous response, static shaker

excita tion data and observation of the 2/rev vibration, the rotor-dynamics model
was considered acceptable for analytic examination of possible elastomeric-damper
designs.

Previous elastomer material testing of shear specimens (ref. 1) was reviewed

for candidate elastomer material selection. Viton-70 at 320C showed the largest
measured value of loss coefficient for the entire range of frequencies expected.
As the loss coefficient is a measurement of the ratio of material damping to
stiffness, a higher value of loss coefficient reflects an increased capacity to
dissipate energy. Temperature increases make the selection of material for damping
somewhat arbitrary as the loss coefficient for all materials, Buna-N, EPDM, Neoprene
and Viton ranges between 0.1 and 0.2 for the entire frequency range at a temperature
of 88C. Therefore the decision to use Viton as the damping material was based upon
the following:

Viton-70 shows superIor damping properties compared

to the other materials tested for 320C or 66C
operation (figs. 4 and 5).

Viton-70's dissipation characteristics are as good

as any of the other materials tested for operation
at 880C.

Viton-70 was previously used as a damper in success

ful testing of high-speed rotor dampers (refs. 1 and
5) .

Unused Viton material was available from MT! elastomer

test activity under NASA Programs NAS 3-18546.

442
 Based on data obtained from shear specimen shaker testing of Viton-70 (ref. 5)
a range of loss coefficients from 0.15 to 0.75 could conceivably be expected for
the 0-16,000 RPM operation of the CCAD balancing rig (fig. 6). Accordingly, the
T-55 power-turbine rotor-dynamics model was modified to determine the optimal damper
design (figs. 7 and 8). All system damping was assumed to be due to the elastomer
damper, consequently no structural damping was modeled for the turbine or gearbox
bearing. However, the two disk pack couplings were included as elastic elements
which offered a minimum of bending rigidity but offered full shear restraint.

As a result of the rotor damper optimization studies (figs. 7 and 8) a number

of facts surfaced regarding the design of the elastomeric damper:

The first critical speed is a conical precession of

the turbine with its nodal point extremely close to
the elastomer damper test bearing location. Accord
ingly, the first critical speed offers little damping
and is insensitive to the elastomer damper stiffness
or damping characteristics.

The second and third mode shapes offer significant

activity of the elastomeric damper. Consequently,
the critical speed location and log decrement of
those critical speeds are sensitive to damper selec
tion.

A best, or optimal, tradeoff between the second and

third mode is obtained by a Viton-70 elastomeric
6
damper of between 5.25 x 106 - 7.0 x 10 N/m
(30,000-40,000 1b/in.).

It is expected with this value of stiffness, the

T-55 rotor mounted on the elastomeric damper would
traverse the second critical, which is not within
the operational range of the roller-bearing mounted
production T-55 configuration (fig. 3).

The range of log decrement expected for the second

critical speed is between 0.2 to 1.1 dependent upon
the loss coefficient (temperature of operation) for
the elastomer damper.

DAMPER DESIGN AND FABRICATION

Past elastomer designs at MTI consisted of button configurations because of the

need to have either interchangeability for comparisons of different elastomers ot
for direct comparisons with squeeze film bearing dampers. T-55 damper design
however, offered uniquely challenging problems since this configuration had to be
compatible with actual gas turbine design technology and constraints. Most gas
turbines do not have the large amounts of radial envelope required by button designs,
particularly if they are multishaft turbofan or turboshaft engines such as the T-55.
On the other hand, design maneuverability and flexibility are often generally less
restricted in the axial direction. Additionally, the design must accommodate

443
possible thrust loading (although the T-55 roller bearing mount does not take
thrust) in combination with radial loads and be capable of controlled preload.
These requirements were satisfied by the shear damper configuration.

Figure 1 illustrates the concept used for the elastomer damper mount on the
T-55 roller-bearing support. Two Viton-70 shear rings mounted along the entire
circumference satisfied all requirements. Axial preload was accommodated by
controlled machining of the outer flanges which attached to the housing by twelve
equally spaced socket head cap screws. An overload protector was installed by
using an a-ring with a prescribed clearance of 0.254 mm (0.010 in.) to avoid inter
ference with normal operation of the damper while still providing backup protection
should the shear damper or bond fail.

