Journal of Sound and Vibration (1997) 208(5), 671683

THE EVALUATION OF ROTOR IMBALANCE IN

FLEXIBLY MOUNTED MACHINES
A. W. L M. I. F
Department of Mechanical Engineering, University of Wales Swansea, Swansea, SA2 8PP,
Wales
(Received 28 October 1996, and in final form 18 April 1997)
A method is presented to determine the state of imbalance of a rotating machine by using
the measured pedestal vibration. The only requirements of the procedure are a good
numerical model for the rotor and an approximate model for the bearing behaviour. No
assumptions are made concerning the operational mode shape of the rotor and the influence
of the supporting structure is included in a consistent manner. For simplicity the analysis
is presented in a single plane orthogonal to the rotor axis, but no difficulty is foreseen in
extending the method to two planes. Examples are given for a two-bearing system.
7 1997 Academic Press Limited

1. INTRODUCTION

Methods of balancing can be categorized into two groups, the influence coefficient method
which only requires the assumption of linearity of both the machine and measuring system,
and modal balancing which in addition requires a knowledge of the modal properties of
the machine. The former of these approaches has the attraction of requiring less a priori
knowledge of the system and techniques have been well developed to make optimum use
of redundant information [1]. The approach does however suffer from the significant
disadvantage of requiring a number of test runs on site. For machinery with a high
commercial output, this is a significant disadvantage.
Modal approaches require fewer test runs but, as mentioned, require prior knowledge
of the machine. Two methods have been proposed in recent years [2, 3] which offer the
prospect of balancing without test runs. Gnielka [2] used prior knowledge of mode shapes
and modal masses and compared results to those from a numerical model of the machine.
The work of Krodkiewski et al. [3] has similar requirements and seeks to detect changes
in imbalance from running data. Both these approaches place reliance on the model of
the machine. In a real machine, however, the modal properties of the machine may be
significantly influenced by the properties of the structure on which it is supported, and this
has proved extremely difficult to model [4]. Numerical models of rotating machinery have
been used to great effect over a number of years [5, 6], and their accuracy and range of
effectiveness have been steadily developing. The principle limitation is now thought to be
the unknown effects of the foundation and recently several authors have addressed this
issue [79].
In principle, if a numerical model of a machine were sufficiently reliable, rotor imbalance
could be derived directly from the measured vibration levels at the bearing pedestals. In
this paper it is shown how rotor imbalance may be derived from measured data by using
an accurate model of the rotor and some knowledge of the bearing behaviour. No
knowledge is required of the supporting structure: this is represented as stiffness and mass
0022460X/97/500671 + 13 $25.00/0/sv971260

7 1997 Academic Press Limited

672

. . . .

matrices whose coefficients are determined as a part of the calculation. It is shown that
good estimates for the imbalance at each plane may be derived, but information concerning
the phase of the imbalance is not derived in this paper as it is strongly dependent on the
quality of the bearing model. Furthermore it is shown that additional data concerning
model parameters may be introduced into the solution to improve accuracy of results.
2. BEARING FORCES

In this work, the rotor is considered to be adequately modelled. It is assumed that the
freefree modal properties of the complete rotor train may be calculated from the model.
The freefree modes of the component rotors may be measured by suspending the rotor
in slings and can be used to verify the model. Knowledge of these freefree modes is simply
one way of representing the dynamics of the rotor itself; mode shapes and frequencies are
readily transformed into stiffness and mass matrices of the rotor. There is additional
advantage in this approach since rotors are often identical amongst a class of machines
whereas supporting structures are extremely variable [4].
Given the rotor model, the motion of the rotor is the result of two sets of forces. Firstly,
there is a set of imbalance forces which can arise anywhere along the rotor, but over the
running speed range of a machine, the resultant can be represented as a limited number
of modal contributions, which in principle may be different along the rotor. Secondly,
forces arise in the bearings and the time dependent part of these forces arises as a
consequence of the imbalance forces. At any point x along the rotor, the displacement yr
is given by
yr (x) =

G(v, x, x')F(x') dx',

(1)

where F(x') represents the force per unit length along the rotor, and G(v, x, x') is the
Greens function of the rotor, representing the response at point x arising from a unit force
at x'. (A list of notation is given in Appendix 2.) This function depends on frequency, and
one has the standard result
c*
j (x)cj (x')
,
vj2 v 2
j=1
a

G(v, x, x') = s

(2)

where the mode shapes have been normalized to the rotor mass, i.e.,

c*
j (x)r(x)cj (x) dx = 1.

