You are on page 1of 8

The Dynamic Stiffness iVIetliod

for Linear Rotor-Bearing


F. A. Raffa
Assistant Professor.
Systems
The behavior of linear rotor-bearing systems is investigated by using the exact ap-
F. Vatta proach of the dynamic stiffness method, which entails the use of continuous rather
Professor. than lumped models. In particular, the theoretical formulation for rotor systems with
anisotropic bearings is developed by utilizing the complex representation of all the
involved variables. The proposed formulation eventually leads to the 8 X 8 complex
Dipartimento di Meccanica,
Politecnico di Torino,
dynamic stiffness matrix of the rotating Timoshenko beam; this matrix proves to be
Corso Duca degli Abruzzi, 24, related, by a simple rule, to the 4 X 4 dynamic stiffness matrix, which describes
10129 Torino, Itaiy rotor systems with isotropic bearings. The method is first applied to the critical speeds
evaluation of a simple rotor system with rigid supports; for this case, the exact results
of the dynamic stiffness approach are compared to the usual convergence procedure
of the finite element method. Successively, the steady-state unbalance response of
two rotor systems with anisotropic supports is analyzed; for these examples, the
dynamic stiffness results compare favorably with the results of the finite element and
the transfer matrix analysis performed by other authors.

Introduction "exact" referring to the field of validity of the conventional


beam theories.
Several well-established prediction methods are nowadays
The idea that the dynamic stiffness method could be conve-
available for the linear analysis of rotor systems, with the preva-
niently adopted for solving rotor dynamics problems was pro-
lence of those approaches, such as the finite element method or
posed by Dimentberg (1961).
the conventional transfer matrix method, which are based on
However, to the best of the authors' knowledge, the applica-
the approximate representation of the physical behavior of the
tion of the method to the analysis of actual rotor systems has
rotating systems.
been presented only in the works of Thomas (1974) and Rieger
The application of the above approximate methods has pro-
et al. (1976), where the dynamic stiffness concept is applied
duced a great amount of numerical results; in particular, for the
to the calculation of critical speeds, instability threshold speeds
subject of this paper, i.e., for the case of axisymmetric rotors
and unbalance response of rotors with anisotropic, flexible and
supported on isotropic or anisotropic bearings, the existing liter-
damped supports; in both the cited works, the formulation of
ature is very vast indeed and few chances appear to be left for
the beam elements is carried out with reference to the usual
the further investigation of the problem.
BernouUi-Euler model, according to which only the transla-
In this paper, however, the analysis of rotor-bearing systems tional inertia of the distributed beam mass is taken into account.
is carried out in a different style, with respect to the approximate Within the framework of the dynamic stiffness method, the
techniques frequently presented in literature; in particular, the authors have recently presented the formulation of rotating Ti-
authors wish to draw the readers' attention to the exact approach moshenko beam elements, therefore considering, in addition to
of the dynamic stiffness method. the terms related to the translational inertia, those corresponding
It is well-known that, according to this method, a rotor-bear- to the rotational inertia, gyroscopic moments and shear deforma-
ing system is analyzed by subdividing it into a number of beam tion of the distributed beam mass.
elements, whose displacement field is assumed to be coincident This activity has been carried out in several steps: Curti et
with the exact solution of the governing differential equation al. (1992) derived the dynamic stiffness matrix of the rotating
of the adopted beam theory; due to this peculiarity, both the Timoshenko beam and applied the formulation to the analysis
inertia and stiffness properties of the system are included into of the natural frequencies and critical speeds of rotors having
a single dynamic stiffness matrix, unlike other methods where flexible undamped supports. The above formulation was succes-
the mass and stiffness matrices are kept distinct and separately sively utilized for the analysis of the unbalance response of
obtained. rotors with flexible, damped supports (Curti et al., 1993a),
With reference to the above considerations, the basic idea of while a first investigation of the applicability of the dynamic
the present work can be stated in the following way: while it stiffness method to the unbalance response of rotors with aniso-
is not questionable that reliable results can be obtained by means tropic, flexible and damped bearings was proposed by Curti et
of an approximate approach (e.g., finite element method), the al. (1993b). The above works prove that the analytical diffi-
authors believe that the dynamic stiffness method, in addition culties shown for instance by Lund and Orcutt (1967), and
to giving more accurate results, leads to a concise and formally related to the use of beam theories more complete than the
elegant description of the rotating systems; besides, its applica- simple Bernoulli-Euler model, can be satisfactorily overcome.
tion is obviously free from the convergence requirements of the The transfer matrix approach is also to be considered an exact
approximate methods. method when it is based on continuous rather than lumped
Due to the above features, the dynamic stiffness method be- models. In this respect, the formulation presented by Lund and
longs to the class of the so-called "exact" methods, the term Orcutt (1967) can be considered precisely as an extension of
the well-known Myklestad-Prohl method (Myklestad, 1944, and
Contributed by the Technical Committee on Vibration and Sound for publica- Prohl, 1945), since it entails consideration of the distributed
tion in the JOURNAL OF VIBRATION AND ACOUSTICS. Manuscript received Jan. mass of the beam element; in particular, in the cited work, only
1994. Associate Technical Editor: J. Mottershead. the shear effect was added to the Bernoulli-Euler approximation.