To obtain the required stiffness for the Viton-70 damper previous shear specimen
data was used (ref. 5). The resulting design, 6.98 cm (2.75 in.) I.D. and 9.52 cm
(3.75 in.) a.D. for a 3.175 mm (0.125 in.) thick Viton-70 specimen, produced a
stiffness range in shear of approximately 6.12 x 106 - 7.0 x 106 N/m (35,000 to
40,000 lb/in.) depending upon strain amplitude experienced. These analytically
predicted values of stiffness were validated prior to test through static load
testing. The static values determined were between 6.12 x 106 and 6.47 x 106 N/m
(35,000 to 37,000 lb/in.). Axial compression stiffness was also determined and was
found to be approximately 1.26 x 107 N/m (72,000 lb/in.). In all cases the static
loads were applied and held constant for up to 5 minutes to determine if any load
relaxation would occur. No measurable load relaxation was observed.

TEST RESULTS

The baseline run of a balanced T-55 power turbine supported on the production
roller bearing support housing showed the characteristics previously noted in
fig. 2 and reconfirmed the existence of the turbine horizontal and vertical rigid
body precessional modes at 4,000 and 6,000 RPM respectively. Additionally, the
intermediate peaks in the 8,000-9,000 RPM range were again observed though not
confirmed analytically (although suspected to be induced by the rigs' structure).
The data collected during the initial elastomer damper tests showed that a substan
tial reduction in synchronous response occurred for both the 6,000 RPM and 8,000
RPM peaks but that the response near 14,000 RPM increased as can be seen in fig. 9.
Through an unbalance sensitivity study this new peak was determined to be the second
critical speed as was predicted analytically.

A comparison of the analysis performed for the elastomer-damper supported

system with the measured data provides good correlation for mode shapes (fig. 10).
Table I provides a comparison of critical speeds and logarithmic decrements for
the two criticals observed.

There is excellent correlation in whirl speed, but the first critical 's log
decrement is significantly different in value between test and analysis. This is
due to the fact that analysis has indicated a node at the damper for the first
critical. Slight changes in rigidity of the shaft or location of roller-bearing
support would alter the analytically predicted value of log decrement for the first
critical appreciably. Further, consideration of structural damping which occurs at
the turbine end bearing (which is heavily participating in the first critical,
fig. 10) was not included in the analysis. Therefore, the analysis does not reflect

444
this form of structural damping which can greatly effect the measured log decrement
of the first critical speed.

Typical measured response indicated dominant l/rev excitation of the rotor

through the testing with the elastomer damper active. Figure 11 presents a typical
frequency spectrum at 13,000 RPM. As shown, relatively minor subsynchronous and
supersynchronous excitations were experienced and the operation of the T-55 power
turbine was as predicted throughout the entire test phase.

CONCLUSION

This successful application of elastomer damper technology presents an oppor

tunity to reflect upon the many uses for elastomers as alternatives to the conven
tional squeeze film dampers. While this work does establish an initial benchmark
for elastomers as viable alternatives for damping of high speed rotating machinery,
further efforts are required to firmly establish this technology. In particular,
these efforts need to address the testing of elastomers in the rather hostile but
realistic gas turbine environments. The authors are cognizant of the temperature
limitations of elastomers however, newer and greater varieties are being introduced
each year that do offer a wide variety of application opportunities particularly in
the colder (compressor) section of gas turbines. Therefore, this contribution is
offered in the hope that further exploration of "dry damper" technology be
encouraged.

REFERENCES

1. Tecza, J. A.; Darlow, M. S.; and Smalley, A. J., "Development of Procedures

for Calculating Stiffness and Damping of Elastomers in Engineering Applica
tions - Part V." NAS3-l8546, February 1979.

