M. Rades - Dynamics of Machinery 2

Mircea Rade
Dynamics of Machinery
II
2009
Preface
This textbook is based on the second part of the Dynamics of Machinery lecture course given since 1993 to students of the English Stream in the Department of Engineering Sciences (D.E.S.), now F.I.L.S., at the University Politehnica of Bucharest. It grew in time from a postgraduate course taught in Romanian between 1985 and 1990 at the Strength of Materials Chair and continued within the master course Safety and Integrity of Machinery until 2007. Dynamics of Machinery, as a stand alone subject, was first introduced in the curricula of mechanical engineering at D.E.S. in 1993. To sustain it, we published Dynamics of Machinery in 1995, followed by Dinamica sistemelor rotor-lagre in 1996 and Rotating Machinery in 2005. The course aims to: a) increase the knowledge of machinery vibrations; b) further the understanding of dynamic phenomena in machines; c) provide the necessary physical basis for the development of engineering solutions to machinery problems; and d) make the students familiar with machine condition monitoring techniques and fault diagnosis. As a course taught to non-native speakers, it has been considered useful to reproduce, as language patterns, full portions from English texts. For the students of F.I.L.S., the specific English terminology is defined and illustrated in detail. Basic rotor dynamics phenomena, simple rotors in rigid and flexible bearings as well as examples of rotor dynamic analyses are presented in the first part. This second part is devoted to the finite element modeling of rotor-bearing systems, fluid film bearings and seals, and instability of rotors. The third part treats the analysis of rolling element bearings, gears, vibration measurement for machine condition monitoring and fault diagnosis, standards and recommendations for vibration limits, balancing of rotors as well as elements of the dynamic analysis of reciprocating machines and piping systems. No reference is made to the vibration of discs, impellers and blades.
July 2009
Mircea Rade
Prefa
Lucrarea se bazeaz pe partea a doua a cursului de Dinamica mainilor predat din 1993 studenilor Filierei Engleze a Facultii de Inginerie n Limbi Strine (F.I.L.S.) la Universitatea Politehnica Bucureti. Coninutul cursului s-a lrgit n timp, pornind de la un curs postuniversitar organizat ntre 1985 i 1990 n cadrul Catedrei de Rezistena materialelor i continuat pn n 2007 la cursurile de masterat n specialitatea Sigurana i Integritatea Mainilor. Capitole din curs au fost predate din 1995 la cursurile de studii aprofundate i masterat organizate la Facultatea de Inginerie Mecanic i Mecatronic. Dinamica mainilor a fost introdus n planul de nvmnt al F.I.L.S. n 1993. Pentru a susine cursul, am publicat Dynamics of Machinery la U. P. B. n 1995, urmat de Dinamica sistemelor rotor-lagre n 1996 i Rotating Machinery n 2005, ultima coninnd materialul ilustrativ utilizat n cadrul cursului. Cursul are un loc bine definit n planul de nvmnt, urmrind: a) descrierea fenomenelor dinamice specifice mainilor; b) modelarea sistemelor rotor-lagre i analiza acestora cu metoda elementelor finite; c) narmarea studenilor cu baza fizic necesar n rezolvarea problemelor de vibraii ale mainilor; i d) familiarizarea cu metodele de supraveghere a strii mainilor i diagnosticare a defectelor. Fiind un curs predat unor studeni a cror limb matern nu este limba englez, au fost reproduse expresii i fraze din lucrri scrise de vorbitori nativi ai acestei limbi. Pentru studenii F.I.L.S. s-a definit i ilustrat n detaliu terminologia specific limbii engleze. n prima parte se descriu fenomenele de baz din dinamica rotorilor, rspunsul dinamic al rotorilor simpli n lagre rigide i lagre elastice, precum i principalele etape ale unei analize de dinamica rotorilor. n aceast a doua parte se prezint modelarea cu elemente finite a sistemelor rotor-lagre, lagrele hidrodinamice i etanrile cu lichid i gaz, precum i instabilitatea rotorilor. n partea a treia se trateaz lagrele cu rulmeni, echilibrarea rotoarelor, msurarea vibraiilor pentru supravegherea funcionrii mainilor i diagnosticarea defectelor, standarde i recomandri privind limitele admisibile ale vibraiilor mainilor, precum i elemente de dinamica mainilor cu mecanism biel-manivel i vibraiile conductelor aferente. Nu se trateaz vibraiile paletelor, discurilor paletate i ale roilor centrifugale. Iulie 2009 Mircea Rade
Contents
Preface Contents 5. Finite element analysis of rotor-bearing systems
5.1 Rotor component models 5.2 Kinematics of rigid body precession
5.2.1 Main reference frames 5.2.2 Rigid-body precession 5.2.3 Small rotations of the spin axis
i iii 1
1 3
3 3 4
5.3 Equations of motion for rotor components

5.3.1 Thin disks 5.3.2 Uniform shaft elements 5.3.3 Bearings and seals 5.3.4 Flexible couplings
7
7 13 31 35
5.4 System equations of motion

5.4.1 Second order configuration space form 5.4.2 First order state space form
36
37 40
5.5 Eigenvalue analysis

5.5.1 Right and left eigenvectors 5.5.2 Reduction to the standard eigenvalue problem 5.5.3 Campbell and stability diagrams
40
41 42 43
5.6 Unbalance response

5.6.1 Modal analysis solution 5.6.2 Spectral analysis solution
47
47 48
5.7 Kinematics of elliptic motion

5.7.1 Elliptic orbits 5.7.2 Decomposition into forward and backward circular motions 5.7.3 Variable angular speed along the ellipse
49
49 52 54
5.8 Model order reduction
56
iv
5.8.1 Model condensation 5.8.2 Model substructuring 5.8.3 Stepwise model reduction methods
DYNAMICS OF MACHINERY
56 62 68
References
76
6. Fluid film bearings and seals

6.1 Fluid film bearings 6.2 Static characteristics of journal bearings
6.2.1 Geometry of a plain cylindrical bearing 6.2.2 Equilibrium position of journal center in bearing
77
77 78
79 80
6.3 Dynamic coefficients of journal bearings 6.4 Reynolds equation and its boundary conditions
6.4.1 General assumptions 6.4.2 Reynolds equation 6.4.3 Boundary conditions for the fluid film pressure field
83 84
86 87 89
6.5 Analytical solutions of Reynoldsequation

6.5.1 Short bearing (Ocvirk) solution 6.5.2 Infinitely long bearing (Sommerfeld) solution 6.5.3 Finite-length cavitated bearing (Moes) solution
90
90 99 99
6.6 Physical significance of the bearing coefficients 6.7 Bearing temperature

6.7.1 Approximate bearing temperature 6.7.2 Viscosity-temperature relationship
103 107
108 109
6.8 Common fluid film journal bearings

6.8.1 Plain journal bearings 6.8.2 Axial groove bearings 6.8.3 Pressure dam bearings 6.8.4 Offset halves bearings 6.8.5 Multilobe bearings 6.8.6 Tilting pad journal bearings 6.8.7 Rayleigh step journal bearings 6.8.8 Floating ring bearings
112
112 112 114 116 117 124 128 129
6.9 Squeeze film dampers

6.9.1 Basic principle
130
130
CONTENTS
6.9.2 SFD design configurations 6.9.3 Squeeze film stiffness and damping coefficients 6.9.4 Design of a squeeze film damper 132 133 135
6.10 Annular liquid seals

6.10.1 Hydrostatic reaction. Lomakin effect 6.10.2 Rotordynamic coefficients 6.10.3 Final remarks on seals
137
138 139 146
6.11 Annular gas seals 6.12 Floating contact seals

6.12.1 Design characteristics 6.12.2 Seal ring lockup 6.12.3 Locked-up oil seal rotordynamic coefficients
147 150
151 154 155
References
159
7. Instability of rotors
7.1 Whirling of rotating shafts 7.2 Instability due to rotating damping
7.2.1 Planar rotor model 7.2.2 Qualitative effect of damping 7.2.3 Whirl speeds of rotor with rotating damping 7.2.4 Quantitative effects of damping
161
161 164
165 166 168 172 173 174 176 178 179 187 189 194 199
7.3 Whirl in hydrodynamic bearings

7.3.1 Oil-whirl and oil-whip phenomena 7.3.2 Half frequency whirl 7.3.3 Onset speed of instability 7.3.4 Crandalls explanation of journal bearing instability 7.3.5 Stability of linear systems 7.3.6 Instability of a simple rigid rotor 7.3.7 Instability of a simple flexible rotor 7.3.8 Instability of complex flexible rotors
7.4 Interaction with fluid flow forces

7.4.1 Steam whirl 7.4.2 Impeller-diffuser interaction
199
199 201
vi
7.5 Dry friction backward whirl
7.5.1 Rotor-stator rubbing 7.5.2 Dry friction whirl
DYNAMICS OF MACHINERY 203

203 204
7.6 Instability due to asymmetric factors

7.6.1 Parametric excitation 7.6.2 Shaft anisotropy 7.6.3 Asymmetric inertias 7.6.4 Finite element analysis of asymmetric rotors
206
207 207 219 226
References
229
Index
233
5.
FINITE ELEMENT ANALYSIS OF ROTOR-BEARING SYSTEMS
This chapter presents the background of finite element techniques used for the prediction of rotordynamics behavior. The inertia, gyroscopic and stiffness matrices are established for axi-symmetric disks and uniform shaft segments. Journal bearings are described by linearized models with generally non-symmetric stiffness and damping matrices. Seals, dampers and couplings are also modeled by corresponding matrices. The forcing vectors due to unbalance are established for constant angular velocity. Model order reduction and condensation techniques are also considered.
5.1 Rotor component models

For the analytic prediction of the rotordynamic response, the main components of a machine with rotating assemblies must be first identified and modeled. The actual system (Fig. 5.1, a) is replaced by a physical model (Fig. 5.1, b), usually a discrete-parameter model, comprising: a) shaft segments (Bernoulli/Timoshenko, uniform/conical); b) bearings (isotropic/orthotropic, undamped/damped); c) disks (rigid/flexible, thin/thick); d) seals; e) dampers (squeeze-film); f) couplings; and g) pedestals (plus the static support structure). A set of mathematical equations consistent with the modeling assumptions is then generated for each component of the system. In the following, the inertia, gyroscopic and stiffness matrices are established for circular disks and axi-symmetric shaft segments, starting from expressions of the kinetic and potential energies, and using Lagranges equations:
d T T U + = Qi , &i qi qi d t q
i = 1, 2,.., n ,
(5.1, a)
&i where T kinetic energy, U potential energy, qi generalized displacement, q generalized velocity, and Qi generalized force. The response analysis of general unsymmetrical rotors is beyond the aim of this introductory presentation.
b
Fig. 5.1 (from [5.1])
In actual situations, the generalized forces are derived by identifying physically a set of generalized coordinates and writing the virtual work in the form
W = Qk qk .
k =1
(5.1, b)
Rotary inertia and shear deformations are considered for axi-symmetrical shafts, while external and internal damping is neglected. Bearings are described by linearized models with generally non-symmetric stiffness and damping matrices. The forcing vectors due to unbalance are established considering constant angular velocity. Then, the system equations of motion in fixed coordinates are set up. Results of eigenvalue analysis and unbalance response calculations are presented in a form useful for engineering analysis.
5. FEA OF ROTOR-BEARING SYSTEMS
5.2. Kinematics of rigid body precession

In order to write the equations of motion of a rigid disk attached to a flexible shaft, it is necessary first to determine the components of its angular velocity.
5.2.1 Main reference frames
The analysis is carried out with respect to two reference frames: X,Y,Z an inertial frame and x,y,z a frame rotating with the rotor (Fig. 5.2). The analysis is restricted to the case when the disk rotates with constant angular velocity (rad/sec) about the spin axis Ox.
Fig. 5.2
5.2.2 Rigid-body precession
Two possible motions of the rigid disk are shown in Fig. 5.3.
a
Fig. 5.3
The motion in which the Ox axis traces out a cone (Fig. 5.3, a), with angular velocity , is called a precession when point M moves along a closed orbit, and whirling when it moves along a spiralling orbit. In forward precession > 0 , while in backward precession < 0 . In synchronous precession = . In some papers, whirling is used instead of precession. Fluctuations in the cone angle, superposed on precession (Fig. 5.3, b) are generally termed nutation. This is not encountered in most linear systems.
5.2.3 Small rotations of the spin axis
The relative motion of the x,y,z frame with respect to the stationary X,Y,Z frame can be described through a set of three angles , and (Fig. 5.4), corresponding to three successive rotations. Note that the order is important, because the rotations are not commutative. In the following, the first rotation is about the transverse axis which first follows the spin axis in accord with the righthand rule for rotations.
Fig. 5.4
It is supposed that and are small precession angles, while = t , where the disk angular speed of rotation = const . Consider the following sequence of rotations [5.2]: a) a rotation around the Y-axis, transferring the Z-axis into the z1 -axis and the X-axis into the x1 -axis ; the X,Y,Z frame becomes the x1 ,Y , z1 frame;
b) a rotation around the z1 -axis which transfers the Y -axis into the y1 axis and the x1 -axis into the x -axis; the x1 ,Y , z1 frame becomes the x , y1 , z1 frame; c) a rotation about the Ox axis: the x , y1 , z1 frame becomes the x , y , z frame.
b
Fig. 5.5
The axes X,Y,Z have unit vectors I , J , K , the axes x1 , y1 , z1 have unit vectors i1 , j1 , k1 , and the axes x,y,z have unit vectors i , j , k .
& around OY) (Fig. 5.5, a) a) Rotation (angular velocity
The relation between the unit vectors is

I cos = K sin In expanded form sin i1 1 i1 . cos k1 1 k1
I J K
1 = 0
0 i1 1 0 J = T 0 1 k1
[ ]
i1 J . k 1
& around Oz1 ) (Fig. 5.5, b) b) Rotation (angular velocity

The relation between the unit vectors is i1 cos = J sin In expanded form sin i 1 cos j1 i . 1 j1
i1 1 J = k 0 1
1 0
0 i 0 j1 = T 1 k1
[ ]
i j1 . k 1
& = around Ox) (Fig. 5.5, c) c) Rotation (angular velocity
The relationship between the unit vectors is

j1 cos = k1 sin sin j . cos k = [ T i ] j k .
In expanded form 0 i 1 j1 = 0 cos k 0 sin 1 0 sin

i j cos k
The relationship between the unit vectors in the stationary and rotating frames is of the form
I J K = [T i ] j k . .
where the transformation matrix
[ T ] = [ T ][ T
1 sin cos cos + sin [ T ] = cos sin sin cos
The vector of the instantaneous angular velocity is

& J + & k1 . = i +
In terms of the unit vectors of the x,y,z system of coordinates

J = i + cos j sin k ,
k1 = sin j + cos k ,
so that & ) i + ( & cos + & sin ) j + ( & cos & sin ) k . = ( + The components of the angular velocity in the rotor-fixed (mobile) x,y,z reference system are
x y z
& + & cos + & sin = & sin & cos
(5.2)
These will be used in the expression of the disk kinetic energy.
5.3 Equations of motion for rotor components

In this section, the equations of motion of disks, shaft segments, bearings, seals and couplings are set up, together with the corresponding element matrices.
5.3.1 Thin disks
In the following, only axi-symmetric rotors (hence disks) are considered and x,y,z are principal directions of inertia.
5.3.1.1 Inertia and gyroscopic matrices of rigid disks
In the fixed frame X,Y,Z, the disk position is defined by two translations ( v and w) and two rotations ( and ) (Fig. 5.6, a), the displacements of the shaft cross-section at the disk attachment. The angles and are approximately equal to the angular displacements collinear with the Y- and Z-axis, respectively. Neglecting any unbalance, the kinetic energy of a rigid axi-symmetric disk is
Td = 1 1 2 2 2 &2 + w &2 + md v J x x + J y y + J z z , (5.3) 2 2 and J y = J z = J T are principal moments of inertia with respect to
where J x = J P
the x,y,z coordinate frame fixed to the disk and md is the mass of the disk, J P is the axial (polar) mass moment of inertia and J T is the diametral mass moment of inertia. Substituting x , y , z from (5.2) into (5.3), the kinetic energy becomes
Td =
1 1 &2 + w & ) 2 + & 2 + J P ( + md v 2 2 1 & cos + & sin ) 2 + ( & cos & sin + JT ( 2
)2]
or
8
Td =
1 1 1 &2 + w & + & 2 2 + J T & 2 + &2 . & 2 + J P 2 + 2 md v 2 2 2
Neglecting the term

Td =
1 & 2 2 , the kinetic energy becomes JP 2
1 1 1 &2 + w & 2 + & 2 + J P & + J P 2 . & 2 + JT md v 2 2 2

1 &2 + w &2 md v 2
(5.4)
In equation (5.4), the term translation,
accounts for rectilinear - for gyroscopic
1 & 2 + & 2 - for rotary inertia, J P & JT 2 1 coupling and J P 2 - for rotation around the spin axis. 2
a
Fig. 5.6
Applying Lagranges equations, the equations of motion of the rigid circular disk are obtained in the form
= md & & = Ty , v d d T T d && J P & = Mz , = JT & dt d d T = md w && = Tz , & d t w

d dt T d &&) J P & = M y , = J T ( & ) ( T d =0, w T d =0, ( )
d d T & dt v
(5.5, a)
(5.5, b) (5.5, c) (5.5, d)
T d =0, v
where Ty , Tz and M y , M z are the components of the applied force and couple, respectively (Fig. 5.6, b). In matrix form md 0 0 0 0 JT 0 0 0 0 md 0 0 0 0 JT & v & && + w && && - 0 0 0 0 0 0 0 JP & Ty 0 0 v & 0 JP Mz = & Tz 0 0 w & 0 0 - My .
Introducing the 2 1 state vectors in the X-Y and X-Z planes, respectively,
v { u }= ,
d y
w { u }= ,
d z
and the corresponding 2 1 forcing vectors acting on the disk in the two planes T { f }= M
d y y
, z
T { f }= M
d z z
, y
equation (5.5) becomes

md 0 [ 0 ] [ g ] && } & } [ ] [ 0 ] {u {u { f } + = (5.6) [ ] [ m ] [ ] [ ] g 0 & & & { } { } { } u u f
d y d d y d y d d z d d z d z
with the forcing right hand term including mass and static misalignment unbalance, interconnection forces and other external effects on the disk. The submatrices
[ m ] = m0
d
0 , JT
[ g ] = 0 0
d
0 , JP
(5.7)
form the inertia and gyroscopic submatrices, respectively, of an axi-symmetric rigid disk.
5.3.1.2 Unbalance force vectors
For the analysis of the synchronous response of a rotor system, it is important to determine the vectors of mass unbalance and static angular misalignment of a rigid disk.
10 Mass unbalance vectors
Consider the disk shown in Fig. 5.7, a. The mass center G of the disk is located at an eccentricity CG = a from the geometric center C. With respect to the rotating frame x,y,z it is located at ac = a cos , as = a sin (Fig. 5.7, b). At t = 0 (at rest), is the angle of CG with the Y-axis.
a
Fig. 5.7
The unbalanced centrifugal force acting in G along CG is md a 2 . The components along the (spinning) rotor-fixed reference system are md ac 2 and
md as 2 , and along the (inertial) fixed reference system are

f y = md ac 2cos t md as 2sin t , f z = md as 2 cos t + md ac 2sin t . (5.8)
The vector of unbalance forces (neglecting disk skewness) can be written

{f } { f }= {f }
d d y d z
a s 2 ac 2 0 0 cos t + md sin t . (5.9) = md 2 2 ac as 0 0
Permanent disk skew vector
A disk misaligned with its driving shaft receives active gyroscopic moments which force pitch changes in the disk analogous to the translations set up by centrifugal forces due to mass unbalance. Consider a rigid disk (Fig. 5.8, a) which is not perpendicular to the shaft. There is a static angular misalignment between a principal axis of inertia of the disk and the shaft axis.
11
The disk skew can be defined by a rotating vector normal to the spin axis and to the line of maximum skew angle. At t = 0 the vector makes an angle with the rotating z axis, so that the disk skew vector has components c = cos and s = sin in the rotating frame x,y,z (see also eq.(2.84) in Ch. 2).
The components of angular displacements (Fig. 5.8, b) become
y = sin ( t + ) ,
where and are elastic rotations.
z = + cos ( t + ) ,
a
Fig. 5.8
The corresponding angular velocities and accelerations are

&z = & sin ( t + ) , &y = & cos ( t + ) ,
&&y = && + 2 sin ( t + ) , &&z = && 2 cos ( t + ) .
(5.10)
From equations (5.5, b) and (5.5, d), replacing by y and by z , we obtain

&&z + J P &y = Mz, JT &&y J P & z = M y . JT
Substitution of (5.10) yields

&&) J P & + (J P J T ) 2 sin ( t + ) . M y = J T ( && J P & + (J P J T ) 2 cos ( t + ) , M z = JT
The vector of gyroscopic moments acting on the disk is

