Limit Cycle Vibration Analysis of a

Downloaded from https://asmedigitalcollection.asme.org/gasturbinespower/article-pdf/113/4/563/5703202/563_1.pdf by Sulzer Management Ltd. user on 24 October 2019

J. Colding-Jorgensen
Department of Machine Elements,
The Technical University of Denmark,
DK-2800 Lyngby, Denmark
Long Rotating Cylinder Partially
Filled With Liquid
The dynamics of a long, stiff cylinder, flexibly suspended at the ends, partially filled
with liquid, is analyzed for varying rotating speed. It appears from the analysis that
in the absence of external damping, two distinct speed ranges with unstable whirl
are present, as opposed to one instability region for the short cylinder, which has
been analyzed by a number of authors (Kuipers, 1964; Wolf, 1968). This is in
agreement with field experience with centrifuges, where several regions of instability
are often encountered, each corresponding to a particular vibration mode. The results
should also apply to a jet engine with oil accidentally trapped in the rotor, or any
hollow rotor with liquid trapped in the cavity. When external damping is applied
the linear theory predicts the rotor to be unstable at all speeds (Kuipers, 1964).
This is clearly not in accordance with field experience, and other authors have
suggested different types of nonlinear analysis that can give finite amplitude stable
whirl or pulsating whirl [Berman et al., 1985). In the present analysis a simplified
nonlinear analysis known as the hydraulic jump approximation is applied in the two
unstable speed ranges predicted by the linear theory, and a stable whirl finite am-
plitude, dependent on the external damping, follows. It is argued that the amplitudes
found this way should always be higher than those predicted by a more sophisticated
analysis, and also higher than the amplitudes measured by other authors, so that
the procedure described should give a safe worst case prediction of rotor whirl
amplitudes for a given external damping. Finally, an experimental setup intended
to verify the analysis in a quantitative way is presented.

Introduction
The dynamic instability caused by fluid trapped inside a
hollow rotor, but not completely filling the cavity, has inter-
ested a number of authors over the past decades. It is of prime
importance in a number of practical situations, for example
when oil accidentally gets trapped in a jet engine, or during
normal operation of a centrifuge. Kollmann (1962), Kuipers
(1964), Wolf (1966) and Lichtenberg (1982) have described the
phenomenon on the basis of experiments with simple one-
degree-of-freedom short rotors, and linear theoretical models
not including fluid internal damping. Within this context, the
problem was found to be caused by a liquid wave, traveling
on the surface of the spun-up liquid layer, in a direction op-
posite to the rotation. In a certain speed interval relative to
the rotor natural frequency, this wave will interact with the
rotor system to produce an unstable vibration with a frequency
determined by the wave speed.
Already Kuipers (1964) had discovered the peculiar fact that
within the framework of the linear, nonviscous model, external
damping on the rotor will cause the system to become unstable
Contributed by the International Gas Turbine Institute and presented at the
35th International Gas Turbine and Aeroengine Congress and Exposition, Brus-
sels, Belgium, June 11-14, 1990. Manuscript received by the International Gas
Turbine Institute December 26, 1989. Paper No. 90-GT-40.
at all speeds. This paradox was partly resolved by Hendricks
and Morton (1979), who showed that including both external
and internal damping would again produce a finite instability
region. The boundaries of this instability region have been
determined as depending on damping and mass parameters by
Hendricks and Morton (1979), using a boundary layer ap-
proximation, and later by Holm-Christensen and Trager (1989),
who obtained somewhat different results, using the full Navier-
Stokes equations.
However, both of these models still predict the system to be
unstable at some speed range, no matter what external and
internal damping is applied.
If this were really the case, some kind of active control like
that proposed in theory by Hendricks and Klauber (1984) and
in practice by Matsushita et al. (1988), using an electromag-
netically generated cross-coupling stiffness, would be the only
way to control this vibration problem. However, Berman et
al. (1985) have shown that if a nonlinear representation of the
fluid is used, external damping is predicted to have a limiting
effect on vibration amplitudes, even in the absence of internal
damping. Furthermore they have provided a simple expression
for a stable limit cycle vibration amplitude as a function of
mass parameters and external damping.
The fluid model used is known as a hydraulic jump ap-