2. Chiang, T.; Tessarzik, J. M.; and Badgley, R. H., "Development of Procedures for
Calculating Stiffness and Damping Properties of Elastomers in Engineering
Applications - Part I: Verification of Basic Methods." NASA Report CR-120905,
March 1972.

3. Gupta, 'P. K.; Tessarzik, J. M.; and Cziglenyi, L., "Development of Procedures
for Calculating Stiffness and Damping Properties of Elastomers in Engineering
Applications, Part II: Elastomer Characteristics at Constant Temperature."
NASA Report CR-134704, April 1974.

4. Dar1ow, M. S.; and Smalley, A. J., "Development of Procedures for Calculating

Stiffness and Damping Properties of Elastomers in Engineering Application,
Part IV: Testing of Elastomers Under a Rotating Load." NASA Report CR-135355,
November 1977.

5. Rieger, A.; Burgess, G.; and Zorzi, E., "Development of Procedur'es for Calculat
ing Stiffness and Damping Properties of Elastomers in Engineering Applications,
Part VI." NASA Report CR-159838, April 1980.

445
 T ABL E I.

Analysis Measured
Speed Log Speed Log
( RPM ) Decrement ( RPM) Decrement

First Critical 6,022 0.001 6,007 0.090*

Second Critical 13,535 0.201 13,666 0.206*

* Average value for all data obtained.

446
 FIXED OUTER
HOUSING

BOLTS FOR
AXIAL PRELOAD
O-RING FOR
OVERLOAD PROTECTION --1---1-""""'-

ELASTOMER
(FLUOROCARBON [VITON-70j)
% in . 'h in. ALONG THE
ENTIRE CIRCUMFERENCE

ROLLER ___:f
BEARING

T55 SHAFT

102349

Figure 1. - T-55 Elastomer supported roller bearing cartridge.

HORIZONTAL PROBE
10

PROBE LOCATION
VERTICAL ,

1
PROBE

16
SPEED (rpm x 10.3)

Figure 2. - Synchronous response of T-55 on CCAD rig.

447
 40

35

o
:- 25
E
.:
fD 20
UJ
a.
CI)

a! 15
x - MODES OBSERVER ON STRUCTURAL TEST
:c
3: F - FORWARD PROCESSION
B - RETROGRADE PROCESSION

"==5=.!.---if- FIRST CRITICAL SPEED

SPIN SPEED (rpm x 10-')

6%29

Figure 3. - Whirl speed versus spin speed.

0.6
DATA
500 Hz

04
I-
Z
w
0
u:
LL
w
0 0.2
U
(f)
(f)
0
--' EPDM

0.1
0 32 66 80
TEMP (OF) u24 _I

Figure 4. - Temperature variation of elastomer loss coefficient (at 500 Hz)

for shear tests.

448
 0.4
I-
Z I V
V
W
O
U
iL O )
_10

u.. UO
w f\,: \J\"'\O
0.2
0 BUTADIENE
0
en
(POLYBUTADIENE)
en
0 [..../ EPDM
...J NITRILE CHLOROPRENE
(BUNA-N) (NEOPRENE) I -
0.1
50 100 2 4 6 B 1000
FREQUENCY (Hz) 793<78-1

Figure 5. - Shear specimen data at 66C.

1.2
1.0

.8

.6

"
.4

f-
ES
-
u
;:;:
LL
w
0 .2
u
V)
V)
c
-'

0------2------4----
. 1 50------1 L-
O 6 --- 100 0
L
FREQUENCY (liZ) 793474

Figure 6. - Fluorocarbon (Viton-70) shear specimen data.

449
 OEARBOX TEST TURBINE -- LOG DECREMENT
BEARING BEARING
--- _,IILSPEED

a.o 3D,OIIO

--\
_ ...
------
:u 25,000

THIRD
2.0 20:000

e-,

I
I!l
... 15.000
I
!

1.0 ,.
,."
,.,. ' .