Mz 2 = (J T J P ) M y cos sin cos t + sin t . (5.11) sin cos
12
The vector of unbalance forces and moments is

md ac (J J ) P c = 2 T m a d s ( J T J P ) s 2 cos t + md as ( J J ) T P s m a d c ( J T J P ) c sin t .
{f }
d
(5.12)
Note that the moment of inertia difference J T J P can be both negative (thin disk) and positive (thick disk). When J T J P < 0 , the applied moment is restoring, acting to pitch the disk in the direction opposite the initial skew. When J T J P > 0 , the applied moment increases the angular unbalance by pitching the thick disk further in the direction of the initial skew.
5.3.1.3 Flexible disks
Flexible disks can be modeled as two rigid disks connected by springs with rotational stiffness k R (Fig. 5.9).
Fig. 5.9
The inner disk has four degrees of freedom, two translatios v , w , and two rotations , . The inertia properties are m, J P , J T . The outer disk has only two rotational degrees of freedom, and . It has only polar and diametral mass moments of inertia J P and J T . Its mass is lumped in the inner disk. The corresponding 6 6 matrices [5.1] can be reduced to 4 4 matrices by a static condensation, selecting the coordinates of the inner disk as active and those of the outer disk as omitted (see Section 5.8.1.2).
13
5.3.2 Uniform shaft elements
Shaft segments are considered isotropic and axi-symmetric about the spin axis of the rotor. This presentation is restricted to uniform cross-section elements.
5.3.2.1 Timoshenko vs. Bernoulli beam elements
Consider a two-noded uniform shaft segment (Fig. 5.10) modeled by a Bresse-Timoshenko beam element. Dropping the element index, the following notations are used: l - length of shaft element, E Youngs modulus, G shear modulus, E I - flexural rigidity, G As - effective shear rigidity, As - reduced shear area, - shear coefficient, mass density of shaft material, = A mass per unit length, I second moment of = I rotary mass per unit length. area of cross section,
Fig. 5.10
At a given node, the shaft has four degrees of freedom, two transverse displacements v and w , and two rotations and , measured in the fixed coordinate system. Define two 4 1 sub-vectors of nodal displacements in the X-Y and X-Z planes, respectively,
{u }
s y
vi i = vj j
{u }
s z
wi i = w j j
(5.13, a)
and the corresponding vectors of nodal forces
14
{f }
s y
Ty i M zi = Ty j Mz j
{f }
s z
Tz i M yi = T zj Myj
(5.13, b)
Note the minus sign of rotations and moments about the Y-axis whose positive signs are in accordance with the right-hand rule. The sign convention used here is that positive internal forces act in positive (negative) coordinate directions on beam cross sections with a positive (negative) outward normal. In the Bernoulli-Euler beam theory, deformations due to shear are neglected. The Bresse-Timoshenko beam theory is based on the following hypotheses: a) planar cross sections remain undistorted and plane; b) warping is neglected; and c) an average shear strain is considered, independent of Y. The planar section hypothesis introduces additional fake stiffness against warping. In some points, distortions due to shear are underestimated. The relationship for T is not correct. The solution is to use an effective shear area As < A . From energy considerations
1 2 so that
l s
1 2 d A G dx = 2 A
2
T2 dx , GAs
( dA ) A =
2 dA
= A.
For a hollow steel shaft, with outer diameter D and inner diameter d, the shear coefficient is 1 = . 2 d D 1.13 + 3.03 2 1 + (d D )
Equations in the X-Y plane
At any point in the beam cross section, the axial displacement is proportional to the distance from the neutral axis
u = y .
The average shear strain (Fig. 5.11) is
15
xy =
The axial strain is
u v + = + v . y x
u = y . x The bending moment is
x =
M z = x y dA = E I z .
A
The equations in the X-Y plane are presented in Table 5.1. && term, a convenient static relationship can be Neglecting the established between v and
v = -
EI z , GAs
v
(5.14) and are kinematically
though, for the Bresse-Timoshenko beam, independent.
Fig. 5.11
Introducing the dimensionless variable
=
so that
x , l
d d d 1 d = = , d x d d x l d
the static relationship between v and becomes [5.3]
16
Table 5.1 Bernoulli-Euler beam a) Equilibrium Bresse-Timoshenko beam
+ Ty = 0, Mz
&& = 0 , + Ty Mz & = 0. + & Ty v

b) Elasticity
& = 0. + & Ty v
M z = E I z xy c) Kinematics
M z = E I z xy , Ty = G As xy .
xy = ,
0 = v - . Elimination of xy
xy = , xy = v - .
Elimination of xy and xy M z = E I z , Ty = G As (v - ) Equations of motion
M z = E I z = E I z v
Elimination of M z and Ty
Elimination of M z and Ty
&=0 E I z v v + & v
&& = 0, E I z G As ( - v) & = 0. G As ( - v) + & v
17
d 2 1 dv = l d 12 d 2
where
(5.14, a)
= 12
EIz . G As l 2
(5.15)
Equations in the X-Z plane
The axial displacement at any point is

u =z .
The average shear strain (Fig. 5.12) is
xz =
The axial strain is
u w + = + w . z x
x =
u = z . x
The bending moment is

M y = x z dA = E I y .
A
The equations in the X-Z plane are presented in Table 5.2.
Fig. 5.12
&& term, the static coupling between w and is Neglecting the
w = +
EI y GAs
= ( )
EI y GAs
( )
(5.16)
18
Table 5.2 Bernoulli-Euler beam Bresse-Timoshenko beam a) Equilibrium
M y + Tz = 0 ,
&& = 0 , M y + Tz && = 0. Tz w
&& = 0. Tz w
b) Elasticity
M y = E I y xz M y = E I y xz , Tz = G As xz .
c) Kinematics
xz = , = w.
Elimination of xz
M y = E I y = E I y w
xz = , xz = + w.
Elimination of xz and xz
Tz = G As ( + w) M y = E I y ,
Equations of motion Elimination of M y and Tz

&& = 0 E I y w v + w
Elimination of M y and Tz
&& = 0. G As ( + w) w && = 0, E I y G As ( + w)
19
Introducing the dimensionless variable
x , l
the static relationship between w and becomes
d 2 ( ) 1 dw = ( ) l d 12 d 2
where
(5.16, a)
= 12
EIy G As l 2
I y = Iz .
5.3.2.2 Coordinates and shape functions
Consider third-degree polynomials as approximating functions for the displacements and rotations in the X-Y plane
v ( ) = A3 3 + A2 2 + A1 + A0 ,
( ) = B3 3 + B2 2 + B1 + B0 .
Substitution in the static coupling condition (5.14, a)
1 3 A3 2 + 2 A2 + A1 = B3 3 + B2 2 + B1 + B0 ( 6 B3 + 2 B2 ) 2 l
yields
B3 = 0 , B2 =
3 2 1 A3 , B1 = A2 , B0 = A1 + A3 , l l l 2
so that ( ) is of second degree. The translation v and the rotation are approximated in terms of the nodal displacements, using static shape functions:
v ( ) = N1 ( ) vi + N 2 ( ) i + N 3 ( ) v j + N 4 ( ) j = N u s y , ~ ~ ~ ~ ~ ( ) = N1 ( ) vi + N 2 ( ) i + N 3 ( ) v j + N 4 ( ) j = N u s y ,
{ } { }
(5.17)
where
N = N1 N 2 N 3 N 4 ,
N = N1
~ ~
~ N2
~ N3
~ N4 .
20
An example of calculation is given here for N 3 ( ) . According to the general properties of shape functions, N 3 has a unit value at coordinate 3 and is zero at coordinates 1, 2 and 4. The four boundary conditions yield a set of four equations v (0) = 0 A 0 = 0 , v (l ) = 1 A3 1 + A2 = 1 , 2
(0) = 0 A1 =
A3 ,
(l ) = 0 3 A3 + 2 A2 = 0 ,
wherefrom the four constants are obtained as

A3 =
2 , 1+
A2 =
3 , 1+
A1 =
1+
A0 = 0 ,
so that
N 3 ( ) =
1 2 3 + 3 2 + . 1+
The shear-modified shape functions for displacements are 1 1+ 1 N 2 ( ) = 1+ 1 N 3 ( ) = 1+ 1 N 4 ( ) = 1+

N1 ( ) =
[ 1 3 + 2 + ( 1 ) ] , l l ( 2 + ) + 2 ( ) , ( 3 2 + ) , l l ( + ) + 2 ( + ) .
2 3 2 3 2 2 3 2 3 2
(5.18)
For = 0 , the above shape functions become the third-degree Hermite polynomials used for Bernoulli-Euler beams. The shape functions for rotations are [5.3] 1 ~ N1 ( ) = 1+ 1 ~ N 2 ( ) = 1+ 1 ~ N 3 ( ) = 1+ 1 ~ N 4 ( ) = 1+ For = 0
1 2 l 6 6 ,
2
[1 4 + 3 + ( 1 ) ] , 1 ( 6 + 6 ) , l ( 3 2 + ) .
2 2
(5.19)
21
1 N i ~ ( i = 1,..,4 ) . N i ( ) = , l It is useful to introduce a third set of shape functions defined by ~ = N N l d Their expressions are 1 dN
which will be used in the derivation of the stiffness matrix.

( ) = N ( ) = 1 , N 1 3 1+ l 1 ( ) = N ( ) = N . 2 4 2 1+
(5.20)
Similar relations hold for the displacements in the X-Z plane ~ s = N uz ,
w=N
{ u z },
s
{ }
(5.21)
because of the similar coupling relationship (5.16, a) between w and .

5.3.2.3 Inertia and gyroscopic matrices
For an infinitesimal uniform shaft element of length d x , the kinetic energy can be obtained from the similar expression (5.4) derived for a thin disk 1 1 1 &2 + w & 2 + & 2 + I P 2 + 2 & &2 + I dT s = A v dx , 2 2 2 where
P = IP = = I , I P = 2I , = A , 1 s JP . l
(5.22)
Integrating, the kinetic energy for the shaft element becomes
Ts =
( v&
0
& 2 dx + +w
(&
0
2
&2 +
1 s 2 P dx + JP + 2
& d x .
0
a) The translatory energy is T1s 1 = 2
( v&
0
& 2 dx . +w
22
Expressing velocities in terms of nodal coordinates and shape functions
&2 = v &T v &= u &s v y
{ }
{ } & }, N {u
s y
& = N u &T = u &s &s v y =v y

&2 w
{ } N , & } N N {u & }, = {u
T T
s T z T s z
T l
this energy can be written

T1s 1 s &y = u 2
{ } N
T
N d x
10 44 4 244 4 3
{ } { } N
&s u y 1 s &z + u 2
& z }. N d x { u
s
[ ]
s mT
10 44 4 244 4 3
[m ]
s T
The translational mass submatrix of the shaft element
[ ] = l N ( )
s mT 0
N ( ) d
(5.23)
is the same in the X-Y and X-Z planes. b) The rotary energy is T2s 1 = 2
(&
0
& 2 dx . +
Expressing angular velocities in terms of nodal coordinates and shape functions ~ ~ s & = N u & = N u &s &z , y , &2
{ } ~ ~ & } N N { u & }, ={u

s T y T s y
&2
{ } ~ ~ & } N N { u & }, ={u

s T z T s z T
this energy can be written T2s 1 s &y u = 2
{ }
T
~ T ~ N N d x
10 442443
~ ~ & }. { } { } N N d x { u &s u y + 1 s &z u 2

T s z
[m ]
s R
10 442443
[m ]
s R
The rotational mass submatrix of the shaft element
23
~ ~ l N ( ) N ( ) d [ ]=
s mR T
0
(5.24)
is the same in the X-Y and X-Z planes. c) The gyroscopic effect energy is T3s & d x , P =
0
T3s
{ }
s &z u
04 14 4 2444 3
N N d x
~
T
{ u }.
s y
[ gs ]
The gyroscopic submatrix of the shaft element ~ ~ l N ( ) N ( ) d = 2 [ m ]. [ g ]=

s T P s R
0 1
(5.25)
The total kinetic energy of the uniform shaft element is

Ts = 1 s &y u 2 1 & } [ m ] {u & } {u & } [ g ] {u }+ J { } [ m ] {u& }+ 1 { u 2 2
T s s y s T z s s z s T z s s y s 2 P
where
[ m ] = [ m ]+ [ m ] .
s s T s R
(5.26)
The total mass submatrix of the uniform shaft element is
[ ]
where
s s m11 m12 s m22 ms = sym
s m13 s m23 s m33
s m14 s m24 . s m34 s m44
(5.27)
s m11 = 156 + 294 + 140 2 T + 36 R , s m12 s m13

2
( ) = ( 22 + 38.5 + 17.5 ) l + ( 3 15 ) l = ( 54 + 126 + 70 ) 36 ,

T
2
R,
24
s m14 = 13 + 31.5 + 17.5 2 l T + ( 3 15 ) l R , s m22 = 4 + 7 + 3.5 2 l 2 T + 4 + 5 + 10 2 l 2 R , s m24 = 3 + 7 + 3.5 2 l 2 T 1 + 5 5 2
) )l
2
R,
s s s s s s s s m23 = m14 = m11 = m12 = m22 , m33 , m34 , m44 ,
T =
420 ( 1 +
)2
R =
30 l 2 ( 1 +
s g14 s g 24 . s g 34 s g 44
)2
The gyroscopic submatrix of the uniform shaft element is
[ ]
where
s g11 = 72 R , s s g14 = g12 ,
s s g11 g12 s g 22 gs = sym
s g13 s g 23 s g 33
(5.28)
s s g12 = 2 ( 3 15 ) l R , g13 = 72 R , s g 22 = 2 4 + 5 + 10 2 l 2 R ,
s g 24 = 2 1 + 5 5 2 l 2 R , s s s s s s s s g 23 = g12 = g11 = g 23 = g 22 , g 33 , g 34 , g 44 .
Substitution of T s in Lagranges equations yields ms 0 or [ 0 ] [ g ] && } & } [ ] [ 0 ] {u {u { f } + = [ ] [ m ] && } & } [ g ] [ 0 ] {u {u { f }

s y s s y s y s s z s s z s z
&& } + [ G ]{ u & } = { f }, [ M ]{ u (5.29) where the 8 8 inertia matrix [ M ] is symmetric and the 8 8 gyroscopic matrix [G ] is skew-symmetric .
s s s s s
5.3.2.4 Stiffness matrix
For an infinitesimal shaft element, the strain energy is
25
d 2 d 2 1 2 2 + G As xy + + xz E I 2 d x d x Integrating and substituting the shear strains we obtain dU =
) dx .

EI U= 2
(
0
G As dx + 2
[ ( v )
0
+ ( + w) 2 d x .
a) The flexural energy is EI U1 = 2
(
0
EI dx = 2
EI dx + 2
2
( )
0
dx .
Expressing rotations and curvatures in terms of nodal coordinates and shape functions 1 ~ ~ = N u s = N us y , y , l where ( ) = d ( ) , d
{ }
{ }
2 =
1 l2
~ ~ { u } N N { u },
s T y
T
s y
the bending energy can be written
1 s U1 = uy 2
{ }
EI ~ T ~ N N d l 044 14 2444 3
s] [kB
{ } { }
us y 1 s + uz 2
EI ~ T ~ s N N d u z . l 044 14 2444 3
s] [kB
{ }
The stiffness submatrix for bending is
[ ]
s kB
EI = l
~ T ~ dN dN d . d d
(5.30)
b) The shear energy is G As U2 = 2 where
( v) d x + G As 2
2
( + w) 0
dx
26
v =
~ ~ 1 s { u s }, (N N N { u y }= N N ) { u sy }= y l
+ w = N N
~ so that
1 U2 = us y 2
{ u s }, ) { u zs }= N z
{ }
T N d u s + 1 u s G As l N y z 2 04 144 2444 3
N d {u }. { } { } G A l N
T T s s z 04 144 2444 3
[ ]
s kS
[ k Ss ]
The stiffness submatrix for shear is N d . [ ] = G A l N

s kS T s 0 1
(5.31)
The total strain energy of the uniform shaft element is

U= 1 s uy 2
{ } [ k ] { u }+ 1 { u } [ k ]{ u }, 2
T s s y s T z s s z
where the stiffness submatrix is
[ k ] = [ k ]+ [ k ].
s s B s S
(5.32)
or
[k ]
s
1 EI = 1 + l3
6l 12 4l 2 sym
12 6l 6l 2l 2 + 12 6l 4l 2
0 0 l2 sym
0 0 l2 . 0 0 l2 0
Substitution of U in Lagranges equations yields ks 0 or

s
[ ] [ 0 ] {u } {f } = [ ] [ k ] { } { } u f
s y s y s s z s z s s
where K s is the 8 8 full stiffness matrix of the uniform shaft element.
[ ]
[ K ]{ u }= { f },
27
5.3.2.5 Unbalance force vectors
Consider a linear distribution of the mass unbalance along the shaft element, between acL , asL at the left end, and acR , asR at the right end
s ( ) = acL ( 1 ) + acR , ac s ( ) = asL ( 1 ) + asR , as
(5.33)
where
acL = a L cos L , asL = a L sin L , acR = aR cos R , asR = a R sin R ,
aL , a R are the unbalance radii, and L , R are the unbalance phase angles.
The corresponding distributed unbalance forces are
p y ( ,t ) = 2 a s y ( ,t ) ,
where
s ( ,t ) , p z ( ,t ) = 2 a z
s s as y ( , t ) = ac ( ) cos t a s ( ) sin t , s ( ,t ) = ass ( ) cos t + acs ( ) sin t . az
The virtual work of these forces is
W =
( v ) 0
1
py d x +
( w ) 0
pz d x ,
1
W =
{ } N
T us y
p y l d +
0 4 1 4 244 3
{ } N
s T uz
p z l d .
{ f ys }
as y
0 4 1 4 244 3
{ f zs }
1
The kinematically equivalent nodal forces are
{ } = l N
f ys
2 0
d ,
{ } = l N
f zs
2 0
s az d .
The shaft mass unbalance vector has the expression
fs y s fz
{ } { }
2 = l
1 1 s s N T ac N T a s d d 0 2 0 1 cos t + l 1 sin t . T s T s N as d N ac d 0 0
28
An example of calculation is given below
N ( )
1 0
s ac
d = N1 ( ) [ acL (1 ) + acR ]d =
= acL N1 ( ) (1 ) d + acR N1 ( ) d =
1 [ ( 42 + 40 ) acL + ( 18 + 20 ) acR ]. = 120 (1 + ) The vector of unbalance forces can be written
{f }
s
l = 120 ( 1 + )
2
acL asL [ a ] [ a ] a a cR cos t + sR sin t , asL acL [ a ] a ] [ a a sR cR
where
18 + 20 42 + 40 ( 6 + 5 ) l ( 4 + 5 ) l , [a ]= 18 + 20 42 + 40 ( 4 + 5 ) l ( 6 + 5 ) l
so that
{ Q } cos t + { Q } sin t . { f }= { { Q } Q }
s y z z
The unbalance force vector is of the form

l 2 = 120 ( 1 + ) 1 5 l l 2 6 3 7 8l 4l sin t , cos t + 1 5 2l 6l 3 7 l l 4 8
{f }
s
(5.34)
where
29
1 = ( 42 + 40 ) acL + ( 18 + 20 ) acR ,
2 = ( 6 + 5 ) acL + ( 4 + 5 ) acR ,
3 = ( 18 + 20 ) acL + ( 42 + 40 ) acR ,
4 = ( 4 + 5 ) acL + ( 6 + 5 ) acR ,
5 = ( 42 + 40 ) asL + ( 18 + 20 ) asR ,
6 = ( 6 + 5 ) asL + ( 4 + 5 ) asR , 7 = ( 18 + 20 ) asL + ( 42 + 40 ) asR ,

8 = ( 4 + 5 ) asL + ( 6 + 5 ) asR .
5.3.2.6 Rotor modeling
Figure 5.13 illustrates an example of finite element modeling for a low pressure turbine rotor (M. L. Adams, 1980).
Fig. 5.13
For the disks integral with the shaft, note the way in which the outer diameter of each shaft element is established, in order to ensure a continuous stepwise variation along the rotor length, which is important in the calculation of shaft element stiffnesses. The remaining part of each disk is considered in the calculation of the disk mass and mass moments of inertia.
30
Fig. 5.14
For monoblock rotors with integral disks (Fig. 5.14), as used in highpressure turbines, and for shrunk-on disks (Fig. 5.15), the portion of the disk which contributes to the shaft stiffness is sometimes determined using the empirical angle rule. The same approach is used to replace a portion of a shaft with a large diameter step with several steps of slowly increased diameter (Fig. 5.16).
Fig. 5.15
Very often different diameters are used for the stiffness matrix and the mass matrix of a rotor portion. For instance, the wirings of an electrical motor contribute only to the kinetic energy, but not to the potential energy. Therefore, the diameter used in the calculation of the mass matrix is greater than the diameter used in the stiffness matrix calculation.
Fig. 5.16
31
Conical shaft finite elements are available [5.4] but their presentation is outside the scope of this presentation. User supplied elements are also used for complicated geometries where the stiffness matrix is calculated by inverting the experimentally obtained flexibility matrix.
5.3.3 Bearings and seals

Radial bearings are usually modeled by translational stiffness and damping coefficients. Inertia effects are considered only for annular seals in centrifugal pumps.
5.3.3.1 Stiffness matrix of a radial bearing
The force-displacement relationship for a radial bearing (in the Y-Z plane) can be written Ty k yy = Tz k zy k yz v b v = k , k zz w w
[ ]
(5.35)
where the elements of k b are stiffness coefficients, so that kij is the spring force in the ith direction due to a unit displacement in the jth direction.
[ ]
Fig. 5.17 Consider the Y Z - reference frame (Fig. 5.17), rotated through an angle with respect to the Y Z reference frame. The transformation of displacements can be written
v cos = w sin
sin v v = [ R ] . cos w w
The transformation of forces is
32
cos Ty = Tz sin
sin cos
Ty Ty T = [ R ] T . z z
Dropping the b index, the new force-displacement relation is Ty Ty v 1 v = [ R ] = [ R ][ k ] = [ R ][ k ][ R ] . w w Tz Tz Because [ T
] 1 = [ T ] T , the transformed stiffness matrix is

c s k yy = s c k zy k yz c s , k zz s c (5.36)
[ k ] = [ R ][ k ][ R ]T
where c = cos and s = sin . The original non-symmetric stiffness matrix can be split into the sum of a symmetric and a skew-symmetric component: k yy k yz k yy k s 0 ka = , + k k zz k s k zz k a 0 zy 24 3 14243 14243 14 [ ks ] [ ka ] [k ] where
ks = 1 k yz + k zy , 2
(5.37)
ka =
1 k yz k zy . 2
(5.38)
The transformed stiffness matrix becomes ] [ k ] = [ R ] ( [ k s ] + [ ka ] ) [ R ] T = [ k s ] + [ka where, denoting c = cos and s = sin , k yy c 2 + k zz s 2 + 2k s cs [ ks ] = 2 2 k zz k yy s c + k s c s
( k zz k yy ) sc + ks ( c 2 s 2 )
k yy s 2 + k zz c 2 2k s c s
(5.39)
and These results imply that:
] = [ ka ]. [ ka
a) the skew-symmetric part [ k a ] is independent of the rotation [ T ] , hence of the angle ; b) when tan 2 = 2 ks , k yy k zz
33
the symmetric part [ k s ] is diagonal and the angles and + 900 define the principal directions of flexibility. The symmetric matrix [ k s ] can be diagonalized through the coordinate transformation [ R ] , whereas the skew-symmetric matrix [ k a ] remains unchanged. This implies that a non-symmetric stiffness matrix can be transformed into a matrix in which the off-diagonal elements are skew-symmetrical. In the special case when the diagonal elements are identical, two particular cases are of interest: a) when the off-diagonal elements are skew-symmetrical, the transformed matrix is the same as the original matrix:
c s k1 s c k 2
k 2 c s k1 = k1 s c k2
k2 k1
hence the bearing is isotropic. b) when the off-diagonal elements are identical
[ k ] =
c s k1 s c k 2
k2 k1
c s k1 + k 2 sin 2 = s c k 2 cos 2
k 2 cos 2 k1 k 2 sin 2
and for = 450
[ k ] =
k1 + k 2 0
0 k1 k 2
so that the bearing is orthotropic, with principal axes of elasticity at 450 and 1350 . The following six important conclusions can be drawn [5.5]: a) when k yz k zy , the bearing is anisotropic; b) when k yz = k zy and k yy = k zz , the bearing is isotropic; c) when k yz = k zy and k yy k zz , the bearing is orthotropic, with principal directions defined by and + 900 ; d) when k yz = k zy and k yy = k zz , the bearing is orthotropic, with principal directions at 450 and 1350 ; e) when k yz = k zy = 0 and k yy k zz , the bearing is orthotropic, with horizontal and vertical principal directions;
34
f) when k yz = k zy = 0 and k yy = k zz , the bearing is isotropic; Moreover, the condition k yz k zy (non-zero skew-symmetric coupling element k a ) is the major cause of the instability of an anisotropic system.
5.3.3.2 Fluid film bearings
Linearized hydrodynamic fluid film bearings are represented by the incremental force components obtained by a Taylor-series expansion of the forcedisplacement and force-velocity relations about an equilibrium configuration of the journal in the bearing [5.6]:
b b b b & T y k yy k yz v c yy c yz v . + = b b b b & k zy k zz w c zy c zz w Tz 14243 14243
(5.40)
[k ]
b
[c ]
b
In expanded form
kb yy 0 kb zy 0 0 kb yz 0 0
b 0 k zz 0 0
0 v cb yy 0 0 + 0 w cb zy 0 0
0 cb yz 0 0 0 cb zz 0 0
& Ty 0 v & 0 Mz = & Tz 0 w & 0 My -
In partitioned form kb yy b k zy & [ ] [ k ] } {u } [ c ] [ c ] {u {f } + = [ ] [ k ] {u } [ c ] [ c ] {u & } {f }

b yz b zz b y b yy b zy b yz b zz b y b y b z b z b z
(5.41)
where
[ k ] = k0
b ij
b ij
0 , 0
[ c ] = c0
b ij
b ij
0 , 0
( i, j = y, z ).
(5.42)
In the following, bearing/pedestal inertia effects are neglected.