Journal of Engineering for Gas Turbines and Power OCTOBER 1991, Vol. 113 / 563

Copyright © 1991 by ASME

 Liquid
Anti sloshing walls

/
//77

Downloaded from https://asmedigitalcollection.asme.org/gasturbinespower/article-pdf/113/4/563/5703202/563_1.pdf by Sulzer Management Ltd. user on 24 October 2019

Physical model

Fig. 1

Fig. 2 Coordinate system and nomenclature

proximation. Such a model was also used by Inoue et al. (1985)
to predict liquid wave profiles in a rotor undergoing forced
vibrations, and experimental data have been provided to sup-
port the theory. culation. However, a fluid viscous term could be incorporated
The present analysis is based on the model of Berman et al. into the equations, for example, by using a boundary layer
(1985), but it has been extended to cover the two rotor vibration approximation at the wall.
modes of a long, stiff, symmetrically suspended rotor (see Fig.
1). Fluid Equations of Motion
The influence of external damping on the vibration ampli- In the x-y coordinate system moving with the rotating whirl-
tudes is discussed, and with the help of the theory and available ing cylinder, as shown in Fig. 2, the fluid equations become:
experimental data, some conditions and guidelines are drawn
up for the use of external damping to control this vibration Tangential momentum balance:
problem. bu bu du 1 bp ••
— +u —+v — 2wv= — — — A- (1)
dt bx by p ox
Physical Model
The system analyzed, which is shown in Fig. 1, consists of
a long, rotating, stiff cylinder, partially filled with liquid. It Radial momentum balance:
is suspended at the ends with identical springs. The springs bv bv bv I dp
also have the same stiffness in the vertical and horizontal 1T+V TT + " T" -A-f (2)
dt by bx p by
directions. The effect of gravity on the spun-up fluid layer
inside the rotor cavity is neglected.
Fluid axial velocity is set to zero, as if a number of walls Continuity:
separated the cavity into narrow compartments; see Fig. 1. bu bv „
This assumption is of minor importance in the first rotor mode, (3)
but of primary importance in the second rotor mode, where dx by
the rotor is yawing around the center of gravity, introducing The axial acceleration of the cylinder is neglected, and since
axial pressure gradients into the liquid. This effect and the the fluid axial sloshing is suppressed, this two-dimensional
problem of the axial velocity assumption are discussed later representation is sufficient to describe the fluid dynamics in
in relation to projected experiments. any z plane with the local rotor axis deflection A(j).
The fluid is assumed to be nonviscous in the present cal- Even in this two-dimensional form, these equations are quite

Nomenclature
A = vectorial position of rotor axis
A = integration constant axis through center of gravity, at z
3R\oi-Q)2 perpendicular to rotational axis A = length of A
B = 7^52 = whirl constant
hfa / == length of rotor A = vectorial acceleration of rotor
C = damping coefficient M --= rotor mass axis at z
c = relative damping on first rotor P ~- = pressure f = angular coordinate in a system
mode R == inner radius of cylinder moving with the whirl
c„ = relative damping on second ro- = hydraulic jump position 6 = unit vector in tangential direc-
tor mode
st ----
= time tion at point on inner rotor
Fr = radial force in system moving U --= mean velocity in x direction wall; see Fig. 2
with whirl u == fluid velocity in x direction V = yaw angle of rotor
F, = tangential force in system mov- v == fluid velocity in y direction i = unit vector in radial direction;
ing with whirl x == tangential coordinate in system see Fig. 2
h = height of fluid layer moving with the cylinder; see to = rotational angular frequency
h0 = height of undisturbed layer Fig. 2 " 0 1 = first rotor natural frequency
h' = h-h0 y = = radial coordinate; see Fig. 2 " 0 2 = second rotor natural frequency
I = rotor moment of inertia around z -= axial coordinate Q = whirl angular frequency