SECOHO
MODE 10.000

,.
..--..,
FIIIST
MODE

--I 5,000

lHIAD
MODE
SECDNDMODE
FIRST MODE

I!.ASTOMEAIC DAMPER TEST BEARING STlFFNI!S8' 10'_,

Figure 7. - T-55 power turbine - critical speeds and log decrement vs. bearing
stiffness for Viton-70 elastomeric damper (n= 0.15).

1
-.ox. TEST TURBINE
-- _,IIL SPEED
BEARINO BEARINO

30.000

--
----------
25,000

0.002

20 ------4

1ELAST00EAIC DAMPER TEST BEARING STiFFNESs. 1()11b11n.

Figure 8. - T-55 power turbine - critical speeds and log decrement vs. bearing
stiffness for Viton-70 elastomeric damper (n = 0.75).

450
 f 12

i 10
(,(,ELASTOMER
8

l!I 6
\\ -
4

0

0 2 8 10 12 14 16
SPEED (rpm 11)4)
-

Figure 9. - Vibration plot comparing baseline and elastomer mounted rotor

response.

T55 SCHEMATIC

\ TEST DATA (NORMALIZED)

TO MAXIMUM RESPONSE)

FIRST CRITICAL

SECOND CRITICAL
-

Figure 10. - Comparison of test data with analytic predicted modes.

451
 9
13,500 rpm (1/rev)
8

0: 7
I
Q.
!!!.
6
I
.....
Z 5
w
::::i
w
() 4

-l
c..
C/) 3
Ci 5,900cpm
2
2/rev
4/rev
3/rev

0
0 2 3 4 5 6 7 8 9 10

ROTOR SPEED (Hz x 10-2)

802351

Figure 11. - Spectrum plot on elastomer probe #5; 3 gm @ 3000 plane 1.

452
"'U,S, GOVERNMENT PRINTING OFFICE: 1982 6,9-008/19 1-3
 1. Report No. 2. Government Accession No. 3. Recipient's Catalog No.

NASA CP-2250
4. Title and Subtitle 5. Report Date
December 1982
ROTORDYNAMIC INSTABILITY PROBLEMS IN HIGH
6. Performing Organization Code
PERFORMANCE TURBOMACHINERY - 1982
505-32-42
7. Author(s) B. Performing Organization Report No.

E-1287

t-------t 10. Work Unit No.

9. Performing Organization Name and Address

NASA Lewis Research Center 11. Contract or Grant No.

Cleveland, OH 44135

1--_ ___________________
_________ -1 13. Type of Report and Period Covered
12. Sponsoring Agency Name and Address
Conference Publication
National Aeronautics and Space Administration
14. Sponsoring Agency Code
Washington, D. C. 20546

15. Supplementary Notes

The workshop was sponsored by Texas A&M Univers ty, College Station, Texas, and the U.S.'
Army Research Office, Durham, North Carolina, and held at Texas A&M University on May
10-122 1982
16. Abstract
The first rotordynamics workshop proceedings (NASA CP-2133) emphasized a feeling of
uncertainty involved in predicting the stability characteristics of turbomachinery. The present

workshop proceedings demonstrates that many of these uncertainties have been substantially
reduced. To a considerable extent, pap ers presented herein report the results of programs
that have been established to systematically resolve the problems of predicting stability
characteristics of high -p_rformance turbomachinery. The emphasis on experimental
validation of the forces that influence rotordynamics is particularly encouraging.

17. Key Words (Suggested by Author(s)) 1B. Distribution Statement

Rotordynamics Unclassified - unlimited

Turbomachinery STAR Category 37
Instability

*
19. Security Classif. (of this report) 20. Security Classif. (of this page) 21. No. of Pages 22. Price
Unclassified Unclassified 458 A20

* For sale by the National Technical Information Service, Springfield, Virginia 22161
National Aeronautics and SPECIAL FOURTH CLASS MAIL
BOOK
Space Administration

Washington, D.C.
20546
OHIc" Bull_
P..,.lty for Privet. U S300

NI\SI\ POSTMASTER:
If Und""'enble. (s.clion lSI
PoIl.1 Man\Ul) Do NOI Re1u.n