332 / Vol. 118, JULY 1996 Transactions of the ASME

Copyright © 1996 by ASME


Downloaded From: http://vibrationacoustics.asmedigitalcollection.asme.org/ on 04/14/2015 Terms of Use: http://asme.org/terms
In the work of Lee et al. (1991) the transfer matrix approach
based on the continuous representation of the rotor is called M + dM
modified transfer matrix method; the authors derive the transfer y y
matrix of the rotating Timoshenko beam and apply it to the
analysis of rotor systems with anisotropic, flexible and damped
bearings; their transfer matrix proves to be of order sixteen.
The reasons for the limited development of the exact formula-
tions (both dynamic stiffness and transfer matrix method), with
respect to the development of the approximate techniques, can
be twofold: on the one hand, the refinement of the flexure theory
in the analysis of rotating beam elements has been considered
of minor importance in the past, compared with other approxi-
mations involved in the analysis, mainly bearings description,
as stated for instance in the technical report AFAPL (1980);
on the other hand, the authors believe that the success obtained
by the approximate techniques (mainly the finite element
method) has diverted the researchers' attention from going deep
into the theoretical aspects of the continuous representation of y " M M + dM
rotor systems. X X
On examination of the research concerning the use of continu-
ous models for the exact representation of the behavior of rotat-
ing systems, the authors believe that the consideration reported
in the paper of Lee et al. (1991) — "Currently there are three
main methods to analyze rotor bearing systems, i.e., finite ele-
ment method, mode synthesis method, and transfer matrix
method"—can be completed by adding a fourth method, i.e.,
the dynamic stiffness method, to the above list.
In the present analysis, the rotating part of the system is
described according to the Timoshenko beam theory; the bear-
ings are characterized by constant stiffness and damping coeffi- Fig. 1 Sign conventions for generaiized forces and displacements
cients (for a total amount of two and eight coefficients in the
isotropic and anisotropic bearings case respectively). For this
class of rotor-bearing systems, it will prove clear that the exact
results obtained by means of the dynamic stiffness method, structure as the equation in terms of z (Curti et al., 1992); the
while being very close to the results given by the finite element two governing equations are
method, are not influenced by any kind of convergence require-
ments, therefore allowing the simplified representation of the
rotor systems; in addition, it is to be pointed out that with ds'* \ G J ds^dt^ G df
finite elements, for instance, the approximations inherent in the
formulation entail an error whose estimate can be difficult to
establish a priori and which is likely to be influenced by the + pA —- + il,„uj —-— iI,nUJ = 0 (1)
use of condensation algorithms (Cook, 1981); on the contrary, G dp
the dynamic stiffness method does not present any of the above
inconvenience and, therefore, can be conveniently used as a XpEJ\ d*-d XP d%
EJ- J„. +
reference method, which means that the exact results given by G J ds^df
this approach can be utilized for testing the accuracy of other
numerical techniques.
+ pA-—^ + il,„uj —J- - il,„uj ^-— =0 (2)
dt ds dt G dt
Description of the Method
In Eqs. (1) and (2), p, E, G, x are the density, Young's
The first step for the understanding of the method consists modulus, shear modulus and transverse shear factor respec-
in the derivation of the dynamic stiffness matrix of a rotating tively; A is the cross-sectional area, J is the diametral area
beam element, which is supposed to be supported on isotropic moment of inertia, w is the rotor speed, J„, and /„ are the
bearings, therefore allowing for the use of the complex notation transverse and polar mass moment of inertia per unit length of
for the variables involved. the beam, respectively.
The sign conventions for the components of both generalized For the case of harmonic motion, separation of time and
displacements (transverse deflections x, y and bending angles space variables in Eqs. (1) and (2) can be obtained through
'dy, dx) and generalized forces (shear forces F,, Fy and bending the conditions
moments My, A/,) are shown in Fig. I; the use of -'dx> rather
than -d^, as the rotational degree of freedom in the plane iy,s), z ( i , t) = Z(s) expiiXt) i?(j, t) = @is) exp(af) (3)
is worth noticing since it allows for the same formal representa-
tion of the beam behavior in the two planes; this circumstance where Z ( i ) , 0(,s) are the amplitude functions and X. represents
is important for both the isotropic and the anisotropic bearings the whirl speed. Substituting Eq. (3) into Eqs. (1) and (2) leads
case. to
The governing partial differential equation of the rotating
Timoshenko beam, in terms of the complex variable z = x -^ dV V d^Z W
Z = 0 (4)
/}', was given by Dimentberg (1961). If the actual derivation of ds' ^ I' ds' I'
the dynamic stiffness matrix is to be carried out, the additional
equation in terms of the second independent complex variable d'& 2 - — W
© = 0 (5)
7? = ^ - i{) needs to be considered, which exhibits the same ds' "^ P ds' I'