(3)

Here r(x) is the mass per unit length of the rotor at x, vj is the jth natural frequency of
the rotor (freefree), cj represents the corresponding mode shape and * denotes the
complex conjugate transpose. The frequencies and mode shapes may be estimated by using
a discrete approximation via a finite element model of the rotor. In the above
representation, no allowance has been made for damping within the rotor. There would
be little difficulty including such a term but in practice damping from the bearings and
supporting structure will normally be dominant in turbo-machinery. In the present
calculation the damping is neglected for the sake of clarity.
The forces acting on the rotor are of two types, the imbalance at various locations which
are acting at m locations xe1 , . . . . , xem , which are taken as a set of pre-defined locations.

673

Note that for a finite range of running speeds, the effective inbalance distribution may be
represented as a summation over the balance planes [10]. The unknown bearing reaction
forces act at the n bearing locations xb1 , . . . xbn . In many cases n = m, but separate symbols
are retained for generality and clarity. The bearing forces Fb1 , . . . , Fbn have yet to be
determined. The total force acting on the rotor becomes
m

i=1

i=1

F(x, v) = s v 2ei d(x xei ) + s Fbi d(x xbi ),

(4)

where d is the Dirac delta function, and ei is the imbalance (i.e., the product of imbalance
mass and radius) at location xei . Inserting this expression into equation (1) yields an
equation for the shaft motion. At each of the bearings of the system, the force can be
related to the shaft displacement within the bearings. Writing the displacements of the shaft
and bearing pedestal as yr and yp respectively, and combining equations (1) and (4), gives
an expression for the displacement of the shaft at any point x:
m

i=1

i=1

yr (x) = s G(v, x, xei )v 2ei + s G(v, x, xbi )Fbi .

(5)

The bearing forces in this equation may be expressed as

Fbi = ki ( yri ypi ),

(6)

where ki is the stiffness of the ith bearing, yri = yr (xbi ) and ypi is the corresponding pedestal
displacement. Note that this is the force acting on the rotorwith an equal and opposite
force acting on the foundation structure. A speed dependent model including damping
terms is readily included by rewriting equation (6) as
Fbi (v) = (ki (v) + jvci (v)) ( yri (v) ypi (v)),

(7)

where all the frequency dependent terms have been included explicitly. Inclusion of these
terms presents no difficulties to the model. In the numerical examples presented below
constant stiffness values of the pedestal displacements ypi are assumed; then a set of
simultaneous equations may be formed for the shaft displacements at the bearing location
by setting x = xb1 , . . . , xbn . Hence
m

i=1

i=1

yrj = s G(v, xbj , xei )v 2ei + s ki ( yri ypi )G(v, xbj , xbi ).

(8)

Thus, on determining the displacements ypi at each bearing location the forces acting on
each of the bearing pedestals may be determined if all of the imbalance components are
known. Multiplying equation (5) by kj gives
m

i=1

i=1

kj yrj = kj s G(v, xbj , xei )v 2ei + kj s G(v, xbj , xbi )Fbi .

(9)

One can now eliminate yrj by using equations (6) and (9), and, after a little rearranging,
obtain an equation for each plane of vibration at each bearing:
m

i=1

i=1

Fbj + kj ypj = kj s G(v, xbj , xei )v 2ei + kj s G(v, xbj , xbi )Fbi .

(10)

674

. . . .

From this, introducing the abbreviated

Gpeq = G(v, xbp , xeq ), one may write

F
G F1 J
G
G G
g h = [a]1
G G
G
f Fn G
j

Gpq = G(v, xbp , xbq )

notation

m
F
k
y

k
s
G1e v 2ei
1
p1
1
G
i=1
G

G
m
G
kn ypn kn s Gne v 2ei
f
i=1
i

J
G
G
h,
G
G
j

and

(11)

where
aij = dij + ki Gij

(12)

In equation (12) i and j range over all bearings and dij is the Kronecker delta
Note that the expression for the forces exerted on the rotor and the reaction on the
bearing pedestal are a function of the bearing stiffnesses, but are independent of the
foundation stiffness. The foundation will, of course, strongly influence, the pedestal
vibration yp , but this is measured.
A full discussion of this approach and the conditioning of the matrix [a] has been given
by the authors [11, 12].
3. ANALYSIS OF MULTI-BEARING ROTOR