5.3.3.3 Annular seals
Short annular seals in centrifugal pumps are usually considered isotropic. The dynamic seal forces are represented by a reaction force / seal motion of the form [5.7]
35
Ty K = Tz k
k K
& & v v C c v & + M . w+ c C w && & w
(5.43)
The diagonal elements of their stiffness and damping matrices are equal, while the off-diagonal terms are equal, but with reversed sign. The cross-coupled damping and the mass term arise primarily from inertial effects. Cross-coupled stiffnesses arise from fluid rotation, in the same way as in uncavitated plain journal bearings. The dynamic coefficients of equation (5.43) are defined in Section 6.10.2. The impeller-volute/diffuser interaction forces are generally modeled by an equation of the form Ty K = Tz k & M k v C c v + + & mc K w c C w & mc & v . && M w (5.44)
Note the presence of a cross-coupled mass coefficient mc , in comparison with the liquid-seal model of equation (5.43). The sign of mc is negative, which implies that it is destabilizing for forward precession. The model with identical diagonal elements and skew-symmetric off-diagonal elements ensures radial isotropy. The forces developed by centrifugal compressor seal labyrinths are at least one order of magnitude lower than their liquid counterparts. They have negligible added-mass terms and are typically modeled by a reaction force/motion model of the form Ty K = Tz k & k v C c v + . & K w c C w (5.45)
Unlike the pump seal model of equation (5.43), the direct stiffness term is typically negligible and is negative in many cases. Usually only translational coefficients are considered. Angular dynamic coefficients are used only for long annular clearance seals in multi-stage centrifugal pumps, where forces give rise to tilting shaft motions and couples produce linear displacements.
5.3.4 Flexible couplings
A flexible coupling can be modeled as an elastic element with isotropic translational stiffness kT and rotational stiffness k R between station i on one shaft and station j on the other, as shown in Fig. 5.18 [5.1]. The stiffness matrix of such a flexible coupling is of the form
36
[ k ] coupling
kT kT kR kR k k T T k kR R = kT kT kR kR kT kT kR kR
(5.46)
This simple type of coupling model allows modest amounts of relative transverse motion and angular misalignment between the centerlines of the components being coupled, but prevents any relative axial motion, so it does not apply to spline couplings.
Fig. 5.18
Inertia effects can be taken into account by including thin rigid disks at each of the connecting points.
5.4 System equations of motion

The equations of motion for the rotor-bearings-seals system are first obtained in second order form, by assembling the element matrices.
37
5.4.1 Second order configuration space form
It is convenient to use a global displacement vector whose upper half contains the nodal displacements in the Y-X plane, while the lower half contains those in the Z-X plane
{ x }T = { v1 , 1 , v2 , 2 ,L , vn , n , w1 , 1 , w1 , 1 ,L , wn , n }T . 14 444 4 244444 3 144444 4 244444 4 3

{Y }
T
(5.47)
{Z }
Correspondingly, the upper half of the global forcing vector contains the nodal forces in the Y-X plane, and the lower half has the nodal forces acting in the Z-X plane
{ f }T
= Ty1 , M z1 , Ty 2 , M z 2 ,L ,Ty n , M z n , Tz1 , M y1 , Tz 2 , M y 2 ,L , Tz n , M y n T . 144444424444443 14444444 24444444 3 { F y }T { Fz } T (5.48) One can write
{x }=
{Y } , {Z }
{ f }=
Fy . { Fz }
{ }
(5.49)
By combining the component element equations, the assembled system equations of motion can be written & } + [ C ]{ x & } + [ K ]{ x } = { f }, [ M ]{ & x where (5.50)
[M ]=
[C ]=

[ k yz ] [ k + k yy ] [m ] [0 ] , [ K ]= [ k + k zz ] [0 ] [m ] [ k zy ]
[ ] [ c yz ] [0 ] [ g ] + , [ g ] [0 ] [ czy ] [ czz ]
& Z
c yy
(5.51)
{Y } , & }= {x & { }
{Y } &} = {& x &&

&& Z
{ }
are of order N = 4n , where n is the number of nodes. The global inertia matrix is shown in Fig. 5.19. The global gyroscopic matrix is shown in Fig. 5.20. The global stiffness matrix is shown in Fig. 5.21.
38
Fig. 5.19
Fig. 5.20
Fig. 5.21
39
Matrices similar.
[ k yz ] , [ k zy ] and [ k zz ] resemble [ k yy ] . Bearing matrices are
Fig. 5.22
For the rotor model from Fig. 5.22, the equations of motion have the form
Fig. 5.23
Fig. 5.24
40
The vectors of mass unbalance forces are
{ f } = { Fc }cos t + { Fs }sin t
where the vectors in the right hand side are shown in Fig. 5.24.
5.4.2 First order state space form
(5.52)
For computational purposes, the equation of motion (5.50) is transformed into the first order state space form. Introducing an auxiliary equation
& } [ M ]{ x & } = {0 } , [ M ]{ x
equations (5.50) and (5.53) can be combined to give [C ] [M ] or
(5.53)
& } [ K ] [ 0 ] { x } { f } [ M ] {x + & } = {0 } &} [0 ] {& x [0 ] [ M ] {x
(5.54)
& } + [ B ]{ q } = { p } , [ A ]{ q
where the 2 N 2 N matrices [ A ] and [ B ] are real but non-symmetrical
(5.55)
[ A ]=
[C ] [ M ] , [M ] [0 ]
[ B ]=
[ K ] [0 ] . [0 ] [ M ]
The resulting system of equations is non-self-adjoint. Note that equations (5.50) and (5.53) can alternatively give [M ] [0 ]
[0 ] [M ]
&} {x &} {& x
[0 ] + [K ]
[M
[C ]
{x } &} {x
{0 } = {f }
(5.56)
but in the following only the form (5.54) will be used.
5.5 Eigenvalue analysis

Solving the rotordynamics eigenvalue problem, natural frequencies, damping ratios, precession modal forms and stability thresholds can be determined for damped gyroscopic systems.
41
5.5.1 Right and left eigenvectors
The eigenvalues and right eigenvectors are obtained by solving the homogeneous form of equation (5.55) & } + [ B ]{ q } = { 0 } . [ A ]{ q Assuming a solution of the form (5.57)
{ q } = { R }e t ,
equation (5.57) can be written
( [ A ]+ [ B ] ){ R }= { 0 } .
There are r = 2 N eigenvalues r obtained from
(5.58)
det ( [ A ]+ [ B ] ) = 0 ,
(5.59)
and 2 N right eigenvectors rR satisfying the generalized eigenvalue problem
{ }
[ B ]{ R }= r [ A ]{ R },
For the transposed equation & } + [ B ]T { q } = { 0 }, [ A ]T { q assume solutions of the form
(r = 1,...,2 N ) .
(5.60)
{ q } = { L }e t .
This yields
( [ A ]
+ [ B ]T
){ }= { 0 }.
L
(5.61)
The eigenvalues are given by the equation
det ( [ A ] + [ B ] ) T = 0
which has the same solutions as those of equation (5.59). Because the transposed of equation (5.61) is
42
{ } are referred to as left eigenvectors, solutions of the eigenproblem

L
{ } ( [ A ]+ [ B ] ) = 0 ,
L T
[ B ]T { rL }= r [ A ]T { rL },
(r = 1,...,2 N ) .
for r = s for r s for r = s for r s
(5.62)
The 2 N 1 right and left eigenvectors satisfy the bi-orthogonality relations
{ }
so that
L T s
[ A ] { rR }= r rs = [ B ] { rR }= r rs
r = r . r
{ }
T sL
= r 0
(5.63)
(5.64)
These bi-orthogonality relations between the modes of the original system and those of the transposed system can be used to uncouple the system equations.
5.5.2 Reduction to the standard eigenvalue problem
Equation (5.58) can be written in the form
[ R ]{ R }= 1 { R }
where
(5.65)
[ R ] = [ B ]1[ A ] = [ K ] [ C ] [I ]
[K
] 1[ M ] [0 ]
is a 2 N 2 N non-symmetric real matrix. This formulation has the drawback of giving the reciprocals of eigenvalues. The upper half of the eigenvectors yields the N 1 eigenvectors of the original problem. An alternative is to use the form (5.56). For an assumed solution
{ x } = {u } e t ,
the homogeneous form of equation (5.56) becomes
(5.66)
43
or
[M ] [0 ]
[ 0 ] {u } [ 0 ] [ M ] {u } {0 } (5.67) + = [ M ] {u } [ K ] [ C ] {u } { 0 } [ I ] [ I ] [0 ] ] [ M ] 1[ C ] [ ] [ ] I 0
{u } = {0 } . {u }
[0 ] [ M ] 1 [ K
Remember that N = 4n , where n is the number of nodes.
5.5.3 Campbell and stability diagrams
Equation (5.65) has 2 N eigensolutions, where N is the order of the system global matrices. They are purely real for overdamped modes and appear in complex conjugate pairs for underdamped or undamped modes of precession. In general, the complex eigenvalues are of the form
r = r + i r ,
r = r i r ,
(5.68)
and they are functions of the rotating assembly spin speed . The imaginary part r of the eigenvalue is the damped natural frequency of precession (whirl speed or precession speed), and the real part r is the damping constant. A stable mode requires a non-positive value for r . Often, the damping is expressed in terms of the damping ratio
r , r
2 r
or in terms of the logarithmic decrement
r = 2 r
It is common practice to plot both the precession natural frequencies and damping constants as a function of the rotor spin speed . These plots are called whirl speed maps [5.1], and one is shown in Fig. 5.25. In most practical applications, if the system becomes unstable, it is usually the first forward precession mode which yields the instability, while the remaining modes remain stable. In Fig. 5.25, the first mode is illustrated to become unstable at an onset speed of instability, oi , also called a threshold speed, th . Plots of only the precession natural frequencies versus spin speed are commonly referred to as Campbell diagrams. When points are marked on the
44
natural frequency curves, with the corresponding values of the logarithmic decrement, the plots are called Lund diagrams. Plots of only the real part of the eigenvalues versus spin speed are called stability diagrams. Examples of Campbell diagrams and stability diagrams are given for selected rotor models in Sections 3.4 and 4.5 of Part I.
Fig. 5.25
A critical speed of order of a single-shaft rotor system is defined as a spin speed for which a multiple of that speed coincides with one of the systems natural frequencies of precession. An excitation frequency line of equation = is included in Fig. 5.25. When equals r , the excitation r creates a resonance condition. One approach for determining critical speeds is to generate the Campbell diagram, include all excitation frequency lines of interest, and graphically note the intersections to obtain the critical speeds associated with each excitation. The complex conjugate eigenvectors are of the form
{ u }r = { a }r + i { b }r ,
{u }
= { a }r i { b }r .
(5.69)
The free precession solution can be written as the sum of two complex eigensolutions associated with the pair of eigenvalues and right eigenvectors
{ x (t ) }r = { u }r er t + {u }r er t = = ( { a }r + i { b }r ) e ( r + i r )t + ( { a }r i { b }r ) e ( r i r )t .
{ x (t ) }r = 2 e r t { a }r e
= 2e
rt
i r t
e i r t e i r t + e i r t + i { b }r 2 2i
( { a }r cos r t { b }r sin r t ) .
i = (5.70)
Figure 5.26 shows the nature of the motion in an undamped precession mode (node and mode indices are dropped out).
45
Fig. 5.26
The factor 2 e r t does not influence the mode shape, being a common factor for all coordinates. Therefore, the upper half of the modal vector can be approximated by
{ x }r = { uc } cos r t + { u s } sin r t ,
where
{ uc } = { a }r = Re { u }r ,
{ u s } = { b }r = Im { u }r .
This way, the orbit at any station becomes an ellipse instead of a spiral, which is taken into account by plotting incomplete (open) ellipses. An element of the rth mode of precession has the form
x j (t ) = uc cos r t + u s sin r t ,
defining a harmonic motion with frequency r . The two components of the translational motion at any station are
46
v = vc cos r t + v s sin r t , w = wc cos r t + w s sin r t ,
(5.71)
defining the parametric equations of an ellipse.
Fig. 5.27
The resulting displacement vector is

r ( t ) = v( t )+ i w( t ) .
Connecting the end points of the displacement vectors at all stations along a rotor (at a given time t), the mode shape is obtained, which is a curve in space (Fig. 5.27). Examples are given in Sections 3.4 and 4.5 of Part I. The special case of conservative gyroscopic systems is worth mentioning. If a pair of bearing principal axes exists, for an undamped gyroscopic system, equation (5.50) becomes
[m ] [0 ] &&} [ 0 ] [ g ] { Y & } [ k y ] [ 0 ] {Y } { 0 } [ 0 ] { Y && + & + = .(5.72) [ m ] { Z } [ g ] [ 0 ] { Z } [ 0 ] [ k z ] { Z } { 0 }
For an assumed solution of the form (5.66), because the system has pure imaginary eigenvalues, = i , so that equation (5.72) becomes [5.8]
ky 2[ m] i [g ]
[ ]
i [g ] { y R } + i { y I } { 0 } = , k z ] 2 [ m ] { z R } + i { z I } { 0 }
which gives four real coupled matrix equations. It can be shown that the solutions of the equations are proportional to each other
{ yI } = { yR } ,
where is a proportionality constant. The modal vector can be written
{ zR } = { zI },
{ y R } + i { y I } { y R } + i { y R } (1 + i ){ y R } { yR } = = = (1 + i ) { z R } + i { z I } { z I } + i { z I } ( + i ){ z I } i { z I }
47
and can be divided by (1 + i ) . Hence, conservative gyroscopic systems are described by pure imaginary eigenvalues, and eigenvectors are real in the X-Y plane and pure imaginary in the X-Z plane, respectively.
5.6 Unbalance response

The forced response of a rotor system can be determined either indirectly, in modal coordinates, or directly, in physical coordinates.
5.6.1 Modal analysis solution
In the case of synchronous excitation due to mass unbalance
{ f } = { F } e i t ,
equation (5.54) becomes
[C ] i [M ]
{ x } = { X } e i t ,
[M ] [K ] [ 0 ] { X } { F } + = [ 0 ] [ 0 ] [ M ] i { X } { 0 }
( i [ A ]+ [ B ] ){ Q } = { P }
or
where
{Q } =
{X } i { X } {Q } =
2N
{P }=
{F } {0 }
Assume a solution of the form
{ } ,
R r r r =1
(5.73)
where r are modal coordinates. Multiplying to the left by R
{ }
T r
and considering the bi-orthogonality
relations (5.63), the rth decoupled equation is
{ } ( i [ A ]+ [ B ] ){ } = { } { P }
L T r R r r L T r
48
or
r ( i r ) r = L
The rth modal coordinate is
{ } { P }.
T r
r =
r ( i r
{ } { P }
L T r
2N
From equation (5.73), the vector of physical coordinates is
{Q } =
whose upper half is
r =1
{ } { }
R
r ( i r
L T r
{ P },
{ X }=
where
r =1
2N
{ } { }
R
r ( i r
L T r
{ F },
(5.74)
{ }
R
and
{ }
L
are the upper halves of the corresponding modal
vectors.
5.6.2 Spectral analysis solution
Consider again equation (5.50). For a synchronous excitation
{ f } = { Fc } cos t + { Fs } sin t ,
the steady-state response has the form
(5.75)
{ x } = { X c } cos t + { X s } sin t .
Substitution of (5.75) and (5.76) into (5.50) yields
(5.76)
[ M ] ) { X c } + [ C ] { X s } = { Fc }, [ C ] { X c } + ( [ K ] 2 [ M ] ) { X s } = { Fs }.
2
( [ K ]
(5.77)
This is a linear set of equations which can be solved with known routines. The cos and sin components are substituted back in equation (5.76). The two components of the translational motion at any station are given by equations of the form (5.71), wherefrom the elliptic orbits can be determined as shown in Section 5.7.
49
5.7 Kinematics of elliptic motion

In the following, the stationary rectangular coordinate frame is x , y , z .
5.7.1 Elliptic orbits
The simplest steady motion of a rotor (nodal) point is a planar motion with the precession frequency . Its displacement components in the y and z directions are y = yc cos t + y s sin t , z = zc cos t + z s sin t . They define an elliptic orbit, as illustrated in Fig. 5.28. (5.78)
Fig. 5.28
The ellipse results by composition of two harmonic motions of different amplitudes and phase angles
50
z = Az cos ( t + z ) = Az cos z cos t Az sin z sin t ,
y = Ay cos t + y = Ay cos y cos t Ay sin y sin t ,
in two perpendicular directions. It is seen that

2 2 Ay = y c + ys
12
2 2 Az = zc + zs
12
ys , yc z tan z = s . zc tan y =
The parameters of the elliptic motion are generally a function of the spin speed . Two particular cases are of interest [5.1]: a) If y = z , the y and z motions are in phase, and the orbit reduces to a straight line (Fig. 5.29).
Fig. 5.29
b) If y s = zc (+ zc ) and yc = + z s ( z s ) the y motion leads (lags) the z motion by 900 , and the orbit reduces to forward (backward) circular motion (Fig. 5.30).
Fig. 5.30
51
The parametric equations (5.78) define an ellipse, which is inclined with respect to the y-z axes. Solving first for cos t and sin t as
) ( yc z s z c y s ) , sin t = ( z yc y zc ) ( yc z s zc y s ) ,
and then eliminating the time, the orbit equation is obtained as
cos t = ( y z s z y s
(z
2 c
2 2 2 + zs y 2 2 ( y c z c + y s z s ) y z + yc + ys z 2 = ( y s z c yc z s
)2 .
(5.79)
Equation (5.79) is more often expressed in terms of the major and minor semiaxes a and b, and the orbital inclination angle (Fig. 5.31).
Fig. 5.31
In an y1 z1 (principal) coordinate system, with the axes along the ellipse axes, the motion is described as z1 = b sin ( t + ) , y1 = a cos ( t + ) , (5.80)
where is a phase angle (inclination of radius vector at t = 0 ) so that
y1 z1 + = 1. a b
The coordinate transformation y = y1 cos z1 sin , z = y1 sin + z1 cos ,
(5.81)
(5.82)
leads to parametric equations of the form (5.78). By combining equations (5.78), (5.80) and (5.82), it is possible to obtain the four parameters a, b, and in terms of yc , y s , zc and z s . This will be done in the following by a different approach.
52
5.7.2 Decomposition into forward and backward circular motions
Elliptical orbits may be represented with the aid of rotating complex vectors, by treating the y-z plane as a complex plane with a real axis along the yaxis and the imaginary axis along the z-axis.
Fig. 5.32
The resultant of a forward rotating vector r f and a backward rotating vector rb is a rotating vector whose end moves along an elliptical orbit (Fig. 5.32):
r = r f ei t + rb e i t = r f e
i t + f
) + r e i ( t + b ) . b
(5.83)
From equations (5.78) y = yc z = zc so that r = y + iz = 1 1 1 1 = e i t ( yc + z s ) + i ( y s + zc ) + e i t ( yc z s ) + i ( y s + zc ) . 2 2 2 2 It is seen that r is the sum of a forward rotating vector with complex amplitude r f , and a backward rotating vector with complex amplitude rb , where 1 i t 1 i t e e i t , e + e i t + y s 2i 2 1 i t 1 i t e + e i t + z s e e i t , 2 2i
53
1 ( yc + z s ) + i 1 ( ys + zc ) = r f e i f , 2 2 1 1 rb = ( yc z s ) + i ( ys + zc ) = rb e i b . 2 2 rf = The amplitudes and phase angles are given by
(5.84)
rf =
1 2
( yc + z s ) 2 + ( ys + zc ) 2 ( yc z s )
2
y s + zc a+b , tan f = , 2 yc + z s y + zc tan b = s . yc z s
1 rb = 2
+ ( y s + zc )
ab , = 2
(5.85)
The major semiaxis is a = r f + rb , a2 = 1 2 2 2 2 yc + y s + zc + zs + 2 The minor semiaxis is b = r f rb = y 1 det c a zc ys . zs (5.87) (5.86)
1 2 2 2 2 yc + y s + zc + zs 4
( yc z s y s z c
)2 .
The inclination of the major axis (attitude angle) is
1 f + b . 2
(5.88)
From the condition of phase coincidence
= t + f = t + b ,
2 t = b f ,
where t is the time when P is in the point of maximum displacement amplitude. The following useful relations can be established tan 2 t = tan 2 = 2 ( yc y s + z c z s
2 yc
)
,
2 ys
2 ys
+
2 zc
2 zc
2 zs
(5.89) (5.90)
2 ( yc z c + y s z s
2 yc
2 zs
The precession is forward when r f > rb , b>0,
yc z s > y s z c .
54
The precession is backward when r f < rb , b < 0,
yc z s < y s z c .
The orbit is a straight line (degenerated ellipse) when r f = rb , b=0, yc z s = y s z c .
5.7.3 Variable angular speed along the ellipse
Consider an ellipse (Fig. 5.33) described by the parametric equations y = a cos t , z = b sin t . (5.91)
Fig. 5.33
Comparison with equations (5.71) yields

y = Ay cos t + y = a cos t ,
z = Az cos ( t + z ) = b cos t + = b sin t , 2

so that
Ay = a , y = 0 , Az = b, z =
The concentric circles method of constructing an ellipse is illustrated in Fig. 5.34. Two concentric circles are drawn with a center at O. The diameter of the large circle equals the given major axis 2a . The diameter of the small circle equals the given minor axis 2b . A diagonal which makes an angle t with the y-axis intersects the large circle in M and the small circle in P. Point B on the ellipse is plotted by projecting downward from M to intersect with the horizontal line drawn from P.
55
Point A on the y axis corresponds to t = 0 . Point B corresponds to t, and point C to t + 2 . Note that points B and C correspond to a time interval of
2 MON = 900 while the angle BOC is larger than 900 .
Fig. 5.34
, the point B moves along the ellipse with variable angular velocity.
Denoting = AOB , the angular position at time t is given by
If point M moves along the circle of radius a with constant angular velocity
tan =
z b = tan t . y a
(5.92)
The variation of as a function of t is shown in Fig. 5.35. The deviation from the straight line indicates a variable angular velocity.
Fig. 5.35
56
Indeed, the angular velocity of the motion along the ellipse is