Journal of Vibration and Acoustics JULY 1996, Vol. 1 1 8 / 3 3 3

Downloaded From: http://vibrationacoustics.asmedigitalcollection.asme.org/ on 04/14/2015 Terms of Use: http://asme.org/terms


in which / is the length of the beam. For circular section beams
(/„, = 2Jm), the coefficients V and W are F(0) Kn K12 Ki3 Ki4 Z(0)
-M(0) K21 K22 ^23 ^24 0(0)
(15)
V= (j^ + ^Y'-2J^u:\ -F{1) ^31 ^^32 ^^33 K-n z(0
' 2EJ M(l) A4I A42 ^^43 ^44 ©(/)
The dynamic stiffness coefficients in (15) are calculated by
w= 11. 2
(6) applying the usual definition of stiffness coefficient; for the
EJ G amplitudes of the involved displacements and rotations, Eqs.
(11) and (12) are to be used, while the amplitudes of the shear
The solution of Eqs. (4), (5) can be assumed to be of the form force and the bending moment prove to be (Curti et al, 1992)
Z{s) = expi^Ts/l) (7)
M = EJ— (16)
®(s) = exp(frs/l) (8) X \ds ds
Substituting Eqs. (7) and (8) into Eqs. (4) and (5) yields
Carrying out computations leads to the following dynamic stiff-
T^ + 2VT - W =0 (9) ness coefficients
With the condition W > 0, which was verified in all the exam- EJ
ples reported in this paper, the two solutions for T prove to be Kn = K33 = - —fgifiii - giJ^)
real positive ( T I ) and real negative (ra)
X ( / cos 1x2 sinh ii\ - g sin 112 cosh /i,)
n = - V + Vv' + w = /Li?
^22 - ^44
T2 = -V- Vv + w = - Ml (10)
It can be shown that the general solutions of Eqs. (4) and (5) EJ
= — (/Ml - ,gM2)(/sin 112 cosh fJ-i + g cos fj,2 sinh //,)
are
EJ
Z(s) = C, cosh ( ^ I + C2 sinh f ^ ^13 — ^31 — fgiflJ-i - .gM2)(/sinh Ht - g sin 112)
D

K24 — KA2 — El {fill - gfJ.2)(g sinh IJ.I + / s i n IJ.2)


+ C,cos(ff\ + C,sin(!f\ (11) D
K12 - A'21 - K34 — — X43

0(^) : C, cosh (if] + Ce sinh (if^ EJ


/ ? [ ( / M I + gM2)(cos ij,2 cosh Ml - 1)
D
+ (/M2 - gMi) sin 112 sinh /j.,]
+ C, cos ( ^ ) + Cs sin ( ^ ^ ) (12)
K23 = K32 = "~^14 = ~^41

The two sets of integration constants (Cj, . . . , C4) and (C5, EJ


fgifl^i - gM2)(cos ij,2 - cosh //,) (17)
. . . , Cs) are not independent, being linked by the following D
relationships (Curti et al, 1992)
where
Cs=fC2 Q=fC, Q = -gQ C, = gC, (13)
D = /[2/g(cosh M] cos IJ.2 - 1) + (/^ - g^) sin jj,2 sinh //,]
in which
At this point, it can be worth noticing that the choice of the
/ = Bernoulli-Euler model greatly simplifies the above analytical
Ml G V/ formulation. In this case, in fact, Eqs. (6) give V = 0, W =
pAl\^/EJ; in addition, the bending slope is simply given by
the first derivative of the displacement with respect to s and
(14) therefore needs not be considered as a further, independent
M2 G
variable of the problem.
With reference to the orientations of generalized forces (shear With the above coefficients established, the global dynamic
force F and bending moment M) and generalized displacements stiffness matrix of a rotor-bearing system is obtained by assem-
(deflection Z and bending slope 0 ) at the beam ends, shown bUng the individual dynamic stiffness matrices of the beam
in Fig. 2, and to the sign conventions established in Fig. 1, the elements into which the system has been subdivided. The as-
following matrix relationship can be written sembling procedure is carried out according to the well-known
methods of the structural matrix analysis, i.e., by imposing both
the continuity condition (for displacements and rotations) and
i Z(l). F(l)
the equilibrium condition (for forces and moments) at the node
Z(0), F(0)
shared by two adjacent beam elements.
In view of the extension of the formulation to the case of
anisotropic bearings, the dynamic stiffness matrix of Eq. (15)
is indicated as the ' 'forward'' matrix [ Kf] and the corresponding
@(0), M(0) expressions (11), (12) are referred to as the Z/and ©/functions,
J ®(i). m respectively.
Fig. 2 Generalized forces and displacements at the ends of the beam The amplitudes of the backward components of motion, Z/,
element and ©J, are formally obtained from Eqs. (11) and (12), respec-