In previous studies [11], the imbalance was considered to be known. In the present work,
however, the imbalance in each of the pre-set balancing planes may be represented as an
unknown vector {e} = {e1 , . . . , en }T. By using this notation, equation (11) may be
re-written in the vector form
{F} = [a]1 [kB ]{ yp } v 2[a]1 [kB ] [Gbe ]{e},

(13)

where [kB ] = diag (k1 , k2 , . . . , kn ), { yp } = { yp1 , yp2 , . . . , ypn }T and [Gbe ]ij = Giej . Equation
(13) may be written more conveniently as
{F} = [ p]{ yp } [q]{e},

(14)

where [ p] = [a]1 [kB ] and [q] = v2[a]1[kB ] [Gbe ]. The dimension of the vector {F} is the
number of bearings (times two if both perpendicular directions are considered). The vector
{F} represents the force acting on the rotor from the bearing reaction; hence the force
acting on the supporting structure is just {F}. The matrix [Gbe ] represents the relationship
between balance planes and bearings and it is just a subset of the frequency dependent
freefree rotor description, as described by equation (2).
Let the dynamic behaviour of the supporting structure be represented by the
undetermined stiffness and mass matrices [K] and [M]. A damping matrix may also be
introduced, but since the damping of rotating machinery is usually dominated by the
behaviour of the bearing oil film (which is easily included in the analysis), the damping
matrix may be neglected. By using these matrices, a further vector equation may be written
to express the force vector as
{F} = [K]{ yp } v 2[M]{ yp },

(15)

and an equation is formed at each frequency measured. Note that equation (15) limits the
number of modes in the foundation to equal the number of bearings. In the case of a

675

foundation with many modes within the running frequency range, the structural matrices
should vary. This may be treated by subdivision of the frequency range, resulting in a larger
Least Squares (LS) problem, but no additional difficulty in principle. Since the elements
of the matrices are unknown, it is convenient to re-write this equation in the form
{F} = [w(v)]{v},

(16)

where [w] contains all reference to { yp }, and the vector {v} contains the elements of
matrices [K] and [M]. The ordering of this vector is arbitrary, but for the two bearing,
single direction case one can make the choice {v} = {k11 k22 k12 m11 m22 m12 }T, where kij
represents the (i, j)th element of [K]. It has been assumed that both [K] and [M] are
symmetric so that kij = kji and mij = mji . In this case [w(v)] is the 2 6 matrix
[w(v)] =

yp1 (v)
0

v 2yp1 (v)
0

yp2 (v)
0
yp2 (v) yp1 (v)

0
v yp2 (v)
2

v 2yp2 (v)
.
v2yp1 (v)

(17)

Equating the two expressions at frequency yields the equation

[w(v)]{v} = [ p]{ yp } [q(v)]{e}.

(18)

Writing these equations at each frequency monitored gives a single equation,

[W]{v} + [Q]{e} = {P},

(19)

where the matrices [Q] and [W] and the vector {P} are formed by the addition of each
corresponding sub-matrix as

K q(v1 ) L
G q(v2 ) G
[Q] = G G ,
G G
k q(vN )l

K w(v1 )L
G w(v2 )G
[W] = G G
G G
k w(vN )l

K [ p(v1 )]{ yp (v1 )} L

G [ p(v2 )] [yp (v2 )} G
G,
{P} = G

G
G
k [ p(vN )] [yp (vN )}l

and

where vi is the ith of N frequency points at which displacement is measured.

Equation (19) may be written in the form
[W

Q]

$%

v
= {P}.
e

(20)

This represents an over-specified problem, provided sufficient frequencies are measured,

for which one may find the MoorePenrose pseudo-inverse giving the least squares solution
as

67 $

v
W TW
=
e
Q TW

% $ %

W TQ
Q TQ

WT
{P}.
QT

(21)

In many practical situations there may not be sufficient information to determine all
parameters, but this may be overcome by the introduction of other physical information,
as outlined below, or by regularization of the problem by other means. A convenient
approach to this is by the use of Singular Value Decomposition (SVD), and limiting the
number of singular values used for the matrix inversion. Use of this technique was not
necessary for the examples in the paper but the method can be found in reference [13].

. . . .