564 / Vol. 113, OCTOBER 1991 Transactions of the ASME

difficult to solve, but when the spun-up liquid layer is thin, This result is consistent with the resonant wave frequencies
they can be simplified considerably. found by Miles and Troesch (1961). Only the retrograde waves
In Eq. (2), the body force A-J is small compared to the are investigated in the following, since linear models and ex-
pressure gradient, when the layer is thin, so it is neglected, as perimental evidence (Kuipers, 1964; Wolf, 1966) show that
are the nonlinear terms in this equation. Also, in Eq. (2), we these waves can cause dynamic stability problems through in-
set v=y(dh/dt)/hQ, as is normally done in shallow water cal- teraction with the rotor resonances.
culations. Then, Eq. (2) can be integrated between y and h, To determine the wave solution, Eq. (11) is integrated to:
where/7 = 0 on the free surface, to give the pressure:
w Q+
p hl-y2 d2h [h , . du - f
=^ T ^ ^ + 2M
\»dy + Ro>\h-y) (4) M=-R1^h- (M)
p 2rt 0 Ot • Jy
At the free surface, we have the boundary condition: With the initial conditions U= 0 for h' = 0, the exact solution

Downloaded from https://asmedigitalcollection.asme.org/gasturbinespower/article-pdf/113/4/563/5703202/563_1.pdf by Sulzer Management Ltd. user on 24 October 2019

of Eq. (14) is:
dh dh
(5) /?(a>-fi) 1
+ u =v U-- •h' (15)
dt dx ha
Using Eq. (5) and the continuity Eq. (3), differentiating Eq.
(4) with respect to x and inserting into Eq. (1), we get: h0
du du du •• „ „ , dh This is inserted into Eq. (12), and after integration, this
— + u — + v — + A-d = -Ru — + 2co — becomes:
dt dx dy dx dt
1
1 h20-y2 d3h (o)2 + 2a)fi)/!'+i?(co-fi) 2 ; = A02cosf+/l (16)
2 h0 dxdt1 (6)

ho
Now, the equations will be put into a simplified depth-averaged
form, setting: where A is an integration constant.
1 f This is now reduced to a second-order equation in h'/h0.
t/= (7) A Taylor series expansion in h'/ho yields:
h Vdy
and J - — =l-2h'/h0 + 3(h'/h0)2
t>o.-0) = 0 (8)
Also, in the following, the dispersion term d3h/dxdt2 is ne-
glected, in which case the solution always becomes a liquid
wave synchronized to the rotor motion. The implications of When this is inserted into Eq. (16), the following equation
neglecting the dispersion term are discussed later. Finally, we results:
neglect terms of (u- Iff, meaning that the Reynolds stresses ,2 D
caused by y gradients of the tangential velocity are small. With
these simplifications, the fluid equations to solve become:
i*-*^ -o)2 + 2con-—
ho
(w-QY)h'

= AQ2cos{-+A (17)
(9) This has two solutions, and a hydraulic jump may be in-
dt dx
terpreted as a jump between the two branches. Since the so-
dU rTdU •• „ 2dh dh lution must be periodic, the two branches must touch at f = ± IT,
— +U — + A-0=-Rw2—-
-+ 2co • (10)
dt dx dx dt by which we determine the constant A.
Consequently:
Circular Whirl Solution: Hydraulic Jump
-co 2 + 2ufi-(o)-fi) 2 -
Consider a rotor motion where the local rotor deflection h
°/ yJh202Q2A(l+cos j)
moves in a circle of radius A(z), with the angular frequency h'=h\
0. 3(.w-Q¥R 3(oi-QYR
With £=x/R + (u — Q)t, we assume a solution of the form: (18)

h = h0 + h'tf) The jump height is:

U= t/(f) 2A(l+cos S) h0Q

Ah' =2 (19)
This describes a liquid wave synchronized to the rotor vibra- 37? (co-fi)
tion. Equations (9) and (10) become:
where S is the jump position. This is similar to the results of
Berman et al. (1985).
- o + 3^5* + *->W- 0 (11)
Rotor Force and Coupled Rotor-Liquid Equations
dh' U dU
( - V + 2wQ) — + —+ co-fi — = -A£T sin f (12) In a coordinate system moving with the whirl, the fluid force
3f V? 3f per unit length is:
The rotor whirl frequency is determined for small values of zir
A, U, and h', where the determinant of the equations has to
be zero. This gives:
dF
{
— r = p(fi7?)2| h'
0
h' cosfrff (20)

in the radial direction, and

^i'l*i (13)
dF,
S
2-7T

h' sinM" (21)

dz n
The plus sign refers to prograde waves, the minus sign to
retrograde waves, relative to the cylinder surface velocity. in the tangential direction.

Journal of Engineering for Gas Turbines and Power OCTOBER 1991, Vol. 113 / 565
 In order to simplify the rotor equations of motion, it is
assumed that there exist two distinct speed ranges with high
whirl amplitudes, one for each vibration mode. In the first
interval, A = A0, and in the second, A = z-v, because of the
symmetry of the rotor (see Fig. 1). This is based on a linear
analysis published elsewhere, experiments with centrifuges by
Matsushita et al. (1988), and field experience indicating that
there generally exists one speed range with high whirl amplitude
for each vibration mode of the rotor.
For the first vibration mode, the equations of motion are:
(«oi Qz)MA0 = Fr
2MwncQA0 = F,
Equations (28) and (29) are solved for the yaw-whirl amplitude
v, and S, using the approximated whirl frequency fi of Eq.
(13).
Limit Cycle Results
Solving Eqs. (26) and (27) for the first mode, and Eqs. (28)
and (29) for the second, a complete view of the rotor vibrations
as a function of speed and external damping can be found.
This is valid only- when the two speed ranges, where high
vibration amplitudes occur, are well separated, because of the
(22)
simplifying assumptions concerning the mode shapes. If the
(23) two speed ranges with high vibrations approach each other

Downloaded from https://asmedigitalcollection.asme.org/gasturbinespower/article-pdf/113/4/563/5703202/563_1.pdf by Sulzer Management Ltd. user on 24 October 2019

and eventually overlap, as will be the case for high values of
For the second mode: the mass relation JX, or if the rotor is not symmetric, so that
dF\. A0 and v are not decoupled in the rotor modes, a more general
("02" Q2)Iv = zdz (24) and more complicated set of equations will have to be devel-
dz
oped.
The present case is, however, sufficient to illustrate the effect
2Iw02c, zdz (25) of external damping, which is done for the case of a long
J-in dz cylinder with 7=(1/12)M/ 2 . In this case, the solutions for the
where co0i and o)02 are the natural angular frequencies for each whirl amplitudes have been calculated for various values of
mode, the relative external damping c, and the results are shown in
Fig. (3), as a function of speed relative to the first rotor critical
c = 2C/(2Mco0i) a>oi.
and In the case calculated, ho/R = 0.13. The whirl frequency is
calculated as that of small amplitude whirl, given by Eq. (13)
cv=l-Cl2/(2Iu02) and the amplitude calculated is that of the end of the rotor:
Hl/2)-= A 0 + j ' - / / 2 .
Fr and F, are found by integration of Eqs. (20) and (21), with
h'(i) given by Eq. (18). Discussion
Then, the following coupled rotor-liquid equations result
for the first mode: The results show that the whirl amplitude increases with the
square of /*, which is the mass ratio, defined as the mass of
2 the amount of fluid needed to fill the cavity completely, relative
«01 fi' to the rotor mass. Also, the maximum whirl amplitude de-
creases approximately with the square of the external relative
damping c, and the whirl amplitude curve flattens out for
increasing damping; see Fig. 3.
4(2)
While the linear analysis of Hendricks and Morton (1979)
\° — - sin
3
(26)
and Holm-Christensen and Trager (1989) cannot predict the
finite amplitudes, the flattening of the curves is consistent with
and their findings of stability limits, and also with free whirl ex-
2 _4_
periments by Cheng et al. (1985). The strong destabilizing
2 Cdoi 0
C= n (1+cosS)3 (27) influence of the external damping predicted by Hendricks and
3ir Morton (1979) is, however, contradicted by the present theory.
R (co-Q) 2 Within the range of the model, the whirl amplitude depends
where u = irpR l/M and -8 = 3 , , on a balance between the external relative damping c and the
mass ratio /*. However, the range of the model also depends
This is similar to the results of Berman et al. (1985). Equations on this balance between p and c: It has been calculated that
(26) and (27) can be solved for jump position S and whirl for
amplitude A0.
For the second mode it is found in the same way that: ^ > 0.25