& = d = d = dt d ( t )
b sec 2 t a . 2 b 2 1 + tan t a
(5.93)
& as a function of t is plotted in Fig. 5.36. It is seen that The variation of & varies between = (b a ) and = (a b ) so that = .
Fig. 5.36
& - that of In conclusion, the angular speed of the precession motion is not the point B along the elliptic orbit, but the speed of the points M and P along the generating circles of radii a and b, respectively.
5.8 Model order reduction

The initial finite element discretization of a rotor system needs a relatively large number of degrees of freedom (DOFs) for satisfactory accuracy. A reduction of the number of DOFs is sometimes necessary because, in actual applications, only a few lower modes are of concern, giving little justification for solving the complete equation of motion in the dynamic analysis.
5.8.1 Model condensation
Static and dynamic condensation techniques can be used to produce reduced models possessing eigensystems that approximate that of the original full system model.
57
5.8.1.1 Formalism of coordinate reduction
Consider equation (5.50) in the form

& }+ [ C ] { x & }+ [ K ] { x }= { f }. [ M ]{ & x
N N N 1 N N N 1 N N N 1 N 1
(5.94)
A rectangular transformation matrix [ T ] is seeked, which relates the N = 4n elements (coordinates) of the vector { x } to a smaller number L < N of elements (coordinates) of a vector { u }, so that
{ x } = [ T ] {u } .
N 1 N L L1
(5.95)
The transformation is time independent, so that

& } = [ T ] {u & } , {& & } = [ T ] {u && } . { x } = [ T ] {u } , { x x
(5.96)
Substituting equations (5.96) in (5.94) and premultiplying by [ T energy equivalence yields the reduced equations of motion
&& }+ [ C ] { u & }+ [ K [ M ]{ u ] { u }= { f },
red red red red L L L1 L L L1 L L L1 L1
]T , the
(5.97)
where
[M ]= [T ]
red
[ M ][ T ] , [ C red ] = [ T ]T [ C ][ T ] , [ K red ] = [ T ]T [ K ][ T ] ,
{ f }= [ T ]
red
{ f }.
A proper choice of [ T ] will drastically reduce the number of DOFs without altering the lower eigenfrequencies and the mode shapes of interest.
5.8.1.2 Guyan/Irons condensation
The basis for the Guyan/Irons reduction is to follow a standard procedure used in static structural analysis, namely, elimination of DOFs at which no forces are applied, whence the name of static condensation. The coordinates (DOFs) are partitioned into two groups: a) the active (master, retained) coordinates, and b) the omitted (slave, discarded) coordinates, denoted by a and o, respectively. Partitioning the equation (5.94) accordingly
58
[M aa ] [M ] oa
&a } [Caa ] [Cao ] {x &a } [K aa ] [K ao ] {xa } { f a } [M ao ] {& x . + + = &o } [Coa ] [Coo ] {x &o } [M oo ] {& x [K oa ] [K oo ] {xo } { f o }
(5.98) Assuming
{ f o } = { 0 } , the static force-deflection relationship reduces to

[K aa ] [K ] oa
[K ao ] {xa } { f a } = . [K oo ] {xo } {0 }
(5.99)
The lower partition provides a static constraint equation
[ K oa ] { xa } + [ K oo ] { xo } = { 0 }
which can be written
(5.100)
{ xo } = [ K oo ] 1[ K oa ] { xa }. (5.101) The original set of coordinates { x } can be related to the subset of active
coordinates by the equation
[I a ] {x } {x }= a = {xa } = [ T ]{xa }. 1 {xo } [K oo ] [K oa ]
(5.102)
Equation (5.102) may be referred to as a Ritz transformation. The Ritz basis vectors, which are columns of the Ritz transformation matrix [ T ] , are the displacement patterns associated with unit-displacement of the respective acoordinates while the o-coordinates are released
{ x } = [ T ]{xa } = [ { t1 } { t2 }
xa1 L {tL } ] L = x aL
{t j } xaj .
L j =1
(5.103)
The reduced set of equations has the form (5.97), where
[ K ]= [ K
red
{ u } = { xa } ,
aa
] [ K ao ][ K oo ] 1[ K oa ] ,
(5.104) (5.105)
[M ]= [ M
red
aa
and C red has an expression similar to (5.105).
] [ K ao ] [ K oo ] 1[ M oa ] [ M ao ] [ K oo ] 1[ K oa ] + , + [ K ao ] [ K oo ] 1[ M oo ] [ K oo ] 1 [ K oa ] ,
59
Replacing the dynamic relationship between active and omitted DOFs by a static relationship, the Guyan/Irons reduction is an incomplete extension of the Static condensation, with inherent loss of accuracy. One exception is worth mentioning: a lumped-mass model, consisting of point masses at the nodes where translational displacements are defined (mass moments of inertia neglected). With all the rotational DOFs as o-DOFs and all the translational DOFs as a-DOFs,
[ M oo ] = [ 0 ] , [ M oa ] = [ 0 ] , [ M ao ] = [ 0 ] ,
and the Guyan reduction is accurate.
[M ]= [ M
red
aa
Drawbacks: a) misapplication can lead to serious modeling errors; b) destroys the banded form of matrices; and c) requires insight, experience and skill to partition the DOFs, though automatic procedures exist for the selection of active DOFs. In conclusion, the accuracy of the refined finite element model one tries very hard to achieve may be lost by the Guyan/Irons reduction.
5.8.1.3 Use of macroelements
Rotating shafts have variable cross sections and usually it is necessary to use a large number of finite elements to obtain a good model of the rotor. The number of elements can be reduced by introducing macro-elements [5.9]. Several short cylindrical elements can be treated as one element. Formally, this is done by a static condensation, treating the interior coordinates at the steps of the cross section as o-DOFs and the boundary coordinates as a-DOFs. This increases numerical economy without loss of accuracy in the results.
a
Fig. 5.37
For the stepped shaft from Fig. 5.37, a, a macro-element is shown in Fig. 5.37, b. Considering only the motion in the Y-X plane, the 8 8 macro-element matrix has a banded form (Fig. 5.38). Reordering the nodal displacements, moving up the external DOFs selected as a-DOFs and moving down the internal DOFs selected as o-DOFs, destroys the banded form.
60
Fig. 5.38
Elimination of internal DOFs using the transformation (5.85) yields a condensed 4 4 matrix. This allows to maintain the banded structure of the system matrix (Fig. 5.39).
Fig. 5.39
5.8.1.4 Modal condensation
Consider the homogeneous part of the equation (5.50) of a damped anisotropic rotor system, written as
& } + ( [ C b ]+ [ G ] ) { x & } + ( [ K s ]+ [ K b ] ) { x } = { 0 } , [ M ]{ & x
(5.106)
where
[K ]
b
s respectively, gyroscopic matrix, [ M ] is the mass matrix, and { x } is the 4n 1 state vector.
[C ] are the stiffness and damping matrices for bearings, [ K ] is the shaft stiffness matrix, [ G ] is the (shaft+disks)
and
b
The corresponding complex eigenvalue problem yields the damped eigenfrequencies and the complex eigenvectors. For rotor systems with a large
61
number of DOFs, the complex eigenvalue procedure may run into numerical difficulties and may be time consuming. One approach to avoid some of these problems is the modal condensation method. One variant is based on the analysis of the isotropic undamped nongyroscopic part of equation (5.106) in the Y-X plane:
&& }+ [ k + k ] {Y } = { 0 }. [ m ]{ Y yy
(5.107)
The shaft is assumed symmetrical, rotor gyroscopic effects are neglected, bearing damping is neglected, and only an average bearing principal stiffness term is considered, usually the symmetric component defined in equation (5.37). This (neighboring) associate conservative system has planar undamped modes entering as columns in the modal matrix
[ ] = [ {1 } { 2 }
{ 2n } ] .
(5.108)
A truncated modal matrix is used as the transformation matrix, using only the L lower modes of the system described by (5.107). Retaining the first L columns of the matrix (5.108)
[ ] = [ { }
{ 2 }
{ L } ] .
(5.109)
{Y } [ ] [ 0 ] { u y } {u y } {x }= = [ ]{ u } = [ ] = { u } { } u { } Z [ ] [ ] 0 z z
where { u } is the reduced state vector.
The coordinate transformation is
(5.110)
Substitution of (5.110) into (5.106) and pre-multiplication by the reduced set of equations of motion (5.97)
[ ]
yields
62
where
[ M ] = [ ] [ M ] [ ] , { f }= [ ] [C ] = [ ] ( [C ]+ [ G ] ) [ ] , [ K ] = [ ] ( [ K ]+ [ K ] ) [ ] .
red
red
{f },
(5.111)
red
red
After determining the modal coordinates { u }, the physical coordinates are calculated from equation (5.110). One can say that up to 10-12 modal vectors j to accurately determine the first 2-3 eigenvalues whose imaginary parts fall within the operating speed range. It is possible to introduce diagonal modal damping values accounting for the external and internal/structural damping.
{ } have to be used in order
The modal method does not require any intuition on mass lumping, component mode selection, and iterative procedures to improve the transformation matrix. The only basic assumption is that the linear combination of the Ritz basis vectors obtained by considering the undamped isotropic rotor system constitutes a good approximation to the complex eigenvectors of the heavily damped anisotropic rotor system.
5.8.2 Model substructuring
The rotor-stator-foundation system can be subdivided into components or substructures, analyzing the components separately, and then coupling them to obtain the mathematical model of the full system. Methods of component mode synthesis (CMS) or substructure coupling for dynamic analysis are used to synthesize the system equations of motion from the characteristic displacement modes of the components. In the CMS method, the displacement of any point in a component is represented as the superposition of two types of component modes: a) constrained normal modes, defining displacements relative to the fixed component boundaries, and b) constraint modes, produced by displacing the boundary coordinates. In addition, the coordinates of any component are classified as: a) boundary (junction) coordinates, if they are common to two or more components, and b) interior coordinates, if they do not interface with any other component. The static constraint modes are determined by giving each of the boundary coordinates a unit displacement in turn, fixing all other boundary coordinates and allowing the internal coordinates to displace. The dynamic constrained normal modes are found by fixing all boundary coordinates and determining the free vibration modes of the constrained component.
63
The reduction of the number of DOFs of the system is provided by the constrained normal modes. The assumption is that the system response can be found accurately enough by retaining only a properly selected reduced number of constrained normal modes. Each component is transformed in terms of its constrained normal modes and assembled into a lower order set of system equations of motion. The analysis of a full rotor-bearing-foundation system is beyond the aim of this presentation. The subsystem rotor-bearings will be considered separately to emphasize the steps in the CMS analysis of a substructure [5.9]. Consider a uniform shaft (Fig. 5.40) supported by three bearings. Neglecting coupling effects, consider only the vibration in the Y-X plane.
Fig. 5.40
64
The shaft is divided into 7 beam finite elements. The nodal coordinates v1 , v2 ,, v8 , 1 , 2 ,, 8 are listed in the upper part of Fig. 5.40, a. Let the boundary coordinates v1 , v4 and v8 be the a-DOFs, and the interior coordinates be the o-DOFs:
{Ya } = { v1 {Yo } = { 1
v4 v8 }T ,
v2 2 v3 3 4 v5 5 v6 6 v7 7 8 }T .
Order the elements of the global displacement vector so that active DOFs are collected at the top
{Y } {Y } = a . {Yo }
(5.112)
The equation of motion can be written in partitioned-matrix form as [maa ] [m ] oa

&& } [caa ] [0] {Y & } [k aa ] [k ao ] {Ya } [mao ] {Y {Fy a } a a + = {F } . & + && [moo ] {Yo } [0] [0] {Yo } [koa ] [koo ] {Yo } yo
(5.113) Matrices are shown in Fig. 5.41 where the shaded areas represent shaft elements while circles represent bearing elements.
Fig. 5.41
5.8.2.1 Constrained normal modes
If translational displacements in bearings (boundary coordinates) are blocked, then v1 = v4 = v8 = 0 , {Ya } = { 0 } and equation (5.113) reduces to that of a conservative auxiliary system. The homogeneous form
65
&& }+ [ k ] {Y } = { 0 } [ moo ]{ Y o oo o
(5.114)
yields the so-called constrained normal modes, i.e. the eigenvectors
( j = 1,....,13) of the rigidly supported (constrained) shaft (Fig. 5.40, b).
{ j }
By truncation of these component modes, a transformation matrix is set up by using, for example, only 5 (out of 13) vectors in a reduced modal matrix
[ ] = [ { }
{ 2 }
{ 5 } ].
(5.115)
The coordinate transformation is written as
{Yo } = [ ] { q }
131 135 51
(5.116)
where { q } is a column vector of constrained modal coordinates. Equations (5.112) and (5.116) yield
(5.117)
5.8.2.2 Constraint modes
The static behavior of the rigidly supported shaft is described by the static part of equation (5.113), where Fy o = { 0 }
{ }
[k aa ] [k ] oa
[kao ] {Ya } {Fy a } = . [koo ] {Yo } { 0 }
(5.118)
66
The interior coordinates {Yo } can be expressed in terms of the boundary coordinates {Ya } using the lower part of equation (5.118)
[ koa ] {Ya } + [ koo ] {Yo } = { 0 }.

This can also be written as
{Yo } = [ koo ] 1[ koa ] {Ya } = [ stat ] {Ya } .

The constraint modes are the columns of the matrix [I a ] where
(5.119)
[ ]
stat
[ stat ] = [ koo ] 1[ koa ] .
(5.120)
The static constraint modes are defined by producing a unit displacement of each boundary coordinate in turn, with all other coordinates blocked and all interior coordinates unconstrained and unloaded (Fig. 5.40, c). They represent global shape functions or Ritz vectors.
5.8.2.3 Combined static and modal condensation
The constrained normal modes and the constraint modes [ stat ] are superposed to obtain a modal transformation of the component physical coordinates. Equation (5.117) becomes
[ ]
(5.121)
or
{Y } = [ T ]{ u }.
(5.122)
67
Substitution of (5.122) into (5.113) and premultiplication by [ T the reduced set of equations of motion
]T
yields
(5.123)
which can be written as m red aa m red qa where
[ [
[ c ] ] [ m ] {u && } + ] [ m ] [0]
red aq red qq aa
[ k red ] [0] aa {u & }+ [0] [0]
[k ]
red qq
[0]
{u } = { M },
[k [k [c [m [m [m
red aa
red aa
red qq
red aa red qq red qa
red aq
] = [ k ] [ k ][ k ] [ k ] , ] = [ ] [ k ] [ ] , ]= [c ] , ] = [ ] [ m ] [ ] , ] = [ ] ( [ m ] [ ] + [ m ]= [m ] ,
1
aa ao oo oa
oo
aa
oo
(5.124)
oo
stat
oa
]) ,
red T qa
[ m ] = [
stat
]T ( [ moo ] [ stat ] + [ moa ] ) + [ mao ] [ stat ] + [ maa ]
Once the truncated component equations are developed for each component of the system, the next step in the analysis is to assemble them to form the truncated system equations. They are solved in terms of the system coordinate vector, composed of the boundary coordinates of the system and the retained constrained modal coordinates of each component. Subsequent development may proceed in first order form. Concluding, in the Craig-Bampton component synthesis methods, the displacement of any point in a component is represented as a superposition of two types of displacement modes: a) constrained normal modes displacements
68
relative to the fixed component boundaries, and b) static constraint modes displacements produced by displacing the boundaries. The number of degrees of freedom for the whole system is reduced by truncating the number of constrained normal modes. Usually, overdamped or heavily damped modes are truncated first, then the highest frequency modes.
5.8.3 Stepwise model reduction methods

The static or Guyan reduction simply neglects the inertia terms associated with the omitted degrees of freedom (o-DOFs). In the Improved Reduced System (IRS) method, the inertial effects of o-DOFs are taken into account. Its robustness is dependent on the observability of the structure from the selected active degrees of freedom (a-DOFs). The Iterative IRS method (IIRS) is based on a serial elimination of o-DOFs and an automatic selection of the number and location of aDOFs [5.10]. The IIRS method converges to a reduced model which reproduces a subset of the modal model of the full system. In the equation of motion (5.50) of a damped rotor-bearing system, the size of matrices is 4 N , where N is the number of nodes in the finite element model. For a large complex system, this leads to a large size eigenvalue problem, while only the first several eigenproperties are of practical interest. In order to reduce the size of the eigenvalue problem, the conservative non-gyroscopic homogeneous part of equation (5.50) is considered first
[ M ] { &x& }+ [ K ] { x } = { 0 } ,
where [M ] and respectively.
(5.125)
[K ]
are the symmetric parts of matrices
[M ]
and
[K ] ,
The criterion is the value of the ratio k jj m jj of the diagonal elements of [M ] and lower location of the displacement vector for elimination. Equation (5.125) can be written
& }+ [P ] T [ K ] [P ][P ] T { x } = { 0 } , [P ]T [ M ] [P ][P ]T { & x
The physical coordinates in the vector
{ x } are eliminated one at a time.
[K ] . The DOF for which this ratio is highest is denoted xo and moved into the
where [P ] is a permutation matrix. It can be partitioned as
69
[M aa ] M oa
&a } { M ao } { & x
M oo
[K aa ] + & &o K oa x
{ K ao } { xa }
K oo
{ 0 } x = o 0
(5.126)
where xo , K oo , and M oo are scalars, and { xa } is the column vector of the remaining a-DOFs. If several DOFs have the same ratio k jj m jj , then the one with the smallest index is considered first. If this ratio is greater than a cut-off frequency squared 2 c , then the corresponding DOF is eliminated.
5.8.3.1 Stepwise Guyan Reduction (SGR)
From the static part of equation (5.126), a constraint equation between the o-DOF and the column vector of a-DOFs is obtained as
1 ST xo = K oo K oa { xa } = Goa
{ xa } .
{ x } = [P] [TS ]{ xa } .
The reduction to a-DOFs in the SGR method is defined by { xa } [I a ] = ST {xa } = [ TS xo Goa
]{xa } ,
After one reduction step, the SGR reduced equation (5.125) becomes
&a } + [ K ST ] { xa } = { 0 } , [ M ST ] { & x
where
(5.127)
[ M ST ] = [ TS ]T [P]T [ M ] [P][ TS ] , [ K ST ] = [ TS ]T [P ]T [ K ] [P][ TS ] . Each reduction step is characterized by the product [P ][ TS ] . The whole
process is described by a transformation matrix of the form
[ T ] = ( [P][ TS ] )1 ( [P ][ TS ] )2 .........( [P ][ TS ] )n
where n is the number of o-DOFs. At the end of the stepwise elimination of o-DOFs, the full reduced matrices are
[ M red ] = [ T ]T [ M ][ T ] , [ K red ] = [ T ]T [ K ][ T ] , [ Cred ] = [ T ] T [ C ][ T ] ,(5.128) where the same matrix [ T ] has been used to reduce in [ C ] the damping and
gyroscopic matrices.
The physical coordinate reduction is defined by
{ x } = [ T ] { xred }.
(5.129)
70
After solving the reduced eigenvalue problem, the full size modal vectors can be recovered by back-expansion, using the inverse SGR method (ISGR) based on the transformation (5.129).
5.8.3.2 Stepwise Improved Reduction (SIR)
If the acceleration vector from equation (5.127)
&a } = [ M ST ] 1[ K ST ] { xa } , {& x
and the acceleration scalar
1 & &o = K oo x K oa [ M ST
]1[ K ST ] { xa }
are substituted into the lower part of equation (5.126), then a new constraint equation is obtained between the o-DOF and the column vector of a-DOFs
(1) xo = Goa
{ xa }
oa
where
[G ( ) ] = K [ K
1 oa
1 oo
( M oa + M oo Goa
ST
) [ M ST ] 1 [ K ST ] ].
(5.130)
In the SIR method, the reduction to a-DOFs is defined by { xa } [ I a ] = (1) {xa } = [ T1 ]{xa } , xo Goa
{ x } = [P] [T1 ]{ xa }.
After one reduction step, the reduced homogeneous equation of motion is