676

The direct solution method has been used in the present calculations in which [W] and
[Q] are both real. In practice these matrices will usually be complex, and in that case
equation (21) must be divided into real and imaginary parts. The unknown parameters are
real and are given by the solution to the equation

Re [W]
Im [W]

%6 7 6

v
Re (P)
=
.
e
Im (P)

Re [Q]
Im [Q]

(22)

This solution may easily be modified to take account of any knowledge of the structure
which may be available, such as one or more of the direct stiffness terms. Such knowledge
may arise from modelling or prior tests. If part of [K] is known then one column of [W]
is removed and [W] becomes [WR ]. If the column removed is a vector {Wj } then

67 $

v
WRT WR
=
e
Q TWR

%$

WRT Q
Q TQ

WRT
{{P} vJ WJ },
QT

(23)

where vJ is the value of the known parameter. Hence in principle, all desired parameters
can be determined. Care is required to optimize the conditioning of the regression matrix
in equations (21), (22) or (23). This is achieved by appropriate scaling of parameters. A
scaling for the spatial displacement may be derived from that given by a unit imbalance
at unit scaled frequency with the reference stiffness. Hence xr = 1/mr . After inserting these
equations, the original equation becomes

$ %6 7 0 1 $ %6 7 6
x
v

xr
vr

K
mr vr2

M
mr

x
F
=
,
xr
mr vr2 xr

(24)

and with these appropriately chosen values, each of the terms in this equation is of order
of unity. The choice of reference frequency vr is the only arbitrary choice. This choice is
made in such a way as to minimize the condition number of the matrix to be inverted.
Equation (19) remains valid for the study of two orthogonal directions. If both
orthogonal directions are considered, and there is damping present in the bearing model,
[W], [Q] and {P} all become complex. The parameters to be identified remain real. The
vector {v}, neglecting damping, is now {k11yy k11zz k22yy k22zz k11yz k12yz k22yz k12yy k12zz k12yz }T
with corresponding mass terms and each component of w(v) is given by the relation
[w(v)] = [[w ] v2[w ]],
where

K
G
[w ] = G
G
k

yp1

zp1

yp2

zp1

yp1

yp2

zp2

yp2

zp1

zp2

yp1

zp2

yp2

zp1

L
0 G
G,
0
G
yp1 l
zp2

(25)

and z is the transverse direction orthogonal to y. The imbalance vector now becomes a
four component vector {e} = {e1y e1z e2y e2z }T where in each of the balance planes there
are two component to the imbalance, referred to a reference shaft marker position. The
resulting equation is solved as before (equation (22)).

677

k1

k2

m1

m2
kf1

kf2

Figure 1. Two-bearing rotor.

4. A SIMPLE EXAMPLE

The example considered is illustrated in Figure 1. A simple rotor 4 m long having mass
1450 kg is mounted on two bearings of stiffness 177 and 354 MN/m respectively. These
bearings are independently supported on foundations which are each represented by a
single mass and stiffness. These two support stiffnesses are both set at 177 MN/m whilst
the two pedestal masses are 90 and 135 kg respectively. With these parameters the natural
frequencies of the rotor alone are those of two rigid modes, 67 and 183 Hz, whilst the
frequencies of the rotor mounted on the bearings but with rigid pedestals are 29 and
100 Hz.
The bearing stiffnesses were held constant over the frequency range. It is recognized that
this is unrealistic, but the simple model suffices to illustrate the important features of the
method. Table 1 shows the range of tests considered. The first test uses data from 120
frequency points over the range of 0120 Hz, which contains the first two modes. Figure 2
shows the displacement of the two bearings. In the next six cases studied, the effects of
restricting the frequency range and the number of data points is examined in the presence
of noise. Noise is added as a factor times the high frequency response of the system.
Figure 3 shows the noisy data for the two bearings for test case 3.
Since the method requires input of the bearing stiffnesses, tests 912 consider the effects
of an error in these stiffnesses. Perfect data is assumed in case 9 and a noise factor of 5%
T 1
Summary of cases considered for two-bearing example
Case

Frequency range
(Hz)

No. of frequency
steps

Noise (relative to
high freq. level)

Bearing error
(%)

1
1a
2
3
4
5
6
7
8
9
10
11
12
12a
12b

120
120
120
120
120
80
50
80
120
120
120
120
120
120
120

120
120
120
120
50
50
50
30
120
120
120
120
120
120
120

0
0
002
005
005
005
005
005
02
0
005
005
002
002
005

0
0
0
0
0
0
0
0
0
10
10
20
50
50
50

. . . .