334 / Vol. 118, JULY 1996 Transactions of the ASME

Downloaded From: http://vibrationacoustics.asmedigitalcollection.asme.org/ on 04/14/2015 Terms of Use: http://asme.org/terms


tively, provided the appropriate condition \ = — w is used for For the bending slope i9(s, t), in close analogy with Eq.
all the relevant parameters of the formulation (V, W and conse- (18), it is possible to write
quently Hu t^2,f, g)', obviously two different sets of integration
constants have to be used in place of (Ci, . . . , C4) and (C5, i5(i, 0 = ®f{s) exp(iu)t) + ©bis) expi-iiot) (27)
. . . , Cg), for Zj, and 0*, respectively. The coefficients of the where 0 / and ©j, are the forward and backward components of
"backward" dynamic stiffness matrix, [K^], are then formally the bending slope, respectively.
obtained by using Eqs. (17). From the definition of V and W, Proceeding as with the total displacement z(s, t) leads the
Eq. (6), and the related coefficients /i,, fJ.2<f< 8, Eqs. (10) and following expression
(14), it is clear that the use of the Rayleigh or Timoshenko
beam approximation results in the functions Zj, 0;, and the •d(s, t) = A{s) cos ujt + iT{s) sin ujt (28)
matrix [Ki,] being different from Zf, 0/and [Kf], respectively.
where A and T are the auxiliary complex parameters
A direct extension of the dynamic stiffness method to rotor
systems with anisotropic bearings meets with some additional A = ©/ -I- 0 i r = % - ©„ (29)
difficulties, since, at first sight, the results obtained for the iso-
tropic case cannot be directly utilized; in fact, when dealing All the equations from (23) to (26) can be formally extended
with this problem, the usual approach in literature consists in to the bending slope components, by replacing X, F, UR, VR,
considering the sine and cosine components of the physical U/, V; with 0y, &x, AR, TR, A;, F,, respcctivcly, 0y and &x
variables (generalized forces and displacements) in the two being the complex amplitudes of the bending slopes.
coordinate planes separately, relinquishing the advantages of Since the shaft is considered to be axisymmetric, the deriva-
the unifying complex variable notation. The present formulation tion of the dynamic stiffness matrix can be performed with
for the anisotropic case is based on a different, new use of the reference to any appropriate system of orthogonal axes, not
complex notation. necessarily coincident with the fixed axes ix,y,s). The choice
The starting point of the formulation consists in writing the of the axes of the still unknown elliptical orbit of the beam
general shape of the elliptic whirl orbit in the following form center (x', y', s), Fig. 3, proves to be particularly convenient;
in fact, in this coordinate system the parameters 0, V, A, F
z{s, t) = 2.f{s) exp((a;0 + Z,,{s) e.x^{-iujt) (18) are real and therefore Eqs. (22) and (29) represent identity
conditions among moduli. Consequently, as reported in the Ap-
where 2f and Zt are the forward and backward displacement pendix, the actual computation of the dynamic stiffness coeffi-
components, respectively, the bar being used for denoting the cients, with all the involved variables, i.e., displacements, bend-
complex character of the variables; in phasor notation ing slopes, shear forces and bending moments, has been carried
out just with reference to the axes (x',y', s).
Z / = \Zf\ exp(!a) Z i = |Zt| exp(j/9) (19)
In addition, it can be worth noticing that consideration of the
Furthermore, from the definition of the complex variable above Eq. (24), and the corresponding equation for the bending
z(s,t) slope components, is necessary since the equilibrium equations,
presented later as Eq. (34), contain the external generalized
x(s,t) = mz{s,t)] y(5,0 = »[z(^,0] (20) forces related to bearings and disks. These forces are defined
with reference to coordinate axes whose position is well-known
For the problem to be solved by means of the dynamic stiffness a priori, as for instance the fixed axes (x, y, s); consequently,
method, an appropriate relationship has to be established be- the complex variables X, F, 0y, 0 ^ are the appropriate vari-
tween the generalized forces and the generalized displacements ables, which appear in the vector {q} of Eq. (34).
at the ends of the beam element. To this purpose, it proves
The generalized forces and displacements at the beam ends
useful to rewrite Eq. (18) in the following form
are related to each other by the matrix relationship
z{s, t) = 0{s) cos ojt + iV(s) sin ujt (21)
where V and V are the auxiliary complex parameters [^21] [^22] ]{i::!}={i^:i} -
U = Zf+Z, V=Z,-Z, (22)
In Eq. (30) the subvectors {qi} ({Qi}) and (qs} ({Q21) con-
From Eq. (20) it follows that
x{s, t) = UR{S) cos u!t — V,{s) sin ujt
yis, t) = U,is) cos uit + Vg{s) sin u>t (23)
where the subscripts R and / refer to the real and imaginary part
of the complex parameters, respectively; in complex notation,
x(s, t) = 'iR[^(s) exTp(iujt)]
y(s, t) = m[7is) exp(jW)] (24)
in which the complex amplitudes X and F have the following
expressions
X{s) = C/« + iV, ns) = U,- iVx (25)
The relationships between the amplitudes of the displacements,
X and F, and the auxiliary complex parameters, 0 and V, are
easily derived from Eq. (25) and prove to be
X + i7= 0+V X+ iY= U - V
where the double bar is used for indicating the complex conju-
(26)
J
gate variables. Fig, 3 Parameters of the elliptical orbit