&a } + [ C1 ] { x & a } + [ K1 ] { xa } = { 0 } , [ M1 ] { & x
where
[ M1 ] = [ T1 ]T [P]T [ M ] [P][ T1 ] ,
[ K1 ] = [ T1 ]T [P]T [ K ] [P][ T1 ] ,
[ C1 ] = [ T1 ]T [P]T [ C ] [P][ T1 ] .
The SIR transformation matrix is of the form
[ T ] = ( [P][ T1 ] )1 ( [P ][ T1 ] )2 .........( [P][ T1 ] )n

and the full reduced system matrices are given by (5.128).
(5.131)
Modal vector back-expansion is carried out using the Inverse SIR (ISIR) method based on the transformation (5.129) where [ T ] is given by (5.131).
71
5.8.3.3 Stepwise Iterated Improved Reduction (SIIR)
&a } and & &o into equation (5.126) is If substitution of accelerations { & x x repeated, for the subsequent iterations the constraint equation becomes
(i +1) xo = Goa
{ xa } ,
where
[G ( ) ] = K [ K
i +1 oa
1 oo
oa
1 (i ) ( M oa + M oo Goa ) [ M i ] [ K i ] .
The reduction to a-DOFs becomes { xa } [ I a ] = (i ) {xa } = [ Ti xo Goa
]{xa } ,
{ x } = [P] [ Ti ]{ xa },
where the subscript i denotes the ith iteration. After one reduction step, the SIIR homogeneous equation of motion of the damped gyroscopic system is
&a } + [ Ci ] { x & a } + [ K i ] { xa } = { 0 } , [ Mi ] {& x

where
[ M i ] = [ Ti ]T [P]T [ M ] [P][ Ti ] ,
[ K i ] = [ Ti ] T [P]T [ K ] [P ][ Ti ] ,
(5.132)
[ Ci ] = [ Ti ]T [P]T [ C ] [P][ Ti ] .
At each reduction step, the effect of the removed o-DOF is redistributed to all the remaining a-DOFs, so that the next reduction will remove the o-DOF with the highest k jj m jj ratio in the reduced mass and stiffness matrices. The procedure is applied until the highest ratio k jj m jj is equal to or less than 2 c . At this point the a-DOFs represent the selected active DOFs of the reduced model. Indeed, for sinusoidal excitation with frequency , the lower part of equation (5.126) yields
1 1 xo = K oo 1 2 K oo M oo
) ( K
1
oa
2 M oa ) { xa }. (5.133)
If the first parenthesis in (5.133) is approximated by the truncated binomial expansion
(1
1 K oo M oo
1 1 + 2 K oo M oo
(5.134)
and the terms in 4 are ignored, then a constraint equation is obtained as
72
1 xo = K oo
[ K
oa
( M
oa
+ M oo Goa
ST
) ] { xa } .
(5.135)
When { xa } is a mode shape of the reduced conservative problem, equation (5.135) becomes equation (5.130). But equation (5.134) is valid only for frequencies 2 << K oo /M oo . This means that if a limit 2 c is established, then the elimination of o-DOFs can be done until the ratio K oo /M oo is equal to or smaller than this value. In this way, the minimum number of a-DOFs is automatically determined, as well as their location. For conservative non-gyroscopic systems, the SIIR method converges monotonically to a reduced model that preserves the lower eigenvalues and the corresponding reduced eigenvectors of the full system. For damped gyroscopic systems, the arbitrary reduction of damping and gyroscopic matrices in (5.132) may lead to better prediction accuracy of the SIR method for some eigenmodes. After solving the reduced eigenvalue problem, equation (5.129) is used in the Inverse SIIR (ISIIR) method to expand the a-DOF vector to the size of the full problem, using the transformation matrix
[ T ] = ( [P][ Ti ] )1 ( [P][ Ti ] ) 2 .........( [P][ Ti ] )n ,

where the subscript i is the number of iterations in the SIIR method. A measure of the accuracy of the expanded mode shapes is given by the relative mode shape error abs ( FEM ) abs exp anded
Example 5.1
FEM 100 (% ) .
Figure 5.42 shows the single-disk multi-stepped shaft rotor system with two identical isotropic bearings of reference [5.8]. It was modeled with 76 DOFs, using 18 Timoshenko shaft elements including gyroscopic effects and neglecting internal damping [5.11]. The reduction procedure, applied with a cut-off frequency of 8000 rad/sec, selected 8 active DOFs: 3, 9, 25, 33, 41, 47, 63, 71. The damped eigenfrequencies, computed at a spin speed n = 30000 rpm , are given in Table 5.3. The "true" values listed in the second column correspond to the full eigenvalue problem (76 DOFs). Columns three to five list natural frequencies computed using SGR, SIR and SIIR.. Note that the 8 translation DOFs are located at 4 nodes, the a-DOFs selected in one plane being selected in the other plane too.
73
Fig. 5.42
Columns six to eight list the relative error of expanded mode shapes by Inverse Stepwise Reduction. For some modes, the iterations in SIIR (and ISIIR) do not improve on the values in SIR (and ISIR).
Table. 5.3 Damped eigenfrequency, Hz Mode
1 2 3 4 5 6 7 8
Mode shape error, % SIIR 5 iter

246.21 296.74 772.79 807.20 1166.8 1369.7 1958.7 2020.7
True FEM
246.01 296.49 774.00 808.31 1165.7 1367.4 1959.9 2020.6
SGR
246.78 296.91 777.26 808.38 1287.3 1425.4 1997.6 2054.1
SIR
246.27 296.81 773.00 807.32 1174.9 1366.1 1960.0 2020.0
ISGR
2.04 1.54 3.59 2,16 24.96 14.71 3.59 3.29
ISIR
1.46 1.82 1.07 0.91 7.57 0.69 1.26 0.61
ISIIR 5 iter
1.32 1.66 1.23 1.12 3.44 3.35 0.57 0.53
Example 5.2
Figure 5.43 shows the four-disk three-bearing rotor from reference [5.12] with variable inner diameter. The rotor was modeled with 13 nodes (52 DOFs), using Timoshenko shaft elements with consistent mass and gyroscopic matrices.
74
Fig. 5.43
The reduction procedures applied with a cut-off frequency of 3000 rad sec selected 24 a-DOFs. Further reduction applied to select only 8 active DOFs, using the K M criterion, selected the four DOFs of node 1, and the translational DOFs of nodes 10 and 12 [5.11].
1 2 3 4 5 6 7 8
Mode shape error, % ISGR

0.28 0.50 4.71 4.83 18.5 34.2 43.5 46.7
True FEM
46.391 59.573 178.44 179.41 380.29 441.37 461.39 463.96
SGR
46.419 59.611 179.77 180.76 399.07 469.27 507.28 508.54
SIR
46.395 59.580 178.42 179.39 380.75 443.79 463.35 465.16
SIIR
46.394 59.578 178.40 179.38 380.40 441.66 459.26 461.70
ISIR
0.10 0.13 0.17 0.19 3.33 9.59 13.8 14.1
ISIIR
0.07 0.09 0.10 0.07 1.31 3.11 1.21 1.70
Table 5.4 lists the damped eigenfrequencies computed at a spin speed of 3000 rpm for the first eight modes of precession. Again, the accuracy is very good for the SIIR method with only 5 iterations. Mode shape expansion by the ISIIR method gives excellent results.
Example 5.3
The reduction procedures have been applied to the rotating shaft of a vertical Kaplan hydraulic unit (Fig. 5.44) from [5.13]. The shaft was modeled with
75
14 Timoshenko elements (60 DOFs) and simplified equivalent constant properties for the three horizontal bearings [5.11].
Fig. 5.44
Using a cut-off frequency of 500 rad/sec, the selected a-DOFs were: 1, 9, 21, 29, 31, 39, 51, 59, six of them for the three large masses: turbine, generator and auxiliary rotor. The eight lowest eigenfrequencies, computed for an angular speed of 3000 rpm, are listed in Table 5.5 together with the relative error of the expanded mode shapes.
1 2 3 4 5 6 7 8
Mode shape error, % SIIR

25.371 28.796 38.084 38.379 45.045 47.226 65.705 68.198
True FEM
25.337 28.762 38.092 38.386 45.026 47.195 65.724 68.203
SGR
25.357 28.884 38.681 38.946 45.779 48.015 66.776 69.263
SIR
25.368 28.799 38.084 38.389 45.040 47.245 65.774 68.203
ISGR
1.21 2.91 9.54 10.3 7.94 8.49 7.43 6.88
ISIR
1.49 1.61 0.38 1.50 1.37 2.59 2.38 1.20
ISIIR
1.61 1.58 0.71 0.75 1.68 2.02 0.90 0.95
76
References
5.1 Ehrich, F. F. (ed.), Handbook of Rotordynamics, McGraw Hill, New York, 1992. 5.2 Lalanne, M., Ferraris, G., Tran, D. M., Quau, J. P., and Berthier, P., Comportement dynamique des rotors de turbomachines, I.N.S.A. Lyon, 1982.
5.3 Marguerre, K. and Wlfel, H., Mechanics of Vibration, Sijthoff & Noordhoff, Alphen aan den Rijn, 1979. 5.4 Genta, G. and Gugliotta, A., A conical element for finite element rotor dynamics, J. Sound Vib., vol.120, no. 1, p. 175-182, 1988.
5.5 Lee, C.-W., Vibration Analysis of Rotors, Kluwer Academic Publ., Dordrecht, 1993. 5.6 Someya, T. (ed), Journal-Bearing Databook, Springer, Berlin, 1988. 5.7 Childs, D., Turbomachinery Rotordynamics: Phenomena, Modeling and Analysis, Wiley, 1993.
5.8 Wang, W., and Kirkhope, J., New eigensolutions and modal analysis for gyroscopic/rotor systems, Part I: Undamped sytems, J. Sound Vib., vol.175, no.2, pp 159-170, 1994. 5.9 Gasch, R., and Knothe, K., Strukturdynamik, Band 2, Kontinua und ihre Diskretisierung, Springer, Berlin, 1989. 5.10 Shah, V. N., and Raymund, M., Analytical Selection of Masters for the Reduced Eigenvalue Problem, Int. J. Num. Methods in Engineering, vol.18, pp 89-98, 1982. 5.11 Rade, M., Rotor-bearing model order reduction, Proc. 5th Int. IFToMM Conference on Rotor Dynamics, Vieweg, pp 148-159, 1998. 5.12 Rajan, M., Rajan, S. D., Nelson, H. D., and Chen, W. J., Optimal Placement of Critical Speeds in Rotor-Bearing Systems, ASME Journal of Vibration, Acoustics, Stress, and Reliability in Design, vol.109, pp 152-157, 1987. 5.13 Gmr, T. C., and Rodrigues, J. D., Shaft Finite Elements for Rotor Dynamics Analysis, ASME Journal of Vibration and Acoustics, vol.113, pp 482-493, 1991.
6.
FLUID FILM BEARINGS AND SEALS
This chapter is a brief introduction to hydrodynamic fluid film journal bearings and seals for rotating machinery. The presentation describes how the fluid film bearings and seals work and some of the basic principles that underline their operation. The focus is on their dynamic characteristics and the influence of different bearing and seal configurations on the dynamics of rotor-bearing systems.
6.1 Fluid film bearings

Fluid film bearings operate with a thin film of lubricant between two surfaces which are in relative motion. The lubricant may serve some or all of the following purposes: a) support the rotor load; b) reduce support friction; c) provide radial stiffness and radial squeeze-film damping; d) cool the bearing; and e) attenuate transmitted rotor vibrations. Fluid film bearings may be classified into several main types according to: a) principle: hydrodynamic (self-acting), hydrostatic (externally pressurized), or hybrid (combination of the two); b) fluid: liquid or gas; c) flow regime: laminar or turbulent; d) geometry: plain cylindrical, four axial groove, elliptical (lemon), three-lobe, pressure dam, Rayleigh step, tilting pad, floating ring, others; and e) load relative direction: radial or axial. In the following, attention will be restricted to hydrodynamic, incompressible, laminar, oil-lubricated journal bearings. In this case, the load carrying capacity derives from the hydrodynamic pressures generated in the lubricant film by the rotation of the journal. Their operation is based on the formation of an oil wedge which lifts the journal. Configurations giving rise to more wedges ensure better stability. Tilting pad designs are intrinsically stable. Solutions with Rayleigh steps have high load carrying capacity.
78
6.2. Static characteristics of journal bearings

Consider a horizontal bearing with a journal rotating at a constant angular speed . Assume that the journal is loaded vertically and downwards by an external force of constant magnitude W (Fig. 6.1).
Fig. 6.1
The center O J of the loaded journal shifts from the center O B of the bearing. A convergent clearance space is formed between the journal and the bearing surface. Due to adhesion and viscosity, the lubricant is forced into the wedge-like clearance so that a hydrodynamic pressure p is generated in the fluid film. The load carrying capacity of the bearing is generated by the pressure forces that compensate the friction forces produced in the viscous fluid by the relative motion of the journal and bearing surfaces. In the remaining divergent clearance space, the lubricant flow is usually cavitated by aeration (expulsion of air or gas) and the fluid film is ruptured. The wedge film pressure p has an asymmetrical distribution with respect to the O B O J line. The resultant force F of the pressure p is balanced by the static external load W. The equilibrium position of the journal center is dependent on the load W and the spin speed .
6. FLUID FILM BEARINGS
79
6.2.1 Geometry of a plain cylindrical bearing

Figure 6.2 illustrates the cross section of a "plain" journal bearing [6.1]. The bearing main dimensions are: DB = 2 RB - inner bearing diameter, D = 2 R outer journal diameter, L - bearing length, C = RB R - bearing clearance,
e = O B O J - eccentricity, - attitude angle and h - fluid film thickness.
Fig. 6.2 It is useful to introduce the following dimensionless geometrical parameters: L/D - length-to-diameter ratio; C/R - clearance ratio; = e C eccentricity ratio; and h = h C - normalized film thickness. The journal is assumed to rotate anticlockwise with a constant angular speed = 2 N (rad/sec), where N represents the journal spin speed in rps. In the triangle O B O J M , O B M = R + C , O J M = R + h , so that
(R + h )2 = e 2 + (R + C )2 + 2e (R + C ) cos .
Expanding and discarding higher-order terms in e, h R , and C R , the film thickness can be approximated by h C ( 1 + cos ) , which describes the film shape to within 1% accuracy. The minimum film thickness is (6.1)
hmin = C ( 1 ) .
(6.2)
80
6.2.2 Equilibrium position of journal center in bearing
The load carrying capacity of a bearing is conveniently characterized by a dimensionless duty parameter called the Sommerfeld number. The mean pressure on the projected area of the journal is
p = W DL p C W C , = N R N DL R
2 2
or, in dimensionless form,

pm =
where W is the radial load, N, and is the lubricant absolute viscosity, Nsec m 2 . The Sommerfeld number is commonly defined as
S=
1 N DL R = . pm W C
(6.3)
A smaller value of S indicates a heavily loaded bearing with a lower rotational speed. In some publications the inverse of S is referred to as the Sommerfeld number.
a
Fig. 6.3 (from [6.2])
81
The locus of the static equilibrium position for the journal center is not a straight line, but a curve resembling a semi-circle (Fig. 6.3, a). Figure 6.3, b shows the static equilibrium loci for five values of L D . The locus is defined by the eccentricity ratio 0 and the attitude angle 0 . Both parameters are functions of load and speed through the parameter S.
Fig. 6.4 (from [6.3])
For W=const. and increasing , the journal center moves upwards the equilibrium locus. Beginning at the bottom of the bearing ( 0 = 1) , when the rotor starts up, and rising as the speed increases, the journal reaches the concentric position ( 0 = 0 ) in the limit, as the spin speed becomes infinitely large. For = const . and increasing W, the journal center moves downwards the equilibrium locus. Beginning in the concentric position ( 0 = 0 ) , where the
82
bearing is unloaded, the journal center moves downwards as the load is increased, approaching the bottom of the bearing ( 0 = 1) as the load goes to infinity.
Fig. 6.5 (from [6.2])
Fig. 6.6 (from [6.2])
83
Figure 6.4 gives the eccentricity ratio 0 of a plain cylindrical bearing as a function of the Sommerfeld number, for three values of L/D. Figure 6.5 gives the attitude angle 0 as a function of S, while Fig. 6.6 shoves 1 0 vs. S. For known bearing geometry, lubricant viscosity and static load, the equilibrium position, given by 0 = 0 (S ) and 0 = 0 (S ) , is a function of the speed only. Note that the force and the displacement are not in the same direction. Although for horizontal shafts the load W is vertically down, the journal center shifts in an oblique equilibrium position. The journal bearing behaves as an asymmetric element, with cross-coupled as well as direct stiffnesses. The bearing is anisotropic, having different stiffnesses in different radial directions.
6.3 Dynamic coefficients of journal bearings

Consider that the journal moves in the bearing along an orbit around its steady-state equilibrium position (Fig. 6.7). The resultant reaction force of lubricant film has a variable magnitude and direction (it is no more vertical). Its components Fy and Fz are non-linear functions of the journal center
& and z &: displacements y and z, and its velocity components y &,z &) , Fy = Fy ( y , z , y
&,z &) . Fz = Fz ( y , z , y
For a small amplitude motion, the bearing reaction may be expressed by the first order Taylor series expansion of its components around the static equilibrium position (note the direction of the force components here, selected to avoid the minus sign in the following expressions):
& + c yz z &, Fy = Fy 0 + k yy y + k yz z + c yy y & + c zz z &. Fz = Fz 0 + k zy y + k zz z + c zy y
(6.4)
The static reaction force components are Fy 0 = W and Fz 0 = 0 . The eight coefficients of the linearized force components are computed as the gradients &=z & = 0) : in the static equilibrium position ( y = z = y
Fy k yy = y Fy c yy = y &
Fy , k yz = z 0 Fy , c yz = z 0 &
Fz y , k zy = 0 Fz y , c zy = & 0
Fz z , k zz = , 0 0 Fz z , c zz = & 0 . 0
They result from the solution of the lubrication equation.
84
The linearization of the bearing reaction forces has the advantage of decoupling the rotor and the bearings. Otherwise, the rotor equations must be integrated simultaneously with the lubrication equation.
Fig. 6.7
In matrix form, equation (6.4) can be written

Fy W Fy . = + Fz 0 Fz The increments of the film force due to small movements around the position of static equilibrium are expressed in terms of stiffness and damping coefficients Fy k yy = Fz k zy k yz y c yy + k zz z c zy c yz y & & b y b y = k + c . c zz z & & z z
[ ]
[ ]
(6.5)
The eight linearized stiffness and damping coefficients depend on the journal steady-state operating conditions, hence upon the rotational speed. The dimensionless stiffness coefficients are defined as
[ K ] = K K
b
yy zy
K yz C = kb , K zz W C yz C b = c . C zz W
[ ]
(6.6, a)
while the dimensionless damping coefficients are defined as
[C ] = C C
b
yy zy
[ ]
(6.6, b)
For a given set of geometrical parameters and lubricant viscosity, these eight coefficients are functions of the Sommerfeld number S or the eccentricity
85
ratio. They are referred to as the eight dynamic bearing coefficients. Values of these coefficients are given in the book edited by Someya [6.1]. For flexible shafts or slightly tilted rigid journals, the journal axis may not be parallel to the bearing axis. The pressure distribution along the journal length gives rise to reaction moments. The corresponding Taylor series expansion defines four moment stiffness coefficients and four moment damping coefficients. Generally, compared to the radial coefficients, they are smaller by a factor of (2L/ l ) where l is the span adjacent to the bearing and L is the bearing length. Only for long bearings or for higher order shaft modes, the influence of moment coefficients may become significant. The stiffness matrix of journal bearings is non-symmetric. This is the cause of rotor instability above a limiting running speed called the onset speed of instability. The unstable motion of journal bearings is called oil whirl and involves large-amplitude subsynchronous motion at the rotor critical speed. Vertical rotors without side loads may experience a motion consisting of a limit cycle whose frequency tracks at approximately one-half running speed which is also called oil whirl. At speeds above twice the rotor's natural frequency, the rotor subsynchronous motion stops tracking running speed and precesses at the natural frequency, motion that is called oil whip. These unstable motions are presented in more detail in Chapter 7. Cavitation, compressibility, and turbulence are phenomena which complicate the modelling and analysis of hydrodynamic bearings. Their account is beyond the aim of this presentation.
6.4 Reynolds equation and its boundary conditions

In the following, certain facts relating to rotor-bearing interaction will be presented, in order to understand the methodology and simplifying assumptions used in the calculation of bearing dynamic characteristics. First, the Reynolds equation for laminar flow is derived. It is the governing equation for the pressure field within a bearing with isoviscous flow. Second, by integrating this pressure field, the non-linear bearing forces acting upon the journal are determined. The boundary conditions for the pressure field distribution are shortly described. Third, the non-linear dynamic fluid film reaction force is expanded in a Taylor series and the terms of second and higher powers are disregarded. The stiffness and damping coefficients (6.5) are determined from the first order terms of this series [6.4]. Note that u, v , w are fluid velocity components in the x, y, z directions, and not displacements.
86 6.4.1 General assumptions
The following 13 assumptions are used to derive Reynolds equation from the general Navier-Stokes equations: 1. The lubricant is isotropic. 2. The lubricant is Newtonian. It obeys Newton's viscosity law of friction
= & ,
1/sec, and is the lubricant dynamic (absolute) viscosity, Nsec m 2 . Sometimes Nsec m 2 ).
(6.7)
where is the shear stress in the fluid film, N m 2 , & is the shear strain rate, is measured in centipoises (1cP = 1.0054 10 3
3. The dynamic viscosity does not vary across the film thickness. 4. The fluid film is very thin compared to its length and width:
hmin << R ,
(C R )2 << 1 .
R 0.001 and Re < 2300 )
5. The flow is laminar and free of inertia:
(C R ) Re << 1 , (C
where the Reynolds number
Re =
wt C
or
Re =
wt hmin
(6.8)
is a measure of the ratio between the inertia forces and the viscous friction forces acting on a lubricant volume, is the lubricant mass density, kg m3 , wt = R is the journal surface velocity, m/sec, and C is the radial clearance, m. 6. There is no slip between lubricant and the journal/bearing surfaces. 7. The curvature of journal and bearing surfaces is neglected. 8. The deformation of journal and bearing is neglected. 9. The variation of bearing clearance in the circumferential direction is small. 10. The relative inclination of journal and bearing axes (hence the restoring moment) is neglected. 11. The fluid velocity component across the thickness of the film is small in comparison to the circumferential and axial components.
87
12. The pressure is constant across the film thickness. 13. First and second order velocity gradients in circumferential and axial directions are negligible compared to those across the thickness. Supplementary assumptions: 14. Incompressible lubricant ( = const .) . 15. Isothermal and isoviscous flow. In order to determine the approximate bearing temperature, a simplified version of the energy equation is used. Because the predominant flow component is in the circumferential direction, usual assumptions include: a) neglecting the small contributions from the pressure gradients to both flow and friction, and b) ignorance of any heat conduction along the film or to the surfaces.
6.4.2 Reynolds equation
Reynolds equation is a greatly simplified version of the Navier-Stokes equations and the equation of continuity.
Fig. 6.8
Neglecting body forces and (both temporal and convective) inertia forces (see assumptions 4 and 5), the equilibrium of pressure forces and friction viscous forces acting on an elementary fluid volume (Fig. 6.8) yields
p zy = z y
so that
where zy =
w , y
p w = , z y y
p u = , x y y
p = 0, y
According to assumption 3
88
2 u 1 p , = y2 x
2 w 1 p , = y2 z
(6.9)
where the pressure gradients p x and p z are not functions of y. For non-slip boundary conditions
y=0, w=0,
v = 0,
u = 0 , (stationary bearing) u = 0 , (assumption 9)
y = h , w = wh , v = h t ,
the velocity field is
w=
y 1 p y ( y h ) + wh , 2 z h
u=
1 p y ( y h) . 2 x
The volume flow per unit bearing width, in the direction of motion, is
qz =
w dy = wh
h h3 p , 2 12 z
(6.10)
where the first term in the right hand side is the shear flow (Couette flow) and the second term is the pressure gradient flow (Poiseuille flow). The volume flow per unit bearing width along the journal length is
qx =
u dy =
h3 p . 12 x
(6.11)
For an infinitesimal film volume of height h and area dx dz , the equation of continuity is
( q z ) + ( q x ) + ( h ) = 0 . z x t
(6.12)
Substitution of q z and q x from (6.10) and (6.11) yields