(WQ2-W )

Ml2 24 1 &&) 1Q 3
ho > 2
- - I sin
5 7T

2 .
- - sin3 (28)
3 ^
and
I
vl\ 2 t0()2 0
r —c*
! I20 3
'° ' ".'o i.y~
Ml1 Calculation based on 1. mode j ] Calculation based on 2. mode
- / * •
(1 + cos Sf (29)
15TTV2 Fig. 3 Limit cycle amplitudes

566 / Vol. 113, OCTOBER 1991 Transactions of the ASME

Table 1 Comparisons of analysis predictions (present theory)
and experimental data by Inoue
A/7? 0.0033 0.0066 0.0115
0.40 0.58 0.76
(AA'/Ao)max.
present theory
0.35 0.43 0.50
(A/!'//l 0 )max,
measured
Fluid force(max), 49 70 92
present theory
Fluid force(max), 35 50 75
measured (Newton)
waves of approximately the same height as the liquid layer will
be generated.
Judging from experimental results published by Inoue et al.
(1985) and Cheng et al. (1985), waves of such height, and the
resulting fluid force, will be limited by the thickness of the
fluid layer and the cavity dimension, rather than the rotor
dynamics, so on the basis of the present theory, it can be
estimated that the balance between /x and c must be:
- < 0.25
c
for this external damping to have a significant limiting effect
on whirl amplitudes.
The present analysis should be expected to overestimate wave
height and whirl amplitudes, since both viscous and Reynolds
stress terms in the equations have been neglected. This is con-
firmed by comparison to forced vibration experiments on a
rotating cylinder partically filled with water by Inoue et al.
(1985), who measured wave heights of 70 to 90 percent and
forces of around 80 percent of those predicted by the present
theory (see Table 1).
Berman et al. (1985) also reported that measured free whirl
amplitudes were about half those calculated by a nonviscous
hydraulic jump model, indicating a considerable damping in-
fluence from the viscosity of the fluid.
Table 1 presents measured maximum hydraulic jump height
and associated fluid force reported by Inoue et al. (1985),
compared to values calculated by the present theory. Correc-
tion has been made for the fact that the experiments were done
with forced oscillation in one plane, instead of circular whirl,
setting a whirl amplitude of A equal to an oscillation of 2A,
as shown by the mentioned authors. The test fluid was water
and other test parameters were:
1=49 mm, # = 152 mm,
h0 = 20 mm, 0 =63 rad/s
If the dispersion term is retained in the equations, Berman
et al. (1985) have shown that this will lead to a pulsating wave
solution. However, the whirl amplitude resulting from this
solution is smaller than that predicted by the simplified hy-
draulic jump solution. The measured pulsating whirl reported
by Berman et al. (1985) is also generally smaller in amplitude
than when the whirl is steady. Therefore, it seems that the
present hydraulic jump calculation can be used as a guide when
estimating the maximum whirl amplitudes resulting with a
given external damping c and mass ratio /x, and to establish
whether or not external damping can be used at all to limit
the vibration amplitudes of the system.
Journal of Engineering for Gas Turbines and Power
An experimental setup is presently being constructed, and
future experiments will focus on whirl amplitude measure-
ments. Also, since the present theory does not take fluid axial
sloshing into account, it is important to determine whether
this has any effect on the second vibration mode of the rotor.
This will be done, measuring with and without an antisloshing
device, dividing the cavity into narrow compartments, as in-