Journal of Vibration and Acoustics JULY 1996, Vol. 1 1 8 / 3 3 5

Downloaded From: http://vibrationacoustics.asmedigitalcollection.asme.org/ on 04/14/2015 Terms of Use: http://asme.org/terms


tain the generalized displacements (forces) of planes (x, s) and Table 1 Forward critical speeds [rad/s] of the rotor system of Fig. 4
by the Timoshenko beam approximation: dynamic stiffness and finite
{y, s), respectively; the 8 X 8 dynamic stiffness matrix has element results
been accordingly partitioned into four square submatrices.
The dynamic stiffness coefficients for the rotating Timo- Mode Consistent Finite Element Method Dynamic Stiffness
shenko beam are computed by using the definition of stiffness number Number of elements Method
coefficient. Performing the computations leads to the following 2 3 6 12
new result: the 8 X 8 dynamic stiffness matrix for the aniso-
tropic case is related to the 4 x 4 dynamic stiffness matrix for 1 69.8 69.8 69.8 69.8 69.8
the isotropic case by the simple rule 2 1140.3 941.3 935.2 934.7 934.7
3 4163 3482 3079 3063 3062
[Kn] =-2([Kf] + [K,]) [K22] = [Ku] 4 - 7129 5258 5190 5182
5 - 12277 7274 7120 7099

[Kn] = i-2aKf] - [K,]) [K2>] 6 - - 13223 11734 116467


[Kr. (31)
7 - - 18454 15495 15301
8 - - 23392 18844 18509
It is worth noticing that, since the off-diagonal submatrices 9 - - 34449 26256 25433
[Ku] and [^21] correspond to the interaction between the two 10 - - 47550 31627 30407
planes, their elements are not zeros only if the effects of the
gyroscopic moments of the distributed beam are taken into ac-
count, either by themselves (Rayleigh beam approximation) or
together with the shear deformation (Timoshenko beam approx- As for the excitation of the rotor, the usual effect of an
imation); on the contrary, no coupling exists between the two unbalance mass u was considered. Because of the bearings an-
planes when either the Bernoulli-Euler approximation is consid- isotropy the centrifugal force corresponding to u has to be re-
ered or the shear effect only is added to it. solved into its X and y components, which prove to be u(^h
Details on the actual computations which have to be carried and -iuLu^e, respectively, e being the eccentricity; the above
out for obtaining Eq. (31) are reported in the Appendix. components are inserted in the appropriate positions of the ap-
plied loads vector, for each node of the rotor model where an
Bearings, Disks, Unbalance unbalance is specified.
The well-known linearized model of the bearing forces, re-
ported by Lund and Orcutt (1967), is used; consequently, each Solution
anisotropic bearing is described by four stiffness (K^^, K^y, Ky^, After considering the effects of all the previous factors, the
Kyy) and four damping coefficients {C^^, C^,, C,^, Cy,), which behavior of the rotor-bearing system proves to be described by
were assumed to be constant in the present formulation. The the following linear system of algebraic equations
effect of the bearing forces can be included into the global
dynamic stiffness matrix by adding to it the bearing stiffness [^]{ql [F] (34)
matrix according to the usual rules of the matrices assembly;
where [K] is the global dynamic stiffness matrix of the system,
the bearing stiffness matrix proves to be
{q} and ( F } are the vectors of the generalized displacements
(K,,, + i\C,,) 0 {K,y+ i\C,y) 0 and unbalance forces, respectively.
0 0 0 0 In particular, after solving the system (34), {q) will contain
(32) the values of the complex amplitudes X, 7, @y, @x at each
{K,:, + i\Cy,) 0 {Kyy "H ikCyy^) 0
0 0 0 0 node of the model; the harmonic motions along the coordinate
axes can now be readily obtained by using Eqs. (24).
The disk is assumed to be a concentrated, rigid body; the effect In addition, the parameters of the elliptic whirl orbit described
of the forces and moments applied by the disk on the shaft is by the rotor axis in the steady-state unbalance response condi-
inserted into the global dynamic stiffness matrix by adding to tion can be derived from the solution of system (34) in a
it, for the anisotropic bearings case, the following disk stiffness straightforward way. In fact, evaluation of the functions Zf and
matrix Zt, from X, Y or U, V, by means of Eqs. (22) and (26), leads
to the complete description of the orbits, if the complex form
Mok" 0 0 0 of the ellipse equation is used (Smirnov, 1979)
0 -JX 0 iIoU)\
(33)
0 0 -Mo\' 0 z(s, t) = e x p ( / $ ) ( | Z ^ exp[((a;« + * ) ]
0 —iIpLj\ 0 -Jo\'
+ \Z,,\ exp[-i(wf-f- * ) ] ] (35)
in which Mo is the disk mass, Jo and /» are the transverse and
polar mass moment of inertia of the disk, respectively. From Eq. (35), the fundamental parameters of the elliptic orbit,
shown in Fig. 3, can be computed by means of the following
The obvious condition X. = a; is to be used in the previous
formulas
equations, when dealing with the syncronous unbalance re-
sponse or the critical speeds computations.
$
a + ^
* P

a= \Zf\ + IZi, (36)


^d
In particular, the rotor will exhibit backward whirl at those
nodes where Z> < 0.
^
/7777 ^
Numerical Examples
First, the simple rotor system with rigid bearings shown in
2L Fig. 4 was considered in order to show the performance of the
proposed analytical approach in terms of accuracy and conver-
Fig. 4 Rotor system with rigid bearings ( t = 0.3 m, d = 0.02 m) gence.