3 3 h p + h p = 1 (wh h ) + ( h ) . t z 12 z x 12 x 2 z
For an incompressible fluid ( = const . ) the laminar-flow Reynolds equation in Cartesian coordinates (with time-dependent effects) becomes

3 3 h p + h p = wh h + h . z 12 z x 12 x 2 z t
89
(6.13) equation in
Replacing z = R and circumferential coordinates becomes
wh = R ,
the
Reynolds
3 3 1 h p + h p = h + h , 2 12 x 12 x R 2 t
(6.14)
where h = C ( 1 + cos
) and
is the angular coordinate defined in Fig. 6.1.
6.4.3 Boundary conditions for the fluid film pressure field
The circumferential boundary conditions commonly used for the plain cylindrical bearings in the case of steady state operation are shown in Fig. 6.9.
b
Fig. 6.9
a) The Full-Sommerfeld boundary condition ( 2 - film bearing) is antisymmetrical with respect to ' = (Fig. 6.9, a):
p ( ' , x ) > 0 for 0 < ' < ,

p ( ' , x ) = p (2 ' , x ) .
A negative pressure of the same order as the positive one is generated for angles < ' < 2 , though, because of the film rupture, the pressure is equal to the atmospheric one in this region. Large negative pressures do not exist in the diverging region and the predicted load capacity is zero what is physically unrealistic. This condition is seldom applied, except when higher pressure is supplied. b) The Gmbel (Half-Sommerfeld) boundary condition ( -film bearing) replaces the negative pressure in the Sommerfeld boundary condition by an atmospheric pressure (Fig. 6.9, b):
90
p ( ' , x ) > 0 for 0 < ' < ,

p ( ' , x ) = 0 for ' 2 .
Although it is physically inappropriate, because the flow continuity is not satisfied at ' = , it is often used because of its simplicity. The errors introduced are small. c) The Reynolds boundary condition (Fig. 6.9, c) imposes a zero pressure gradient at the point where the fluid film pressure equals the ambient pressure:
p ( ' , x ) > 0 for p ( ' , x ) = 0 for p =0 '
0 <' < *,
* ' 2 , ' = * .
at
6.5 Analytical solutions of Reynolds equation

Methods of solving Reynolds' equation include the finite difference method, the finite element method and the infinite series function method. In the following, some analytical solutions are presented.
6.5.1 Short bearing (Ocvirk) solution

In a very short cylindrical bearing, the Poiseuille flow component (xcomponent) in the axial direction becomes much larger than the circumferential one ( -component). If the pressure-gradient contribution to the circumferential velocity w is neglected, the first term in Reynolds' equation (6.14) can be ignored:
h3 2 p h h . + = 12 x 2 2 t
(6.15)
For = const . the equation (6.15) is easily integrated

1 L2 p = 3 x2 3 h 4 so that, along the x axis (Fig. 6.10), h h + 2 t (6.16)
p ( , L 2) = p ( , L 2 ) = 0 .
91
The fluid film forces are calculated by integrating the pressure over the film domain. It is convenient to calculate the force components in the radial and tangential directions, F and F , respectively (Fig. 6.11).
Fig. 6.10
Fig. 6.11
The force components are:
F =
L p cos' R d' dx ,
2
' 1
L 2
' 2
F =
L p sin' R d' dx ,
2
' 1
L ' 2 2
(6.17)
where ' = is measured from the O B O J line. Substitution of the normalized film thickness
h= h = 1 + cos ' C
2
into the pressure equation (6.16) yields
R p = 3 C where
L x D R
&=
1 & ) sin ' 2 & cos ' ] 3 [ ( 1 2 h

&=
(6.18)
1 d , dt
1 d , dt
are the derivatives of and with respect to the dimensionless time t . The pressure is positive when
& ) sin ' 2 & cos ' 0 . ( 1 2
92
The film begins at ' = '1 and ends at ' = ' 2 where p = 0 :
= tan 1 1
& 2 , &) ( 1 2
= 1 + . 2
, the slope of the pressure profile must be positive: At ' = 1

p >0, = 1
& ) cos1 & sin1 + 2 = ( 1 2
&) ( 1 2 > 0, cos1
so that
= cos1
& )2 + (2 & )2 2 ( 1 2
&) ( 1 2
= sin1
& )2 + (2 & )2 2 ( 1 2
& 2
. which define the quadrant for 1 By substituting the pressure (6.18) into (6.17), the force components become
L F = W 2 S D
L F = W 2 S D
2
' 1
' 2
1 h3
& ) sin ' 2 & cos ' ] cos ' d ' , [ ( 1 2
(6.19, a)
' 1
' 2
1 h3
& ) sin ' 2 & cos ' ] sin ' d ' . [ ( 1 2
(6.19, b)
& = & = 0 , the equilibrium position of the Under static loading conditions = 0 and 2 = . Using journal center is given by the coordinates 0 and 0 , 1 this position, a coordinate system (Fig. 6.12) can be defined with the r-axis in the radial direction ( 0 -direction) and the t-axis in the tangential direction ( 0 direction). The components of the pressure resultant along these axes are Fr cos ( - 0 ) sin ( - 0 ) F = . Ft sin ( - 0 ) cos ( - 0 ) F (6.20)
For motions with small amplitudes, e = C in the r-direction and e0 = C 0 in the t-direction. A first order Taylor series expansion around the static equilibrium position yields
93
Fr Fr = Fr0 + Ft Ft = Ft 0 +
Fr Fr + 0 + 0 0 & Ft Ft + 0 + 0 0 &
Fr &+ & , (6.21, a) 0 & 0 0 Ft &+ & , (6.21, b) 0 & 0 0
In evaluating the derivatives, equations (6.19) must be differentiated with & and & , using Leibniz's rule for the respect to the three variables , differentiation of integrals.
Fig. 6.12
& = & = 0 in the static = , corresponding to = 0 and 2 With 1 equilibrium position, equations (6.19) may be integrated and, for = 0 and = 0 , the film forces are Fr0 = (F )0
2 4 0 L = W S 2 D 1 0 2
(6.22, a)
Ft 0 = F
L ( )0 = W D
(1 )
2 32 0
(6.22, b)
From Fig. 6.12 it is seen that Fy cos 0 = Fz sin 0 sin 0 Fr . cos 0 Ft (6.23)
At the static equilibrium, Fz = 0 and Fy = W ,
94
tan0 =
Fr0 W which implies that Fr0 W Ft 0 W
Ft 0 Fr0
2 1 0 , 4 0
cos 0 +
Ft 0 W
sin 0 = 1 ,
= cos0 =
(1 ) (1 )
0
2 4 0
2 2 0
(6.24, a)
= sin0 =
2 32 0
(6.24, b)
Eliminating 0 between equations (6.24), the modified Sommerfeld number becomes
L = S = D 0
(1 ) 16 + (1 )
2 2 0 2 2 0 2 0
(6.25)
which defines the relationship between 0 and S. Equations (6.21) may be written in matrix form as Fr W Fr0 W K rr = + Ft W Ft 0 W K tr
K rt Crr + K tt 0 Ctr
& Crt . (6.26) & Ctt 0
The dimensionless stiffness and damping coefficients in equation (6.26) are related to the actual coefficients by equations similar to (6.6). They are obtained from a differentiation of equations (6.20) and with substitution from equations (6.24):
Kr r =
1 W 1 W 1 W
Fr
1 F =W 0
2 2 8 0 1 + 0 2 1+ 0 = = 2 3 2 0 1 0 0 1 0
( (
) )
Fr 0 W
Krt = Kt r =
Fr 1 Ft 0 = , 0 0 W Ft 1 F = W 0
2 2 1 + 2 0 1 + 2 0 = = 2 52 2 0 1 0 0 1 0
Ft 0 W
95
Kt t = Cr r = Cr t =
1 W
Ft 1 F 1 Fr 0 =W = W , 0 0 0 1 =W 0 F & = 2 Kt r , 0
1 W 1 W
Fr &
Fr 1 =W & 0 Ft &
F 8 0 = 2 & 0 1 0
2 Fr 0 = 2 K t t , 0 W 2 Fr 0 = Cr t , 0 W 2 Ft 0 = 2K r t . 0 W
Ct r =
1 W 1 W
1 F = W 0 &
8 0 = 2 0 1 0
) )
Ct t =
Ft 1 F 2 = = 2 & 0 & W 1 0 0
32
As the r-t coordinate system changes orientation with the equilibrium position, and therefore, with the rotor speed, it is convenient to make a transformation into the fixed y-z coordinate system. cos 0 = 0 sin 0 sin 0 y . cos 0 z (6.27)
The transformation of forces is shown in equation (6.23) so that the y-z dimensionless coefficients are obtained from the transformation K yy K zy
K yz cos 0 = K zz sin 0
sin 0 K rr cos 0 K tr
K rt cos 0 K tt sin 0
sin 0 . cos 0
The analytical form of the eight dynamic coefficients in y-z coordinates is
K yy = K yz =
C k yy W C k yz W C k zy W
4 Q ( 0 )
2 1 0
[ + ( 32 + ) + ( 32 2 ) ],
2 2 2 0 2 4 0
Q ( 0 ) 0
2 1 0
[ + ( 32 + ) + ( 32 2 ) ] ,
2 2 2 0 2 4 0
K zy =
Q( 0 ) 0
2 1 0
2 4 2 2 0 16 2 0 ,
) ]
96
K zz =
C k zz 2 = 4 Q ( 0 ) 2 2 + 16 2 0 , W
) ]
(6.28)
2 0 4 + 2 0 ,
C yy = C yz =
C zz =
C c yy W C c yz W
2 Q ( 0 )
2 1 0
[ + ( 48 2 )
2 2
2 = 8 Q ( 0 ) 2 + 2 2 16 0 = C zy ,
2 1 0 Q ( 0 )
) ]
C c zz 2 = W
[ + ( 2
2
2 16 0 ,
) ]
where
2 2 Q ( 0 ) = 16 0 + 2 1 0
)]
3 2
Figure 6.13 shows an alternate form of the dimensionless stiffness and damping coefficients
Ki j = Ki j 2 , Ci j = Ci j 2 , i, j = y, z (6.29)
as a function of 0 for -film (Gmbel) short bearings.
a
Fig. 6.13 (from [6.5])
The short bearing solution proves to be a valid approximation for plain journal bearings with an L/D ratio less than 0.5 and < 0.7 . As this condition is satisfied in many practical applications, the Ocvirk solution is frequently employed in the analysis of rotor-bearing systems.
97
a
Fig. 6.14 (from [6.5])
More often, the eight dynamic coefficients given by equations (6.28) are plotted versus the Sommerfeld number S for specified values of the L/D ratio. Figure 6.14 shows the plots for L/D = 0.25. For any specific application, it is more useful to plot the physical stiffness and damping coefficients kij , cij against the rotational speed .
Example 6.1
Plots of the physical dynamic bearing coefficients are shown in Fig. 6.15 for a bearing with L = 20 mm, D = 80 mm, C = 0.05 mm, = 0.7 Ps, and W = 417.5 N.
Fig. 6.15 (from [6.5])
98
The static characteristics are given in Table 6.1 [6.5].

Table 6.1 Spin frequency, Hz
1 5 10 20 30 40 50 60 70 80 90
S
0.1085 0.5423 1.0846 2.1692 3.2538 4.3384 5.4230 6.5076 7.5923 8.6769 9.7615
0
0.9512 0.8805 0.8292 0.7593 0.7079 0.6661 0.6305 0.5992 0.5714 0.5462 0.5232
0
0.2750 0.4178 0.5019 0.6049 0.6758 0.7316 0.7781 0.8184 0.8540 0.8859 0.9150
It is interesting to study the asymptotic behaviour of the bearing coefficients as the eccentricity tends to zero. From equation (6.25) one can obtain the asymptotic relation between 0 and S given by
0 S
1 D . 2 L
Then it can be shown that, as 0 approaches zero (or equivalently S tends to infinity),
K zz 2 K yy C yz C zy 8
,
2
L 2 K yz 2 K zy C yy C zz 2 S , D
k zz 2k yy 8W , C
3
k yz k zy , C
D L
8
, C
(6.30)
c yy c zz
D L
4
c yz c zy
8W . C
As 0 0 ( S or W 0 ), the direct stiffness terms, k yy and k zz , become negligible compared with the cross-coupled stiffness terms, k yz and
99
k zy , the direct damping terms, c yy and c zz , tend to a non-zero limit, while the cross-coupled damping terms, c yz and c zy , tend to zero. It is known that a journal bearing must have a radial restoring force in order to have a finite stability threshold. If the journal is operating at zero eccentricity, or if the fluid film is not allowed to cavitate, then the principal stiffness terms vanish, the journal is inherently unstable and the system will exhibit a half frequency whirl. It is also known that the cross-coupled stiffness terms, k yz and k zy , are the major sources of instability. In order to generate instability, the cross-coupled term k zy must be negative. The largest degree of instability occurs when k yz = k zy . When k yz becomes negative, the system stability rapidly improves.
6.5.2 Infinitely long bearing (Sommerfeld) solution

The Sommerfeld solution to the Reynolds equation is obtained by considering p x = 0 , hence dropping the second term on the left in equation (6.14), and integrating with respect to to obtain the pressure field. A short presentation is given in Section 7.3.4.1. The Sommerfeld boundary condition leads to an erroneous result (journal displacement always perpendicular to the direction of the static load W) so that the Gmbel boundary condition is usually used.
6.5.3 Finite-length cavitated bearing (Moes) solution

The previous asymptotic solutions have proven to be useful in the dynamic analysis of bearings: a) the Ocvirk (short) bearing solution - for small eccentricity ratios and very small L/D values, and b) The Sommerfeld (long) bearing solution - for large eccentricity ratios and large L/D values. Using a weighted sum of the asymptotic solutions mentioned above, Moes and Childs [6.6] obtained an accurate finite-length analytic solution which is valid for general finite-length bearings at both large and small eccentricity ratios. The results for a cavitated - bearing are given below. A dimensionless force F0 is used F0 = F0 C LC R
3
(6.31)
100
where F0 = W is the static reaction load. A bearing impedance vector is also defined, whose magnitude at the static equilibrium position is Z0 = F0
In terms of the Sommerfeld number 1 DL R S= . = 0 Z0 2 F0 C

2
(6.32)
Figure 6.16 illustrates S = S ( 0 ) for some L/D values. For a given applied load and bearing, it can be used to determine 0 from the value of S calculated from (6.32). The equilibrium attitude angle 0 is given by
0 = tan 1
where
4a
2 1 0 3b 0
a = 1 + 2.12 B ,
2 L B = 1 0 D
b = 1 + 3.60 B ,
2
The equilibrium magnitude of the impedance vector is Z0 = where d = 1 0 ,

G0 = 3 0 1 + 3.6 Q0 , 4 ( 1 0 )
1 0.15
2 E0 2 32 + G0 d
E0 = 1 + 2.12 Q0 ,
L Q0 = ( 1 0 ) D
2
0 = 0 cos0 ,
0 = 0 sin0 .
101
Fig. 6.16 (from [6.3])
The physical stiffness and damping coefficients are given by

ki j = F0 Ki j , C ci j = F0 Ci j , C i, j = y, z
(6.33)
where the dimensionless stiffness coefficients are K zz = 1
cos 0
sin 0 , (6.34)
1 K zy = sin 0 + cos 0 , 0
K yz = 1
0
1
sin 0 +
Z 1 sin 0 , Z 0
K yy =
cos 0 +
Z 1 cos 0 , Z 0
and the dimensionless damping coefficients are
102
C zz = C zy =
2 sin 0 , 0 2 cos0 , 0
2 Z 1 cos 0 + sin 0 , 0 Z 0
(6.35)
C yz =
C yy = where
Z 1 2 sin 0 cos 0 , Z 0 0
Z Z , = Z Z Z , + = Z Z Z , + = 1
2 1 0
4 2 ( b a ) a 1 = cos 20 2 2 3 b 0 b
= 1 + 0 + sin 1 0 0 2
Z = Z0 1 2 2 E0 + G0
1
2 1 0
, ,
2 3G L 3 2 . 12 E 0 0 0 2 4 D d 2d
G2 1 Z , = Z0 2 0 2 E0 + G0 0
sin 0 , = cos 0 0
= 0 sin 0 ,
= 0 cos 0 .
cos 0 , = sin 0 + 0 Note that C yz C zy .
The physical coefficients of the fluid film bearing from Example 6.1 are listed in Table 6.2.
103
Table 6.2
Freq.
[Hz]
10 20 30 40 50 60 70 80 90
k yy
k yz
k zy
k zz
c yy
c yz
c zy
c zz
[N/m] 10 7
8.4844 5.8945 4.7415 4.0521 3.5814 3.2344 2.9658 2.7504 2.5734 4.6560 4.0755 3.8018 3.6379 3.5298 3.4553 3.4034 3.3677 3.3445 0.6521 0.3442 0.1424 -0.0188 -0.1586 -0.2854 -0.4034 -0.5153 -0.6227 1.5066 1.5749 1.6261 1.6678 1.7033 1.7341 1.7614 1.7857 1.8076 1.0840 4.8069 3.0295 2.2017 1.7295 1.4269 1.2177 1.0652 0.9495
[Nsec/m] 105
0.2292 1.1958 0.8249 0.6361 0.5208 0.4426 0.3858 0.3426 0.3085 0.2430 1.2656 0.8703 0.6692 0.5466 0.4636 0.4036 0.3579 0.3219 0.1333 0.8750 0.6978 0.6008 0.5387 0.4953 0.4631 0.4382 0.4184
6.6 Physical significance of the bearing coefficients

In order to better understand the physical significance of the eight linearized dynamic bearing coefficients, it is useful to calculate the work done along a closed orbit by the bearing forces [6.7]. First, the non-symmetric stiffness and damping matrices are decomposed into their symmetric (s) and skew-symmetric (a) parts
[ k ] = [ k ]+ [ k ] ,
b b s b a
[ c ] = [ c ] + [ c ],
b b s b a
(6.36)
where [ k ]+ [ k ] [k ]= 1 2
b s b b T
[ k ] [ k ] [k ]= 1 2
b a b
b T
(6.37) (6.38)
[ c ]+ [ c ] [c ]= 1 2
b s b
b T
[ c ] [ c ] [c ]= 1 2
b a b
b T
The damping matrix can be written
[c ]
b
c yy = 1 c yz + c zy 2
1 c yz + c zy 0 2 + 1 c yz c zy c zz 2
1 c yz c zy 2 , 0
104
[c ]
b
b is not a true damping matrix, in the dissipative sense. It The matrix ca embodies a conservative gyroscopic-like force field which, in the absence of other true gyroscopic effects, produces the bifurcation of eigenfrequencies along forward and backward precession branches. Indeed, the incremental work done by the corresponding forces at any point on any orbit is zero:
[ ]
cs = yy s c zy
0 cs yz + s a c zz c zy
ca yz . 0
(6.39)
0 ca yz dW = a c zy 0
y & dy & dt y a & & c zy z y , = ca yz & dz & dt z z
a & d t ) + c zy & (z &y &z & (y & d t ) = c a &y & ) dt 0 . dW = c a y yz z yz ( z
(6.40)
This conservative property shows that the force vector is always perpendicular to the velocity vector { }. For most journal bearings the crosscoupling damping coefficients are equal and the skew-symmetric matrix vanishes. The stiffness matrix can be written
[k ]
b
k yy = 1 k yz + k zy 2
)
s yy s zy
1 k yz + k zy 0 2 + 1 k zz k yz k zy 2
1 k yz k zy 2 , 0 (6.41)
[ k ] = k k
b
b The matrix k a is not a true stiffness matrix since it provides no radial restoring force. It embodies a non-conservative circulatory force. The incremental work done by this force at any point on any orbit can be written as
[ ]
s 0 k yz + s a k zz k zy
a k yz . 0
0 dW = a k zy
a k yz y dy d y a a z k zy y , = k yz dz 0 z dz
a a dW = k yz z dy + k zy y dz = f y d y + f z d z .
(6.42)
fz , dW is not an exact differential. The work done z y between two points is path dependent and thus the force is non-conservative. Because
fy

b The net energy exchange per cycle to the rotor resulting from k a is
a Ecycle = k yz
105
[ ]
( z dy y d z ) .
(6.43)
Splitting this integral into two line integrals between points A ( y A = ymin ) and B ( y B = ymax ) (Fig. 6.17), and integrating the dy terms by parts, yields ( z2 > z1 ) :
Ecycle =
a 2 k yz
yA
(z
yB
z1 ) d y .
(6.44)
The integral equals the closed orbit area. It is positive for forward a precession and negative for backward precession. Generally k yz 0 for journal
b bearings. Thus the k a effect represents negative damping on forward whirls and positive damping on backward whirls. Only for straight-line (degenerated)
[ ]
orbits, where
yA
yB
dy =
yA
z d y , is the net cyclic exchange of energy zero.