678

(a)

20

Displacement (dB re 1 mm)

40

60

80
(b)

20

40

60

80

40

80

120

Frequency (Hz)
Figure 2. Absolute pedestal displacements for case 1. (a), Bearing 1; (b), bearing 2.

is used in the remaining cases. The error in the stiffness of both the bearings is 10, 20 and
50% respectively.
5. RESULTS

Table 2 shows the results for the cases given in Table 1. Each calculation using Gaussian
random noise was the average of twenty independent calculations. The table shows the
means and standard deviations for each of the parameters. The convergence of the
statistics was checked with some longer runs of one of the examples. It was established
that, whilst the results are not fully converged for 20 samples, the figures are representative
and considered sufficiently accurate for the present study. In cases 112 shown in Table 1,
one bearing is of the same stiffness as the pedestal, whilst the other one is double the
stiffness. This is a demanding example for the study as the treatment of the bearing forces
becomes a better approximation for a soft foundation. The study of machines in which
the supporting structure is more flexible than the bearing is covered by cases 1a, 12a and
12b. This is representative of many large modern turbo-generators.
In case 1, the identification calculation is carried out with no noise present. No difficulty
is experienced in the correct identification of both the elastic coefficients of the model and
the unbalance at the pre-determined balance planes. Cases 2 and 3 cover the same
frequency range, covering two natural frequencies, and both cases show excellent treatment
of the noisy signals, particularly in establishing the imbalance.
Case 4 utilizes fewer frequency steps but retains the same frequency range. The effect
of the noise produces larger errors in the estimated parameters, although the results are
still acceptable. Case 8 considers the highest level of noise; mean imbalance predictions,
however, remain within 5% of the exact result, whilst the standard deviation is within 30%.

679

Note that this level of noise is unlikely in practice. Nevertheless, a machine balance on
this would reduce vibration substantially.
As the frequency range is reduced to 50 Hz, case 6, the accuracy of the model parameters
becomes unacceptable. Note that this reduced frequency range covers only one of the
natural frequencies of the system. It is worthy of note, however, that the standard deviation
of the imbalance is only 5% of the true level and the mean is 20% in error. Cases 5 and
7 cover a wider frequency range but with many fewer frequencies. It is not surprising that
the noise degrades the estimate of imbalance.
It was considered important to examine the sensitivity of identified parameters to
uncertainties in the bearing stiffnesses. These are highly uncertain in a real machine, partly
as a result of variations in static loading. Therefore, identification of model parameters
and imbalance was carried out with varying error levels in the assumed bearing stiffness.
Cases 1012 have bearing stiffness errors of 10, 20 and 50% respectively and the resulting
errors are modest, the largest being 20%. Note that with this same noise and error levels
for a more flexible support structure (with a stiffness of 295 MN/m) the error in imbalance
reduces to about 6% as shown in cases 1a, 12a and 12b.
Whilst the standard deviation of structural parameter, particularly the masses are large
in many cases, smaller deviations are observed in the imbalance estimates. It is this feature
which shows some promise for technique.
6. DISCUSSION

It has been established in section 5 that the method presented in this paper for the
identification of imbalance is appropriate for use on machinery in which the bearing
oil films are at least as stiff as the foundation over the speed range of interest. This is
because the estimation of bearing force levels is insensitive to assumed levels of bearing

(a)

20

Displacement (dB re 1 mm)

40

60

80
(b)

20

40

60

80

40

80
Frequency (Hz)

Figure 3. As Figure 2 but for case 820% noise added.

120

. . . .