336 / Vol. 118, JULY 1996 Transactions of the ASME

Downloaded From: http://vibrationacoustics.asmedigitalcollection.asme.org/ on 04/14/2015 Terms of Use: http://asme.org/terms


Bearings location Disk location Bearings location

^f=^^.
h
diskl

disk2 disi<3
-3:-—
Fig. 7 The Nelson-UcVaugh rotor system

The dynamic stiffness method was also applied to the steady-


state unbalance response of rotor systems with anisotropic bear-
ings, by using two examples taken from the literature.
In the first case the simplified system of Fig. 5 was consid-
ered, which appears to have been first proposed, and experimen-
tally studied, by Kikuchi (1970); the unbalance response of
Fig. 5 Tliree-disl< rotor system ( t ' 0.131 m,L = 1.062 m, h = 0.04 m, this system was analyzed by Lee et al. (1991) by means of
d = 0.04 m) the modified transfer matrix method. The system consists of a
con.stant section shaft with three equal disks, the unbalance
mass being applied at disk 1 only. The modulus of elasticity,
This example and the data concerning the shaft geometry shear factor and density of the shaft are 2.07 X 10" N/m^,
and the thin disk were taken from Dimentberg (1961), where, 1.47 and 7750 kg/m-*; the disk mass is 13.47 kg and its polar
however, the shaft is assumed to be massless; the modulus of and diametral mass moment of inertia are 1.02 X 10 ' kgm^
elasticity of the continuous shaft is 2.06 X 10" N/m^ and to and 5.11 X 10"^ kgm^, respectively. The unbalance at disk 1
the mass density and the Poisson coefficient the values 7800 is 1.347 X 10-" kgm.
kg/m^ and 0.3 were given. The disk mass is 15.4017 kg and The following set of bearings properties was considered (Lee
its diametral and polar mass moments of inertia are 2.4065 X e t a l , 1991)
10 ' kgm^ and 4.8130 X 10 ' kgm^ respectively.
In Table 1, a comparison between finite element and dynamic "•XX '*-V\ lO^N/m K„ K„ = 5X 10<* N/m
stiffness results is presented with respect to the first ten forward
critical speeds of the rotor system. As for the finite element C,,V = Cyy = 2 X W^ Ns/m C,y = C,, = 0
computations, four models with a progressively increasing num- The dynamic stiffness model of the rotor system consists of six
ber of consistent beam elements were defined, with the refine- elements, corresponding to twenty-eight complex degrees of
ment criterion of using equal length elements. freedom. The unbalance response of the system at disk 1, com-
On the contrary, the dynamic stiffness model is composed of puted according to the Timoshenko beam theory, is shown in
two elements only, regardless of the number of critical speeds Fig. 6 where the absolute values of both semi-axes of the whirl
to be computed; this circumstance is obviously related to the orbit are given, for the sake of comparison with the correspond-
pecuharity of the exact formulation of the dynamic stiffness ing results reported by Lee et al. ( 1 9 9 1 ) ; in particular, the
method. In particular, the finite element approach proves to curve of the major semi-axis a and the corresponding amplitude
produce results nearly as accurate as the dynamic stiffness response curve, shown in the paper of Lee et al. ( 1 9 9 1 ) , prove
method only by using the usual convergence procedure, which to be practically indistinguishable.
entails the definition of models with an increasing number of
degrees of freedom.

0.40

0.05

0.00 500 1000 1500 2000 2500 3000


(J (rad/s)
u (rad/s)
Fig. 8 Normalized unbalance response at the disl( location (Nelson-
Fig. 6 Unbalance response at disk 1 location (three-disk rotor system) McVaugh rotor)

Journal of Vibration and Acoustics JULY 1996, Vol. 1 1 8 / 3 3 7

Downloaded From: http://vibrationacoustics.asmedigitalcollection.asme.org/ on 04/14/2015 Terms of Use: http://asme.org/terms


Table 2 Dynamic stiffness and finite element results for the Nelson- lated to the isotropic dynamic stiffness matrix by the simple
McVaugh rotor by the Rayleigh beam approximation
law reported as Eq. (31).
(c) Due to the use of the complex representation of the vari-
CO 20000 rpm 40000 ipm ables involved, the dynamic stiffness matrix derived for
Rotor length a/e blE a/e b/e
the case of anisotropic bearings proves to be of order eight.
(d) In all the numerical simulations the proposed method has
355 mm (a) 2.04141 1.48724 0.74465 0.69940 produced an excellent agreement with the results obtained
355.6 mm (b) 2.04087 1.48664 0.74521 0.69958 by means of other techniques; it should be noted that, since
358.1 mm (c) 2.03855 1.48404 0.74765 0.70063 the dynamic stiffness matrix of the shaft is derived from the
exact continuous-system representation, one can actually
Results from (d) 2.04120 1.48820 0.74602 0.70050
dispense with the usual convergence procedure of mesh
(a) Curti et al. (1992) (b) Cossalter and Da Lio (1986) refinement, which proves to be necessary for the correct
(c) Nevzat and Levent (1984) (d) Nelson and Meacliam (1981) use of other approximate techniques (e.g., finite elements).