1
yB
Fig. 6.17
For a closed precession orbit described by the parametric equations

z = Az cos ( t + z ) , y = Ay cos t + y ,
(6.45)
the work done by the bearing forces (pay attention to the negative sign)
106
& c yz z &, Fy = k yy y k yz z c yy y & c zz z &, Fz = k zy y k zz z c zy y
(6.46)
over one cycle is

W=
(F d y + F d z ) = (F y& + F z&)dt .
y z y z 0
(6.47)
The following integrals can be easily calculated:

& + k zz z z & ) dt = 0 , ( k yy y y
0 T
& + k zy y z & ) dt = ( k yz k zy ) Ay Az sin ( y z ) , ( k yz z y

0
& ( c yy y
0 T
2 2 & 2 dt = c yy Ay + c zz z + c zz Az ,
& + c zy y &z & ) dt = ( c yz + c zy ) Ay Az cos ( y z ) . ( c yz z& y

0
The net energy exchange per cycle to the rotor system is

W = k yz k zy Ay Az sin y z
2 2 c yz + c zy Ay Az cos y z c yy Ay + c zz Az ,
(6.48)
where the positive term represents supplied energy, while negative terms represent dissipated energy. With k yz positive, the cross-coupling stiffness coefficients can supply energy to the motion, hence provide negative damping. The first term may also be written
Wsup = k yz k zy Ay Az sin y z =
= k yz k zy
( )( Aysin y Az cos z Ay cos y Azsin z ) = = ( k yz k zy ) ( y s zc + z s yc ) = ( k yz k zy )a b ,

( )
2
Wsup = k yz k zy rf
rb
(6.49)
107
where the elliptical orbit has been decomposed into two circular orbits, one with forward precession and radius r f , and one with backward precession and radius
rb .
For k yz > 0 , if negative damping.
r f > rb , b > 0 , the precession is forward and
Wsup > 0 . Thus, it is the forward whirl component which is responsible for the
6.7. Bearing temperature

In any specific application, the bearing geometry, the static load, the type of lubricant and its properties are given. To obtain the bearing coefficients at any selected speed it is necessary first to determine the lubricant viscosity. This could be done by an iterative procedure [6.4].
6.7.1 Approximate bearing temperature

A temperature value is estimated and the corresponding viscosity value is found from a viscosity-temperature diagram for the lubricant (Fig. 6.18). So it is possible to compute the Sommerfeld number and thereafter to enter the tables of bearing data to find the following parameters: - the dimensionless side flow
Q= Q 1 N D LC 2 , (6.50)
where Q is the lubricant flow obtained by integrating the pressure gradient at the edge of the bearing; - the dimensionless friction power loss P= CP , N 2 L D3
3
(6.51)
where P is the power loss by friction; - the dimensionless temperature rise
108
T =
T R
cv C
2
(6.52)
The temperature rise T through the lubricant film can be calculated from a simplified version of the energy equation, neglecting heat conduction. For an infinitesimal axial film strip of thickness hmin and width R d , the heat balance yields
cv q z
( R ) 2 , dT R R d hmin
where cv is the specific heat per unit volume and T is the temperature. Under the above assumptions q z is equal to the shear flow
qz =
1 hmin R . 2
(6.53)
The heat balance equation reduces to dT 2 2 = R 2 . d c v hmin (6.54)
The viscosity is a given function of the temperature T, for any particular lubricant. The temperature, hence, the viscosity variation along the film can be computed from
T1
dT
2 2 = R cv
d . 2 hmin
However, it is common practice to assume = const . , equal to an average value, so that the total increase in temperature
2 T = R 2 cv
d . 2 hmin
(6.55)
By assuming that, say, 80 per cent of the friction heat is carried away by the lubricant, the operating temperature of the bearing, Toper , results from a simple heat balance as
109
Toper = Tsup + 0.8
R
cv
P 4 , Q C
(6.56)
where Tsup is the lubricant supply temperature. The maximum temperature in the film is Tmax = Toper + T = Toper +
T . cv C
(6.57)
If Tmax differs from the estimated temperature, a new calculation should be performed using a viscosity value based on Tmax and repeated until the two temperatures agree. Once the true Sommerfeld number has been obtained, the corresponding spring and damping coefficients can be found either from tables, usually requiring interpolation, or from formulae.
6.7.2 Viscosity-temperature relationship

The viscosity of lubricating oils is extremely sensitive to the operating temperature (Fig. 6.18). With increasing temperature, the viscosity of oils falls quite rapidly (up to 80% for a temperature increase of 250 C ). It is important to know the viscosity at the operating temperature, because it determines the Sommerfeld number used in the estimation the bearing dynamic coefficients. The oil viscosity at a specific temperature can be either calculated from a viscositytemperature equation or obtained from the viscosity-temperature ASTM chart.
Usually kinematic viscosity is measured in centistokes (cSt) and the dynamic (absolute) viscosity is measured in centipoise (cP). The water at 200 C has a viscosity of 1.0020 cP. In SI units, a millipascal sec , or mPa s , is equal to a centipoise and 1cSt = 1 mm 2 s . Walthers equation [6.8] has the form
( + a ) = b d 1 T
where a, b, c, d are constants, T is the absolute temperature, K, and is the kinematic viscosity, m 2 s . Kinematic viscosity is defined as the ratio of dynamic viscosity to fluid density, = . The typical density of a mineral oil is 850 kg m 3 . The most widely used chart is the ASTM D341 viscosity-temperature chart, which is entirely empirical and is based on Walthers equation.
110
In deriving the bases for the ASTM chart, logs were taken from Walthers equation and it was considered that d = 10 , yielding log10 ( + a ) = log10b + 1 T c .
Fig. 6.18 (from [6.2])
It has been found that if is in cS (centistokes), then a 0.7 . After substituting into the equation, the logs were taken incorrectly as
111
log10 log10 ( cS + 0.7 ) = a c log10T ,
(6.58)
where a and c are constants. To use this equation, the kinematic viscosity should be larger than 2 cS. For lower viscosities, the constant 0.7 increases according to relations in ASTM D341. In the ASTM chart, the ordinate is log10 log10 ( cS + 0.7 ) and the abscissa is log10 T . Despite the mathematical error, the ASTM chart is quite successful and works well for mineral and synthetic oils under normal conditions. Figure 6.18 is a Walther-Ubbelohde chart for oils used in internal combustion engines. A similar chart exists for turbine oils [6.2]. The major factor in lubrication is that viscosity changes very strongly with temperature and pressure. A general rule is that the more viscous the oil the more susceptible it is to change. The thicker the oil the larger the viscositytemperature derivative. The viscosity index (VI) is commonly used to indicate the relative decrease in viscosity with increasing temperature (Dean and Davis, 1929). The standard temperatures used in the oil industry are 100 0 F and 2100 F
( 37.8 C and 99 C ). The best paraffinic mineral oils (of Pennsylvania) were given
0 0
a viscosity index of 100 and the poorest naphtenic oils (of Texas) a V.I. of zero. The American Society of Automotive Engineers (SAE) have divided oils into
grades. A 5W oil has a maximum viscosity of 1200 centipoises at 00 F and a

minimum viscosity of 3.9 centipoises at 2100 F . A 10W oil has a viscosity between 1200 and 2400 centipoises at 00 F and a minimum viscosity of 3.9 centipoises at 2100 F . A 20W oil has a viscosity between 2400 and 9600 centipoises at 00 F and a minimum viscosity of 3.9 centipoises at 2100 F . An SAE 20 oil has a viscosity between 5.7 and 9.6 centistokes at 2100 F . Furthermore SAE 30, SAE 40, SAE 50 oils have viscosities between 9.6-12.9, 12.9-16.8 and 16.8-22.7 centistokes, respectively. Some oils, with polymers added to them, have high viscosity indices (about 150), and are called multigrade oils. This is because they are in one grade at the 00 F end and in a higher grade at 2100 F . As an example, a 10W/30 oil may have a viscosity of 2100 cP at 00 F and 11.5 cS at 2100 F . It falls into the 10W range at 00 F and SAE 30 at 2100 F . For this reason it is called 10W/30 [6.8]. It would be advantageous to produce a thick oil with the viscositytemperature derivative of a thinner oil. To some extent (2-3 per cent) this can be done by the addition of polymer thickeners, often a methacrylate. The result is also a multigrade oil. An oil may be an SAE 10 at 1000 F but SAE 30 at 2100 F , hence the oil is designated SAE 10/30.
112
6.8 Common fluid film journal bearings

In the evolution of rotating machinery many different bearing configurations have been developed in order to achieve acceptable performance, most of then aiming to improve the rotor stability.
6.8.1 Plain journal bearings

The plain journal bearing presented so far (Fig. 6.2) is the simplest fixedgeometry design, suitable for highly-loaded or low-speed rotors which are not subject to rotordynamic instability. Low cost and ease of manufacture are the main advantages. In the centered position, the bearing plain cylindrical shell is concentric with the journal surface, so there is no preload. Plain sleeve bearings have the highest cross-coupling of all bearings, making them the most destabilizing configuration.
6.8.2 Axial groove bearings

The most common cylindrical journal bearing (Fig. 6.19) is horizontally split with axial oil distribution grooves (or relief pockets) along each horizontal split line.
Fig. 6.19
Fig. 6.20
The four-axial-groove journal bearing (Fig. 6.20) incorporates four axial grooves, 90 0 apart, located at 450 from the vertical axis. There is no preload, but the design is more stable than the plain journal bearing for some applications. Table 6.3 lists the dimensionless dynamic coefficients for a two-axialgroove bearing with L D = 0.5 as a function of Sommerfelds number [6.1]. Figure 6.21 presents the dimensionless dynamic coefficients for a bearing with L D =1.
6. FLUID FILM BEARINGS Table 6.3 Dynamic coefficients of two-axial-groove bearing, L D = 0.5 Stiffness coefficients Damping coefficients
113
S
5.96 4.43 2.07 1.24 0.798 0.517 0.323 0.187 0.135 0.0926 0.0582 0.0315 0.00499
K yy
1.77 1.75 1.88 2.07 2.39 2.89 3.65 4.92 5.90 7.35 9.56 13.8 43.3
K yz
13.6 10.3 5.63 4.27 3.75 3.57 3.62 3.88 4.11 4.46 4.92 5.76 9.61
K zy
-13.1 -9.66 -4.41 -2.56 -1.57 -0.924 -0.427 0.0235 0.258 0.527 0.805 1.24 4.26
K zz
2.72 2.70 2.56 2.34 2.17 2.03 1.92 1.83 1.80 1.78 1.73 1.72 1.99
C yy
27.2 20.6 11.2 8.50 7.32 6.81 6.81 7.32 7.65 8.17 9.12 10.6 22.4
C yz
2.06 2.08 2.22 2.32 2.24 2.10 2.08 2.16 2.10 2.05 2.06 2.03 2.93
C zy
2.06 2.08 2.22 2.32 2.24 2.10 2.08 2.16 2.10 2.05 2.06 2.03 2.93
C zz
14.9 11.4 6.43 4.73 3.50 2.60 2.06 1.70 1.46 1.24 1.06 0.846 0.637
a
Fig. 6.21 (from [6.9])
The average values used in simplified calculations [6.9]
K av = C av
1 K yy + K zz + C yz C zy , 2 2 1 1 = C yy + C zz + K zy K yz , 2 2
(6.59)
114
are plotted with broken lines.
6.8.3 Pressure dam bearings

A pressure dam bearing (Fig. 6.22) is basically a plain bearing modified to incorporate a central relief groove or scallop along the top half of the bearing shell, ending abruptly at a step. As the lubricant is carried around the bearing, it encounters the step that produces an increased pressure at the top of the journal (Fig. 6.23), inducing a stabilizing force downwards. This forces the rotor into the bottom half of the bearing to a greater eccentricity.
Fig. 6.22
Fig. 6.23
Pressure dam bearings have greater power consumption than plain bearings and are more expensive to manufacture because of the precise machining required to produce the correct dam geometry. They have high load capacity and are beneficial in improving rotordynamic stability. An improved configuration incorporates a circumferential relief track cut into the lower pad, as shown schematically in Fig. 6.24. This groove reduces the L D ratio of the load-carrying bearing segment and increases the bearing unit loading.
Fig. 6.24
115
Figure 6.25 presents the dimensionless dynamic coefficients for a pressure dam bearing with L D = 1 as a function of Sommerfelds number. Numerical values of these coefficients are given in Table 6.4 [6.1].
Fig. 6.25 (from [6.1]) Table 6.4 Dimensionless dynamic coefficients of pressure dam bearing, L D = 1 Stiffness coefficients Damping coefficients
S
1.72 1.09 0.742 0.517 0.362 0.249 0.165 0.102 0.0558 0.0219
K yy
4.59 4.67 4.72 4.85 5.43 6.20 7.57 9.69 14.10 30.0
K yz
8.55 6.8 5.79 5.22 5.11 4.85 5.15 5.33 6.28 7.86
K zy
-2.77 -0.832 -0.273 -0.0993 0.231 0.414 0.697 0.985 1.38 2.91
K zz
5.3 4.19 3.34 2.85 2.50 2.21 2.17 2.05 1.99 1.82
C yy
14.0 10.8 9.24 8.08 8.30 8.01 8.58 8.95 10.8 13.0
C yz
6.15 4.38 3.46 2.95 2.84 2.63 2.42 2.39 2.33 1.77
C zy
5.90 4.37 3.46 2.95 2.84 2.64 2.43 2.40 2.34 1.78
C zz
8.44 5.18 3.6 2.72 2.11 1.70 1.52 1.21 1.01 0.555
116
DYNAMICS OF MACHINERY Table 6.5 Dimensionless dynamic coefficients of offset bearing, L D = 0.5 , m p = 0.5 Stiffness coefficients Damping coefficients
S
8.519 4.240 2.805 2.081 1.339 0.953 0.717 0.555 0.493 0.353 0.284 0.228 0.182 0.162 0.143 0.126
K yy
47.06 23.60 15.81 11.93 8.08 6.18 5.14 4.63 4.56 4.63 4.85 5.18 5.65 5.93 6.26 6.64
K yz
82.04 41.06 27.42 20.61 13.79 10.39 8.45 7.20 6.72 5.78 5.40 5.15 5.01 4.97 4.95 4.95
K zy
5.48 2.64 1.65 1.12 0.54 0.20 -0.05 -0.09 0.01 0.22 0.33 0.42 0.51 0.55 0.606 0.65
K zz
64.74 32.32 21.49 16.05 10.56 7.78 6.15 5.00 4.53 3.53 3.08 2.74 2.48 2.37 2.27 2.19
C yy
97.59 49.04 32.97 25.01 17.15 13.34 11.29 10.00 9.49 8.51 8.17 7.99 7.95 7.97 8.02 8.10
C yz
45.00 22.62 15.22 11.56 7.98 6.31 5.43 4.76 4.38 3.56 3.18 2.88 2.65 2.53 2.46 2.38
C zy
45.00 22.62 15.22 11.56 7.98 6.31 5.43 4.76 4.38 3.56 3.18 2.88 2.65 2.53 2.46 2.38
C zz
59.71 29.94 20.06 15.15 10.25 7.83 6.51 5.38 4.74 3.40 2.79 2.34 1.98 1.82 1.69 1.56
6.8.4 Offset halves bearings

The offset split journal bearing is shown in Fig. 6.26, a. It is preloaded in the horizontal direction by displacing the upper and lower halves transverse to the journal axis a slight amount d, usually about one half the radial clearance. Offset bearings have increasing load capacities as the offset is increased.
a
Fig. 6.26
117
It is manufactured by offsetting the halves before finish boring so that when the halves are matched, the pad centers will be horizontally offset. The horizontal clearance is smaller than the vertical clearance. The vertical clearance equals the pad machined clearance. The horizontal diametral clearance is equal to the pad machined diametral clearance, minus twice the radial offset. This bearing is more stable than the plain journal bearing, but there is still a tendency for instability and it is unidirectional. Table 6.5 lists the dimensionless dynamic coefficients for an offset bearing with L D = 0.5 and d = 0.5C as a function of Sommerfelds number [6.4].
6.8.5 Multilobe bearings

These consist of a number of fixed bearing segments bored to a larger radius than the bearing set clearance, thus creating a preload. While a grooved bearing consists of a number of partial arcs with a common center, the lobed bearing is made up of partial arcs whose centers do not coincide.
6.8.5.1 Bearing preload
Consider the journal in the concentric position (Fig. 6.27). The bearing pad (lobe), with machined bore radius RP , is centered at point O p . The bearing center is OB OJ . The set bore has a radius RB which is the distance from the bearing center to the pad surface at the point of minimum clearance.
Fig. 6.27
118
If OP coincides with OB , then RP = RB and the preload is zero. In this case, if the journal is centered in the bearing, the pad arc will be concentric with the journal surface. If the pad bore is larger than the set bore, i.e. RP > RB , the centers of curvature will be offset. The preload factor m p is defined as mp = 1 Cb , Cp (6.60)
where C p = RP R is the machined clearance, usually taken as the representative bearing clearance, and Cb = RB R is the assembled clearance. If the pad machined bore is held at a fixed value, the preload is increased by moving the pad radially inward toward the shaft. The distance between the pad center of curvature and the bearing center d = RP RB = C p Cb = m p C p . (6.61)
The preload forces the oil to converge at each pad because of the reduced clearance at the midpoint of the pad. The result is an oil wedge effect.
6.8.5.2 Elliptical (lemon-bore) bearings
The elliptical bearing consists of two partial arcs (Fig. 6.28). The center of the bottom arc is above the bearing center, while the center of the upper arc is below the bearing center. This arrangement has the effect of preloading the bearing. The tight clearances at the minimum points create higher minimum temperatures, but the open clearances at the splits provide large amounts of cool oil.
a
Fig. 6.28
In the elliptical bearing, the shells are first put together with shims, bored, and then assembled after removal of the shims, as shown in Fig. 6.29.
119
The pressure distribution on the journal is shown in Fig. 6.30 for a particular design. The wedge formed in the top half creates a downward pressure which suppresses the whirl in marginally unstable bearing designs. Also, the journal center eccentricity is increased, which results in increased stiffness.
Fig. 6.29 (from [6.1])
Fig. 6.30
Figure 6.31 presents the dimensionless dynamic coefficients for an elliptical (two-lobe) bearing with L D = 0.5 and m p = 3 4 , for Reynolds boundary conditions, as a function of Sommerfelds number. Numerical values of these coefficients are given in Table 6.6 [6.1].
Table 6.6 Dimensionless dynamic coefficients of elliptical bearing, L D = 0.5 , m p = 3 4 Stiffness coefficients Damping coefficients
S
5.06 3.01 2.04 0.983 0.477 0.286 0.197 0.0999 0.0504 0.0306 0.0199 0.00976
K yy
474 282 186 93.7 46.6 29.5 22.4 15.6 14.7 16.9 20.1 29.8
K yz
72 40.6 26.5 12.7 6.83 4.91 3.88 3.82 4.36 5.21 6.32 9.04
K zy
-215 -128 -86.7 -43.2 -20.4 -12.3 -8.44 -3.75 -1.11 0.0379 1.08 2.59
K zz
49.8 29.7 18.2 9.79 4.79 3.24 2.35 1.61 1.4 1.49 1.62 1.93
C yy
548 319 215 106 50.9 33 24.7 16.1 12.3 11.2 12.1 15.8
C yz
-157 -94.3 -63.8 -30.8 -15 -8.45 -5.76 -2.13 0.085 0.582 1.24 2.63
C zy
-157 -94.1 -63.6 -30.7 -15 -8.42 -5.75 -2.09 0.0901 0.586 1.25 2.66
C zz
92.5 54.4 36.9 17.9 8.77 5.45 3.93 2.24 1.32 0.978 0.866 0.84
120
a
Fig. 6.31 (from [6.1])
6.8.5.3 Three-lobe and four-lobe bearings
Three lobe (Fig. 6.32, a) and four-lobe (Fig. 6.33) bearings have fixed segments (called lobes) bored to a larger radius than the bearing radius.
a
Fig. 6.32
There are also configurations with five or over ten lobes. The lobes can be converging and diverging across each segment. Tapered land designs have constantly converging segments. The inner surface of the three-lobe bearing (Fig. 6.32, b) is composed of three arc segments whose centers are displaced from one another and from the common bearing center, providing a preload. The three lubricant wedges around the journal stiffen and stabilize the bearing, even at the centered position. Increasing the number of lobes, more lubricant wedges are formed around the journal, improving the stability. A wide range of preloads is possible by changing the pad and set bores.
121
Fig. 6.33
Figures 6.34 and 6.35 present the dimensionless dynamic coefficients for a three-lobe and a four-lobe bearing, respectively, with L D = 0.5 and m p = 3 4 , for Reynolds boundary conditions, as a function of Sommerfelds number. Numerical values of these coefficients are given in Tables 6.7 and 6.8 [6.1].
a
Fig. 6.34 (from [6.1])
a
Fig. 6.35 (from [6.1])
122
DYNAMICS OF MACHINERY Table 6.7
Dimensionless dynamic coefficients of three-lobe bearing, L D = 0.5 , m p = 3 4 Stiffness coefficients Damping coefficients
S
1.27 0.625 0.299 0.187 0.131 0.095 0.070 0.0516 0.0374 0.0256 0.0204
K yy
73.2 38.4 21.9 17.1 15.3 14.8 15.8 16.0 17.8 20.8 23.2
K yz
36.4 18.2 8.52 5.56 4.28 3.47 3.03 2.93 3.03 3.44 3.86
K zy
-42.4 -23.4 -12.5 -8.56 -6.88 -5.68 -4.72 -4.04 -3.32 -2.41 -1.92
K zz
64.8 30.9 14.2 8.92 6.48 5.04 4.24 3.55 3.10 2.70 2.444
C yy
88.4 48.0 26.2 19.2 16.0 13.84 12.5 11.8 11.4 11.6 12.0
C yz
-3.45 -2.52 -2.16 -1.80 -1.67 -1.52 -1.17 -0.944 -0.688 -0.298 -0.128
C zy
-3.31 -2.44 -2.12 -1.77 -1.67 -1.50 -1.29 -0.972 -0.688 -0.288 -0.112
C zz
85.6 44.0 20.5 12.2 9.04 6.92 5.20 4.28 3.56 2.58 2.28
Table 6.8 Dimensionless dynamic coefficients of four-lobe bearing, L D = 0.5 , m p = 3 4 Stiffness coefficients Damping coefficients
S
1.24 0.614 0.293 0.182 0.123 0.0853 0.0593 0.04 0.0256 0.0147 0.0103
K yy
64.6 33.2 18.7 14.7 13.5 13.6 14.6 16.5 19.8 25.9 31.4
K yz
43.9 22.4 11.8 8.19 6.32 5.36 4.81 4.53 4.56 5.14 5.82
K zy
-44.2 -22.2 -11.3 -7.33 -5.62 -4.61 -3.53 -3.08 -2.62 -1.74 -1.36
K zz
65.0 32.6 16.5 11.1 8.52 6.92 5.85 5.04 4.32 3.61 3.07
C yy
98.5 50.7 26.7 19.1 15.3 13.6 12.7 12.5 13.0 14.6 16.4
C yz
-0.068 0.036 0.136 0.0967 0.196 0.288 0.236 0.376 0.492 0.56 0.776
C zy
0.14 0.14 0.192 0.132 0.224 0.308 0.256 0.392 0.508 0.58 0.792
C zz
98.4 49.3 24.2 15.3 11.3 8.89 6.38 5.32 4.35 2.70 2.28
123
6.8.5.4 Pressure dam multi-lobe bearings
The performance of a series of non-cylindrical bearings such as elliptical, offset-halves, orthogonally displaced and multi lobe bearings is improved by incorporating pressure dams. Figure 6.36 shows a four-lobe pressure dam bearing.
Fig. 6.36 (from [6.10])
Figures 6.37 show the circumferential variation of fluid film pressures at the center line of a conventional four-lobe bearing and a four-lobe pressure dam bearing, for two extreme eccentricity ratios.
a
Fig. 6.37 (from [6.10])
In the ordinary four-lobe bearing (Fig. 6.37, a), lobes 1 and 4 are either cavitated over most of the surface or have low pressures. When pressure dams are
124
incorporated in the lobes 1 and 4 (Fig. 6.37, b), peak pressures are generated at the dam locations, followed by a steep fall in the pressures. These pressures are distributed over a much wider range of the surface and their peaks are much higher than those arising in the conventional design. There is also a comparative decrease of the peak pressures is the lower lobes, where relief tracks are used. The generation of high pressures in the upper half of the bearing improves its stability. The distribution of pressures throughout the circumference ensures a stable operation of the bearing.
6.8.6 Tilting pad journal bearings