680

T 2
Results for two-bearing example: estimated parameters; results of 20 runs of each
casestandard deviations in brackets
Case
1
1a
2
3
4
5
6
7
8
9
10
11
12
12a
12b

k11 108
(N/m)

k22 108
(N/m)

177(0)
029(0)
175
(015)
170
(033)
158
(052)
056
(012)
091
(006)
018
(021)
187
(041)
162
(0)
184
(052)
146
(053)
131
(075)
029
(0002)
029
(0007)

177(0)
029(0)
174
(016)
169
(034)
159
(055)
057
(011)
086
(004)
023
(027)
185
(041)
162
(0)
185
(054)
146
(056)
132
(078)
029
(0002)
029
(0008)

k12 108
(N/m)
0(0)
0(0)
0025
(015)
007
(034)
006
(053)
074
(011)
088
(004)
038
(023)
011
(041)
003
(0)
019
(053)
011
(055)
01
(077)
0007
(0002)
0007
(0007)

m11
(kg)

m22
(kg)

m12
(kg)

95(0) 135(0) 0(0)

95(0) 135(0) 0(0)
85
125
99
(59)
(62)
(61)
68
108
28
(128) (136)
(134)
35
81
35
(208)
(225)
(214)
23
48
683
(65)
(70)
(67)
65
106
186
(216)
(141)
(118)
155 122
511
(116)
(139)
(122)
154
188
30
(164)
(166)
(163)
70
108
15
(0)
(0)
(0)
158
199
71
(210)
(212)
(209)
34
68
47
(212)
(221)
(215)
21
59
45
(298)
(310)
(304)
102
133
83
(27)
(31)
(29)
103
134
87
(95)
(11)
(10)

e1
(kgmm)

e2
(kgmm)

2(0)
2(0)
201
(007)
202
(018)
193
(029)
247
(251)
254
(012)
040
(309)
210
(070)
187
(0)
188
(024)
178
(024)
153
(022)
189
(001)
189
(003)

3(0)
3(0)
302
(009)
303
(021)
296
(030)
276
(254)
243
(013)
078
(309)
311
(076)
281
(0)
279
(016)
267
(022)
234
(026)
292
(001)
291
(003)

stiffness over most of the frequency range. Whilst this assumption covers a range of
turbo-machinery of practical interest, it is not valid in all circumstances. For machines
mounted on very stiff supports, the methodology may still be employed, but either an
accurate assessment of bearing stiffness must be available, or use must be made of absolute
rotor displacement at the bearings. If these absolute measurements are available, a bearing
model is not required for the analysis. Appendix 1 outlines the modifications to the
calculation required in this case.
No damping has been included in the studies in this paper. The inclusion of damping
presents no real difficulties, but for an analysis in a single plane, the prediction of phase
will be examined in a subsequent paper. For a fully self-consistent analysis, it is believed
that consideration and measurements are necessary in both directions perpendicular to the
rotor axis. This analysis follows logically from the methods presented above, but a full
study of the sensitivity of the case is beyond the scope of the present paper.
A notable feature of the method is that in the presence of noise the imbalance estimates
are more accurate than the mass and stiffness parameters. The mass parameters are the
terms most prone to errors in the estimation procedure. Clearly this behaviour is related
to the form of the equation errors which are minimized, and this phenomena will be
investigated more fully in later work. Since imbalance estimation is an important practical

681

problem, the accuracy demonstrated above indicates that a useful technique has been
outlined.
7. CONCLUSIONS

A method has been presented to derive imbalance components by using measured data
and a model of the rotor only. The analysis is insensitive to bearing parameters for
machines with flexible supports. The approach is easily generalized to two directions.
Although stiffness and mass terms show moderate sensitivity to uncertainties, the method
has been shown to produce imbalance estimates which are insensitive to measurement
noise. Data is required covering as many modes as there are balancing planes.
ACKNOWLEDGMENTS