Acknowledgments
In the second example the more complex rotor-bearing sys-
tem shown in Fig. 7 was taken into account; it consists of a The research activity leading to this work was supported by
shaft supported by two equal bearings, carrying a disk at its the Italian Ministry of University and Scientific and Technologi-
overhanging part. As for the shaft configuration the data re- cal Research through the ' '60 percent'' funds.
ported by Nelson and McVaugh (1976) were used, while for
the total length the value 358.1 mm was assumed, following
the indication given in the paper of Nezvat and Levent (1984).
References
A density of 7806 kg/m^ modulus of elasticity 2.078 X 10" AFAPL-TR-78-6, 1980, "Rotor-Bearing Dynamics Technology Design
Guide," Part 1-Flexible Rotor Dynamics.
N/m^ and Poisson's coefficient 0.3 were used for the shaft; the Coolc, R. D., 1981, Concepts and Applications of Finite Element Analysis, 2nd
disk mass is 1.401 kg, its polar and diametral mass moment ed., John Wiley & Sons.
of inertia being 2.03 X 10"^ kgm^ and 1.36 X 10"^ kgm^ Cossalter, V., and Da Lio, M., 1986, "Un Codice per I'Analisi Dinamica di
respectively. Sistemi Rotore-Cuscinetti-Struttura Portante," /I Progettista Industriale, No. 9,
pp. 80-90.
The following set of bearings properties was considered (Nel- Curti, G., Raffa, F. A., and Vatta, F., 1992, "An Analytical Approach to the
son and Meacham, 1981) Dynamics of Rotating Shafts," Meccanica, Vol. 27, pp. 285-292.
Curti, G,, Raffa, F. A., and Vatta, F., 1993a, "Externally Damped Rotor Sys-
K^ = K„ = 3.503 X lO'' N/m tems by the Dynamic Stiffness Method," Proceedings, 11th International Modal
Analysis Conference, Orlando, FL, Vol. I, pp. 538-544.
K,y = K,, = -0.876 X 10' N/m Curti, G., Raffa, F. A., and Vatta, F., 1993b, "Steady-State Unbalance Response
of Continuous Rotors on Anisotropic Supports," Proceedings, I4th ASME Bien-
C,, = Cyy = 1.752 X 10^ Ns/m C^, = C,;, = 0 nial Conference on Mechanical Vibration & Noise, Albuquerque, NM, DE-Vol.
60, pp. 27-34.
The steady-state unbalance response of the rotor system due Dimentberg, F. M., 1961, Flexural Vibrations of Rotating Shafts, Butterworths,
to the disk mass eccentricity was computed; in particular, the London.
Kilcuchi, K., 1970, "Analysis of Unbalance Vibration of Rotating Shaft System
response at the disk location is reported in Fig. 8 in terms of the with Many Bearings and Dislcs," Bulletin of the JSME, Vol. 13, pp. 864-872.
normalized values of both semi-axes and using the Timoshenko Lee, A.-C, Kang, Y., and Liu, S.-L., 1991, "A Modified Transfer Matrix
beam approximation. In a successive calculation, carried out by Method for Linear Rotor-Bearing Systems," ASME Journal of Applied Mechan-
using the Rayleigh beam approximation, the authors verified ics, Vol. 58, pp. 776-783.
Lund, J. W., and Orcutt, F. K., 1967, "Calculations and Experiments on the
that the curves of both semi-axes, obtained by the proposed Unbalance Response of a Flexible Rotor,'' ASME Journal of Engineering for
method, were in excellent agreement with the corresponding Industry, Vol. 89, pp. 785-796.
curves reported by Nevzat and Levent (1984). Myklestad, N. O., 1944, "A New Method of Calculating Natural Modes of
Moreover, a detailed numerical comparison is reported in Uncoupled Bending Vibration of Airplane Wings and Other Types of Beams,''
Journal of the Aeronautical Sciences, Vol. 11, pp. 153-162.
Table 2, where for two values of the rotor speed the dynamic Nelson, H. D., and McVaugh, J. M., 1976, "The Dynamics of Rotor-Bearing
stiffness results are compared with the results obtained by Nel- Systems Using Finite Elements," ASME Journal of Engineering for Industry,
son and Meacham (1981) by applying the component mode Vol. 98, pp. 593-600.
synthesis method to a finite element model of the rotor. On Nelson, H. D., and Meacham, W. L., 1981, "Transient Analysis of Rotor
Bearing Systems using Component Mode Synthesis," ASME Paper No. 81-GT-
examination of Table 2 it is evident that, regardless of the 110.
negUgible discrepancy concerning the rotor length value, the Nevzat OzgUven, H., and Levent Ozkan, Z., 1984, "Whirl Speeds and Unbal-
dynamic stiffness and the finite element results are very close. ance Response of Multibearing Rotors using Finite Elements,'' ASME JouimAL
OF VIBRATION, ACOUSTICS, STRESS, AND RELIABILITY IN DESIGN, Vol. 106, pp.
This particular example has been considered in order to show
72-79.
that the applicability of the dynamic stiffness method is not Prohl, M. A., 1945, "A General Method for Calculating Critical Speeds of
necessarily limited to the analysis of simpUfled systems, but Flexible Rotors," ASME Journal of Applied Mechanics, Vol. 67, pp. A142-
can prove interesting also in the investigation of more signifi- A148.
cant rotor systems. Rieger, N. F., Thomas, C. B., and Walter, W. W., 1976, "Dynamic Stiffness
Matrix Approach for Rotor-Bearing System Analysis," Proceedings, IMechE
Conference Vibration in Rotating Machinery, Paper C187/76, pp. 187-190.
Conclusions Smirnov, V. I., 1979, Corso di Matematica Superiore, Vol. 1, Editor! Riuniti,
Roma.
The main conclusions which can be drawn from the present Thomas, C. B., 1974, "A Unified Formulation for the Unbalance Response of
paper can be summarized in the following points: a Flexible Rotor in Fluid-Film Bearings," Master Thesis, Rochester Institute of
Technology, Rochester, New York.
(a) For the first time, within the scheme of the dynamic stiff-
ness method, the exact dynamic stiffness matrix of the
rotating Timoshenko beam has been derived for the case APPENDIX
of axisymmetric rotors with anisotropic bearings. The algebraic developments which lead to the expression of
(b) Owing to a new- use of the complex notation, the theoretical the dynamic stiffness coefficients, for the anisotropic bearings
formulation proves to be quite different from the usual case, are described in this Appendix. The amplitudes of the
approaches reported in literature and leads to the basic forward components of the lateral displacement and bending
result that the anisotropic dynamic stiffness matrix is re- slope are here rewritten as