dicated in Fig. 1.
Conclusion
An analysis of a long, stiff, rotating cylinder, flexibly sus-
pended and partially filled with liquid has been presented, using
Downloaded from https://asmedigitalcollection.asme.org/gasturbinespower/article-pdf/113/4/563/5703202/563_1.pdf by Sulzer Management Ltd. user on 24 October 2019
an extended version of the nonviscous hydraulic jump model
of Berman et al. (1985).
This approach has an advantage over the different linearized
models previously published, in that it is able to predict limit
cycle vibration amplitudes in the speed ranges where even the
most advanced linearized models including internal and ex-
ternal damping predict unstable whirl with amplitudes growing
to infinity, regardless of the magnitude of the damping pa-
rameters.
The theory predicts a range of mass ratio relative to damping,
ti/c, where external damping can be used to limit vibration
amplitudes.
Within this range, comparison with available experimental
results shows that the present calculation overpredicts desta-
bilizing forces and whirl amplitudes with a factor of 1.1-2.0
for fluids with a viscosity in the same order of magnitude as
that of water.
The calculation can be improved by including internal damp-
ing forces in the liquid. Also, the effect of fluid axial sloshing
on higher rotor vibration modes should be studied, and a more
general rotor configuration should be analyzed.
References
Berman, A. S., Lundgren, T. S., and Cheng, A., 1985, "Asynchronous Whirl
in a Rotating Cylinder Partially Filled With Liquid," Journal of Fluid Me-
chanics, Vol. 150, pp. 311-327.
Cheng, C. A., Berman, A. S., and Lundgren, T. S., 1985, "Asynchronous
Instability of a Rotating Centrifuge Partially Filled With Fluid," ASME Journal
of Applied Mechanics, Vol. 52, pp. 777-782.
Hendricks, S. L., and Morton, J. B., 1979, "Stability of a Rotor Partially
Filled With a Viscous Incompressible Fluid," ASME Journal of Applied Me-
chanics, Vol. 46, pp. 913-918.
Hendricks, S. L., and Klauber, R. D., 1984, "Optimal Control of a Rotor
Partially Filled With an Inviscid Incompressible Fluid," ASME Journal of Ap-
plied Mechanics, Vol. 51, pp 863-868.
Holm-Christensen, O., and Trager, K., 1989, " A Note on Rotor Instability
Caused by Liquid Motions," DCAMM Report No. 390, Technical University
of Denmark.
Inoue, J., Jinnouchi, Y., and Araki, Y., 1985, "Forced Wave Motion of
Liquid Partially Filling a High-Speed Rotor," ASME Journal of Vibration,
Acoustics, Stress, and Reliability in Design, Vol. 107, pp. 447-452.
Kollmann, F. G., 1962, "Kritische Drehzahlen fliissigkeitsgefullter Hohlkor-
per," Forschung auf dem Gebiete desIngenieurwesens, Vol. 28, No. 4, pp. 115-
123; No. 5, pp. 147-153.
Kuipers, M., 1964, "On the Stability of a Flexibly Mounted Rotating Cylinder
Partially Filled With Liquid," Applied Scientific Research, Section A, Vol. 13,
pp. 121-137.
Matsushita, O., Takagi, M., Yoneyama, M., Saitoh, I., Nagata, A., and
Aizawa, M., 1988, "Stabilization by Cross Stiffness Control of Electromagnetic
Damper for Contained Liquid Rotor Unstable Vibration," Proceedings of
IMechE International Conference: Vibrations in Rotating Machinery, Sept. 13-
15, pp. 77-84.
Miles, J. W., and Troesch, B. A., 1961, "Surface Oscillations of a Rotating
Liquid," ASME Journal of Applied Mechanics, Vol. 28, pp. 491-496.
Wolf, J. A., Jr., 1966, "Whirl Dynamics of a Rotor Partially Filled With
Liquid," ASME Journal of Applied Mechanics, Vol. 33, pp. 676-682.
OCTOBER 1991, Vol. 113 / 567