This work forms part of a project funded by Nuclear Electric Ltd. and Magnox Electric
plc to derive methods for inferring the influence of flexible turbo-generator foundations.
The authors thank them for funding this work and for permission to publish this paper.
Dr. Friswell gratefully acknowledges the support of the EPSRC through the award of an
Advanced Fellowship.
REFERENCES
1. J. D 1980 Institution of Mechanical Engineers Conference on Vibrations in Rotating
Machinery, Cambridge, UK, 6570. Processing surplus information in computer aided balancing
of large flexible rotors.
2. P. G 1983 Journal of Sound and Vibration 90, 157172. Modal balancing of flexible rotors
without test runs: an experimental investigation.
3. J. M. K, J. D and N. Z 1994 Journal of Sound and Vibration 169,
685698. Identification of unbalance change using a non-linear mathematical model for rotor
bearing systems.
4. A. W. L and I. C. S 1983 Institution of Mechanical Engineers Conference, C6/83,
London, UK. The dynamics of turbo-alternator foundations.
5. T. H. MC and M. L. A 1992 Institute of Mechanical Engineers Conference on
Vibrations in Rotating Machinery, Paper C432/137, 239250, Bath, UK. Troubleshooting power
plant rotating machinery vibration using computational techniques: case histories.
6. C. F, G. A. Z and A. V 1988 Institution of Mechanical Engineers Conference
on Vibrations in Rotating Machinery, Paper C302/88, 395404, Edinburgh, UK. Some in-field
experiences of non-repeatable behaviour in the dynamics of rotating machinery.
7. A. W. L 1988 Institution of Mechanical Engineers Conference on Vibrations in Rotating
Machinery, Paper C306/065, 209215, Edinburgh, UK. The least squares method applied to
investigating rotor/foundation interactions.
8. G. A. Z 1992 Institution of Mechanical Engineers Conference on Vibrations in Rotating
Machinery, Paper C432/092, 173182, Bath, UK. Identification methods in the dynamics of
turbo-generator rotors.
9. N. S. F and E. J. H 1995 Mechanical Systems and Signal Processing 9, 243256. Including
foundation effects on the vibration behaviour of rotating machinery.
10. W. K 1972 Transactions of the American Society of Mechanical Engineers, Journal
of Engineering for Industry 94, 584560. Should a flexible rotor be balanced in N or N + 2
planes?
11. A. W. L and M. I. F 1996 International Conference on Identification in Engineering
Systems 793803, Swansea, UK. Estimation of forces exerted on machine foundations.
12. M. I. F, A. W. L and M. G. S 1996 NAFEMS Conference on Structural
Dynamics Modelling: Test Analysis and Correlation, 461472, Cumbria, UK. Model updating
techniques applied to turbogenerators mounted on flexible foundations.
13. G. G and C. V L 1989 Matrix Computation. John Hopkins University Press.

. . . .

682

APPENDIX 1: INCLUSION OF ROTOR ABSOLUTE MOTION MEASUREMENT

From equation (8), the rotor displacement at any bearing position is given by
m

i=1

i=1

yrj = s Gjei v 2ei + s Gji Fbi .

(A1)

Then for frequencies away from the freefree resonances of the rotor, the forces may be
written as
{F} = [a(v)]1{ yr [Gbe (v)]v 2{e}},

(A2)

where
[a(v)] =

&

G(v, xb1 , xb1 )

G(v, xbn , xb1 )

'

G(v, xb1 , xbn )

G(v, xbn , xbn )

(A3)

Hence one may write an equation corresponding to equation (14) of the main text:
{F} = [a(v)]1{ yr } [q']{e}.

(A4)

Note that q' differs from q of equation (14) only insofar as the matrix [a'] which replaces
the [a] of equation (12) does not have the delta function term present: i.e.,
[q'] = v 2[a]1 [Gbe ].

(A5)

Then the analysis presented in the paper continues and equation (19) is replaced by
[W]{v} + [Q]{e} = {Pr },

(A6)

F [a(v1 )]1{ yr (v1 )} J

G [a(v2 )]1{ yr (v2 )} G
h.
{Pr } = g

G
G
1
f [a(vN )] { yr (vN )} j

(A7)

where

In equation (A6) the matrices [W] is unchanged whilst [Q] is modified as outlined in
equation (A5). The solution for the simple case with no constraints is

67 $

v
W TW
=
e
Q TW

WTQ
{Pr }.
Q TQ

(A8)

APPENDIX 2: NOTATION
x
y
z
L
G
F
{e}
n
m
k

axial distance along rotor

displacement orthogonal to rotor
second orthogonal direction
length of rotor
rotor Green function
force on rotor
imbalance
number of bearings
number of imbalance planes
bearing stiffness

c
[a]
Gij
[K]
[M]
[w]
[v]
[W]

bearing damping
force sensitivity matrix at a given
frequency
discrete representation of G
foundation stiffness matrix
foundation mass matrix
frequency dependent matrix of displacements
vector of unknown components
concatenation of [w]


[q]
[Q]
[ p]
{P}
Greek
d( )
dij

imbalance influence matrix defined

in equation (14)
concatenation of [q]
force influence matrix
generalized force vector defined in
equation (19)
Dirac delta function
Kronecker delta

v
r
c

angular frequency
rotor mass per unit length
mode shape (freefree)

Subscripts
i, j
bi
ei
r
p

counters
bearing i
imbalance plane i
rotor
pedestal

683