338 / Vol. 118, JULY 1996 Transactions of the ASME

Downloaded From: http://vibrationacoustics.asmedigitalcollection.asme.org/ on 04/14/2015 Terms of Use: http://asme.org/terms


Zf{s) = Ci cosh MI'S
+ C2 sinh lixs F(f, g, s) = cosh ( ^ ) - h(f, g) sinh ( ^

+ C3Cos(^J + C 4 s i n ( ^ ) (37) cos('^)-^M/,5)sin(^| (45)

&f(s) = C, cosh ( ^ ) + Q, sinh ( ^ G ( / , ^, 5) = -Mf, g) cosh ( ^ ] + / s i n h ^ ^'^

M2«
+ C,cos(^J+C«sin(^) (38) + fh{f, g) cos g sin f I <«,
Similarly, for the backward components being
§(cosh /^] - cos 112)
Zh(s) = D, cosh ( ^ I + £>2 sinh (^ hU\ g)
f sin )i2 + g sinh fx^
The dynamic stiffness coefficients of the eighth column can
+ D3 cos I ^ J +D4 s i n f u l (39) now be determined; for instance, the coefficient k2s is defined
as
-M,.(0, t)
©*(j) = D5 cosh ( ^ ) + Ds sinh ^ ^"' km = (47)
m, t)
From the expressions of A and F, Eq. (29), andM,., Eq. (41),
M2* Eq. (47) becomes
+ D^ cos + DHsin(^| (40)
EJ G'{f,g,Q) G'(/,i,0)1
^ R = •{•
=
2 [ G{f, g, I) G{f, g, I) J
(48)
In Eqs. (39) and (40) (Di, . . . , 0 4 ) and (£),, . . . , Dg) are
two further sets of integration constants and the hat is used
to differentiate the backward from the forward parameters. In where prime denotes differentiation with respect to s.
addition, the following expressions for the amplitudes of the After some algebraic manipulations, Eq. (48) proves to sat-
generalized forces hold isfy the general relationship (31); in fact, it can be verified that

dA dT ^28 — ' 2 ( ^ 2 4 -^ K24) (49)


My = EJ— M,. = EJ — (41)
ds ds
in which K^^ and K24 are the coefficients of the "forward" and
"backward" dynamic stiffness matrices, computed from the
F,.
--~(f- (42) coefficient /C24 reported in the paper.
By the way of example, the same procedure is applied for
the coefficient k^g, which is defined as
With reference to Eq. (30), the appropriate boundary conditions
are to be defined for each column of the dynamic stiffness -Fy'U, t)
matrix; as an example, in order to derive the expressions of k^i, = (50)
r ( / , /)
the coefficients of the eighth column, the following boundary
conditions are to be imposed From the expressions of Vand F, Eqs. (22) and (29), and Fy,
Eq. (42), Eq. (50) becomes
(7(0) = V(Q) = A(0) = r ( 0 ) = 0
IGA F'{f,g,l)
U(l) = V{1) = A(l) = 0 fes = - — - 1
2 X Gif, g, I)
Using the expressions of U, V, A, F and Eqs. from (37) to
(40) gives F'(f,g,l)
(51)
Z,is) = C,F(/, g, s) Gif, g, I)

Z,(s) = - C , ^%' ! ' II F(f, g, s) (43) Some algebraic manipulations lead to the following expression,
G(f, g, I) which is in accordance with the general result (31)
&;{s) = QGif, g, s) (52)
kn - 2(^^4 + ^34)

&,is) = - C , ^l^; ! ' II G(f, g, s) (44) where Ki^ and .^'34 are the "forward" and "backward" dy-
G(f, g, I) namic stiffness coefficients, respectively, computed from the
in which coefficient ^"34.

Journal of Vibration and Acoustics JULY 1996, Vol. 1 1 8 / 3 3 9

Downloaded From: http://vibrationacoustics.asmedigitalcollection.asme.org/ on 04/14/2015 Terms of Use: http://asme.org/terms

You might also like