The tilting-pad journal bearing consists of several individual journal pads that can pivot in the bore of a retainer. The tilting pad is like a multilobe bearing with pivoting lobes, or pads. The most common tilting-pad bearing arrangements have four or five pads.
Fig. 6.38
Fig. 6.39
Figure 6.38 illustrates the rocker pivot arrangement. This design is the simplest and least expensive to manufacture. The pad design allows tilting motion in the circumferential direction but none axially. Since the support contact is a line, the pivot stresses may be high if a good alignment is not achieved. The spherical point pivot arrangement is illustrated in Fig. 6.39. A spherical button is mounted in either the pad or the housing and pivots on a hardened flat disk in the opposite member. This allows tilting in all directions. In the spherical surface pivot design (Fig. 6.40), the pad load is transmitted into the housing through a ball and socket arrangement. This allows the pad to pitch in the conventional manner but can also accommodate shaft misalignment. The spherical pivot tends to be more durable than the rocker pivot, as the surface contact of the ball-and-socket joint has a lower unit loading than the line contact of the rocker-pivot. The sliding motion between the ball and socket helps avoiding fretting. Like the fixed-geometry bearings, there is a thin layer of babbitt metal (50 to 125 m ) applied to the bearing surface to protect the shaft. Recently, flexible (flexural) pad bearings are of increased attention.
125
Fig. 6.40
Fig. 6.41
Tilting pads may be mounted offset. Pad offset is a measure of how far along the pad arc the pivot is located (Fig. 6.41) Offset =
pivot . pad
(6.62)
An offset of 0.5 corresponds to a centered pivot, where the pivot is positioned symmetrically between the leading and trailing edges. Offsetting the pivots in the direction of rotation to values grater than 0.5 can raise the minimum film thickness, and hence the load capacity. A common offset value is about 0.6. Application of offset pivots is limited to unidirectional operation. The value of offset is meaningful only when the bearing is preloaded. Preload can be accomplished by boring the pad arcs to a larger diameter than the clearance diameter. Typical preload values range from 0 to 0.5, with the most common being about 0.3. An increase in preload will increase the stiffness, but will result in a reduction of the damping in the bearing.
a
Fig. 6.42 (from [6.11])
The primary advantage of this design is that each pad can pivot independently to develop its own pressure profile. This feature significantly reduces the cross-coupled stiffness. The pressure distributions on the journal of a four-pad bearing with rigid pivots are shown for a vertical load in Fig. 6.42, a and
126
for a load on an arbitrary angle in Fig. 6.42, b. The bearing has L D = 0.7622 , m p = 0.5 , arc length 72 0 , pad angle 450 and offset 0.5. Figures 6.43, a and b present the dimensionless dynamic coefficients of a four-pad tilting pad bearing with load between pads (LBP), as a function of Sommerfelds number, for Reynolds boundary conditions. The bearing has L D = 0.5 , m p = 2 3 , pad angle 800 and offset 0.5. The bearing is isotropic, with equal direct stiffness and damping coefficients, respectively. It is intrinsically stable, having zero cross-coupled stiffness coefficients.
a
Fig. 6.43 (from [6.1])
The pressure distributions on the journal of a five-pad bearing are shown in Fig. 6.44. The bearing has L D = 1 , offset 0.5, pivot angle 360 , and no preload. Figure 6.44, a is for a LOP (load on pad) bearing. Figure 6.44, b is for the same bearing with load between pads (LBP). The attitude angle is zero in both cases.
b
Fig. 6.44 (from [6.11])
127
Figure 6.44, c shows the pressure distribution for the load vector between the pad and the center of the oil groove. In this case, since the load is not symmetric with respect to the pads, the bearing attitude angle is not zero and the cross-coupling stiffness and damping coefficients are not zero. Table 6.9 lists the dimensionless dynamic coefficients for a five-pad tilting pad bearing with L D = 0.5 and m p = 3 4 , for load between pads and Reynolds boundary conditions, as a function of Sommerfelds number [6.1].
Table 6.9 Dimensionless dynamic coefficients of five-pad bearing, L D = 0.5 , m p = 3 4 , LBP Stiffness coefficients Damping coefficients
S
16.1 6.42 3.21 1.57 0.515 0.243 0.120 0.0763 0.0484 0.0356 0.0264 0.0187 0.0128
K yy
2010 803 401 196 66.1 32.8 18.3 14.5 12.9 12.7 13.0 13.8 15.7
K yz
0 0 0 0 0 0 0 0 0 0 0 0 0
K zy
0 0 0 0 0 0 0 0 0 0 0 0 0
K zz
2010 803 401 196 65.2 31.6 16.5 11.7 9.18 8.31 7.95 8,00 8.78
C yy
1230 490 245 120 40.2 19.7 10.5 7.76 6.33 5.82 5.59 5.60 5.80
C yz
0 0 0 0 0 0 0 0 0 0 0 0 0
C zy
0 0 0 0 0 0 0 0 0 0 0 0 0
C zz
1230 490 245 120 39.8 19.2 9.85 6.72 4.93 4.17 3.72 3.48 3.41
The load between pad (LBP) configuration is used when the most important features are higher load capacity and better synchronous response. In this case the damping by the effective support area is larger than for LOP bearings. Heavy rotors running at relatively low speeds (low Sommerfeld number) have LBP tilting pad bearings. Light rotors running at high speeds predominantly have load on pad (LOP) bearings. This configuration provides asymmetry in the support, which enhances the stability, which is the most important required characteristic. For a bearing with five 600 tilting pads, centrally pivoted, with L D = 0.5 , L B = 0.955 (where B is the pad arc length), no preload and pad inertia neglected, the dimensionless stiffness and damping coefficients are
128
shown as functions of the Sommerfeld number for a LBP bearing in Fig. 6.45, a and for a LOP bearing in Fig. 6.45, b.
a
Fig. 6.45 (from [6.12])
For the LOP bearing, a sudden reduction in the coefficients K yy and K zz appears when approaching the condition of pad resonance. This does not entail a loss of the overall bearing stiffness, since at resonance the pad inertia cannot be neglected, and the cross-coupling terms become important and replace the direct-coupling terms. However, in this case the bearing loses its inherently stable characteristic. Pad resonance occurs in general at high Sommerfeld numbers, where the minimum film thickness is large and the damping is high.
6.8.7 Rayleigh step journal bearings

The stepped pad bearing (Rayleigh pad bearing) is known as the bearing with the highest load capacity amongst all other possible bearing geometries. This is based on an observation of Lord Rayleigh (1918) that, when side leakage is neglected (infinitely wide bearing), the step film shape has the greatest load capacity in a slider bearing lubricated with an incompressible fluid. Figure 6.46, a illustrates the journal in a concentric position. In this case, the number of steps placed around the bearing has to be limited to one in order to obtain a non-zero load capacity. The main parameters are Cr - the radial clearance
129
in the ridge region, C s - the radial clearance in the step region, - the angle subtended by the step region, - the angle subtended by the ridge region, and the angle subtended by the lubrication groove. Figure 6.46, b depicts the eccentric Rayleigh step journal, where is the attitude angle and is the eccentricity orientation angle. A pad is defined as the portion of a bearing included in the angle subtended by the ridge, step, and lubrication groove. Each pad acts independently since the pressure is reduced to ambient at the lubrication supply grooves. Figure 6.46, c shows a schematic view of the geometry of a four-pad Rayleigh step bearing.
b
Fig. 6.46 (from [6.13])
In a single step Rayleigh journal bearing, a positive pressure is developed completely around the journal. The pressure distribution is triangular for a radiusto-length ratio equal to zero, with the peak value at the step and there is an optimal step configuration yielding maximum load capacity. Side leakage is usually more pronounced than in other designs. Side rails are placed at the ends of the bearing to restrict the axial flow and increase the load capacity. The study of the dynamic characteristics of the incompressibly lubricated Rayleigh step journal bearing is under development.
6.8.8 Floating ring bearings

The floating ring bearing finds application in turbochargers, where the rotor is light and runs at very high speeds. It is basically a plain cylindrical journal bearing, but with a loose bushing inserted in the clearance space between the journal and the bearing housing (Fig. 6.47). There are two lubricant films, an inner film identified by subscript 1, and an outer film identified by subscript 2. Each film can be considered as a plain journal bearing.
130
The oil-film formed between the ring and the bearing housing provides damping which is thought to be useful in improving the rotor stability as it spins in the inner film.
Fig. 6.47
Almost all rotors of automotive turbochargers exhibit strong sub-harmonic vibrations that cause noise (rumble) and rotor-stator rub. Their study is the subject of intense actual research efforts.
6.9 Squeeze film dampers

Squeeze film dampers are employed to improve the vibration characteristics of rotors by adding damping to either stabilize otherwise unstable rotors or to reduce synchronous rotor amplitudes. These are a special case of a hydrodynamic journal bearing where the rotation speed is zero and the hydrodynamic pressure forces are produced by the lubricant being squeezed between the journal and the bearing surface.
6.9.1 Basic principle

Squeeze-film dampers (SFD) are non-linear elements that are widely used in high-speed rotor-dynamic systems to attenuate vibrations and transmitted forces. The journal of a SFD is formed by a ring (sleeve) that is tightly fitted (fixed) on to the outer race of a rolling-element bearing (or the bush of a sleeve bearing). The SVD itself is an oil-film filling the annular clearance between the journal and the
131
bearing housing. The journal is prevented from rotating by some kind of mechanical restraint, but it is free to orbit within the annular clearance.
Fig. 6.48 (from [6.14])
An SFD is also often used on its own, between a bearing and its housing, to reduce vibrations while running through the rotor critical speeds. Rotation of the inner member of the SFD is in this case prevented by antirotation pins or dogs. The damper becomes part of a series structure of rotor, damper and housing flexibility. Unlike the hydrodynamic journal bearing, an SFD is inherently stable, i.e. it cannot introduce self-excited vibrations. However, in the presence of rotating out-of-balance excitation, the forced periodic vibrations of a squeeze-film damped rotating system can become unstable, resulting in either amplitude jump phenomena or undesirable non-synchronous frequency components that are not simple integer multiples of the rotational speed. In practical applications, an SFD is used either with or without a parallel retainer spring (Fig. 6.48). Retainer springs help to attenuate non-linear effects, especially if preloaded to centralize the journal in its housing. It typically consists of an annulus with a clearance of 25 to 500 m that contains the oil film and surrounds the outer race of the rolling element bearing. The squirrel cage constrains the rolling element bearing from rotating, preventing the shaft to reach the oil film. It also acts as a light retainer spring that centers the journal in the damper clearance. Other designs of squeeze film dampers do not incorporate squirrel cages; they rely instead on a dog mechanism to constrain the outer race of the rolling element from rotating. In such dampers the journal bottoms down in the damper clearance if no precession occurs. The stationary oil film is the main characteristic of squeeze film dampers and is, in fact, the only difference between a squeeze film damper and a journal bearing, both of which are hydrodynamic bearings. The oil film in journal bearings rotates due to the rotation of the rotor even if there is no precession. The rotating
132
oil creates the static load-carrying capacity of the journal bearing and contributes to its stiffness. But oil rotation also causes instabilities such as oil whirl and oil whip. In contrast, in squeeze film dampers the oil does not rotate if the rotor is not precessing. These dampers therefore have no static load-carrying capacity and no stiffness. The oil film dampens motion only when the rotor whirls. If there is no precession (and no retainer spring) the journal would bottom down in a squeeze film damper. Because the oil film is stationary, oil whirl and oil whip do not occur in squeeze film dampers.
6.9.2 SFD design configurations

A locally sealed damper with end hole feed/drain and radial piston rings is shown in Fig. 6.49, a. In such a damper, tightly sealed in the axial direction, the oil will flow circumferentially when squeezed.
Fig. 6.49 (from [6.15])
An open-ended damper, sometimes referred to as a globally sealed damper, with end groove feed/drain and axial face seals is shown in figure 6.49, b. Because the damper is not sealed in the axial direction, the oil can flow axially when squeezed.
133
A different configuration for feeding oil is the central circumferential feed groove; it is most commonly found in dampers without end seals. Other squeeze film damper designs are used with rolling element bearings, where the centering spring is generally eliminated, but antirotation tabs or keys are provided. Common end sealing arrangements include: a) radial O-ring seals; b) piston ring seals; and c) side O-ring seals.
6.9.3 Squeeze film stiffness and damping coefficients

The stiffness K sf and damping coefficient C sf of the squeeze film damper can be derived from a solution of the Reynolds equation for a nonrotating journal bearing. For a squeeze film with no end flow (Fig. 6.49, a), assuming laminar flow, full cavitation, a circular synchronous precession orbit and an inertialess oil film, the stiffness and damping coefficient are found to be [6.16] K sf =
3 2 + 2 1 2 Cr
24 R 3 L
)(
(6.63)
C sf =
3 2 + 2 1 2 Cr
12 R 3 L
)(
12
(6.64)
where R - squeeze film radius, L - axial length of damper, Cr - radial clearance, - oil dynamic viscosity, - eccentricity ratio (orbit radius / radial clearance), whirl speed. For relatively short dampers (L D < 0.5) with open ends, assuming unrestricted end flow (Fig. 6.49, b), the damper stiffness is K sf = 2 R L3
3 Cr 1 2
(6.65)
and the damping coefficient is C sf =
R L3
3 2 Cr 1 2
3 2
(6.66)
For the uncavitated film, K sf = 0 and
134
C sf =
R L3
3 Cr 1 2
3 2
(6.67)
For pure radial squeeze motion, K sf = 0 and for cavitated film
C sf =
R L3 cos 1 ( ) 2 2 + 1
3 Cr
(1 )
).
](
2 52
(6.68)
for uncavitated film C sf =
R L3 2 2 + 1
3 1 2 Cr
52
(6.69)
Note that the above stiffness and damping coefficients apply for a circular orbit of radius Cr . Dampers have no equilibrium position for an applied static load in the sense of a journal bearing. The "stiffness" developed by a damper results from its orbital motion due to a rotating load and could more properly be defined with a cross-coupled damping coefficient. For dampers operating at high speeds and high values of the squeeze-film Reynolds number Re =
Cr2 ,
(6.70)
where is the oil density, fluid inertia forces become significant. Inertia or "added mass" coefficients can be obtained dividing the fluid inertia forces to the radial acceleration 2Cr . The direct inertia coefficient is M sf =
R L3 1 (2 1) , 12 Cr 2
(6.71)
and the cross-coupled inertia coefficient is msf = where
R L3 27 1 1 , 2 + ln Cr 70 1+
(6.72)
= 1 2
12
(6.73)
135
6.9.4 Design of a squeeze film damper

If a squeeze film damper is added to an existing compressor design to cure a subsynchronous vibration problem, a stability analysis is carried out based on a finite element model or a transfer matrix model of the rotor-bearing system. Input parameters are damper radius, R, damper length, L, oil dynamic viscosity, , and rotational angular speed, . Usually, an aerodynamic crosscoupling stiffness is also considered. A practical highest level of cross-coupling for design studies is 1.75107 N/m [6.16]. Output parameters are the radial clearance, Cr , and the eccentricity ratio, 0 , which produce support stiffness and damping coefficients able to eliminate the instability. A computer program helps solving the damped eigenvalue problem and calculating the logarithmic decrement (or simply the real part of eigenvalues) as a function of design parameters. A typical stability map is shown in Fig. 6.50, where the real part of the eigenvalue of the unstable mode is plotted as a function of support damping. Each curve corresponds to a given value of support stiffness. Separate maps are plotted for several assumed values of the aerodynamic cross-coupling stiffness. The rotor and bearing characteristics remain unchanged.
Fig. 6.50 (from [6.16])
Curves crossing the zero level of eigenvalue real part define stiffnesses for which the system can be stable. As the support stiffness is increased, the system
136
becomes less stable, so that there is a threshold value K th above which the system is always stable (the curves do not intersect the zero real part value). On the other hand, for curves crossing the horizontal zero level axis, there is an intermediate range of support damping values for which the system is stable for a given value of support stiffness. For damping less than a certain value, c1, or larger than another value, c2, defined by the crossing points, the system is again unstable so that the actual value of support damping has to be selected between the two limits. Thus, linearized stability maps as that from Fig. 6.50 provide information on the support characteristics needed to promote stability in a given rotor-bearing system.
Fig. 6.51 (from [6.16])
Fig. 6.52 (from [6.16])
In order to further determine the damper design parameters, two more diagrams are needed illustrating the support stiffness (Fig. 6.51) and damping (Fig. 6.52) versus eccentricity, for different values of the radial clearance. First, the diagram from Fig. 6.51 is used. Drawing a horizontal line for the value K th , the values of 0 at the crossing points with curves Cr =const. are determined. Based on these values, the dotted line K th =const. is drawn in Fig. 6.52, which sets an upper border for the range of possible stable operation. Next, two horizontal lines are drawn in Fig. 6.52 corresponding to c1 and c2, the limits of variation of the support damping for stable operation. The area between these two
137
limits and below the K th curve defines the design region of the damper. The design parameters Cr and 0 are selected to correspond to points within this region.
6.10 Annular liquid seals

Plain (ungrooved) annular seals are joints with the inner cylinder (the rotor) rotating in a tight fluid annulus with respect to the outer cylinder (the stator). For a centered seal, there is an axial velocity due to the axial pressure gradient (leakage flow) and a circumferential velocity due to the shaft rotation and shaftfluid friction. In the "bulk-flow" model [6.17], these velocities are constant across the film, and shear stresses are accounted for only at the boundaries, at the shaft and the seal stator.
Fig. 6.53 (from [6.18])
138
For an eccentric seal, the difference between the pressure gradient on the tight and loose sides provides a restoring force, which opposes the shaft displacement, yielding a large direct stiffness. This is the Lomakin effect. For a whirling shaft, a tangential force will appear normal to the shaft radial displacement, yielding a cross-coupled stiffness. In the "short seal" solution, the pressure-induced tangential velocity is neglected. An improved treatment accounts for the development of circumferential flow due to the shear forces along the seal. In the following, the short seal solutions of Black and Childs are presented.
6.10.1 Hydrostatic reaction. Lomakin effect

The steady axial pressure distribution at any circumferential location within a plain non-serrated seal is shown in Fig. 6.53, a for a simple case without shaft rotation. The pressure drops at the seal entrance are shown in Fig. 6.53, b. The pressure at the inlet of the seal is ps and that at the seal exit is pe . As flow accelerates into the seal, a pressure drop occurs due to the Bernoulli effect. An additional decrease at the inlet region occurs due to the gradual growth of a turbulent flow velocity field into its fully developed form. The remainder of the pressure drop through the seal is very nearly linear. The seal differential pressure can be calculated using Yamada's leakage relationship [6.19] for flow between concentric rotating cylinders
p = ( 1 + + 2 )
V 2
2
(6.74)
where is the entrance loss factor, is the fluid density, V is the average fluid velocity, and is a friction loss factor defined by
L
Cr
(6.75)
In the above, L is the seal length, Cr is the radial clearance, and is a friction factor, defined by Yamada to be the following function of the axial and circumferential Reynolds numbers, Ra , Rc :
= 0.079 Ra
1 4
7R 2 8 1+ c , 8R a
(6.76, a)
139
2 V Cr
Ra =
, Rc =
R Cr
(6.76, b)
where is the fluid kinematic viscosity, R is the seal radius and is the rotor angular speed, rad/sec. The friction law definition of equation (6.75) fits the Blasius equation for pipe friction [6.17]. If the shaft moves downward within the seal, the velocity at the top of the shaft, Vtop , increases, because of the larger gap through which the fluid flows. The flow velocity at the bottom, Vbot , decreases because of the smaller gap height there. Thus, pressure drops at the seal entrance are flow-velocity dependent and differ from top to bottom as shown in Fig 6.53, b. At the top, the entrance pressure drop increases. At the bottom, the entrance pressure drop decreases. Therefore, a net increase in pressure at the bottom of the seal produces an upward force tending to restore the shaft to its original centered position. A similar situation occurs when the lateral velocity of the shaft changes. The combined effect of the inlet loss and axial-pressure gradient accounts for the large direct stiffness which can be developed by annular seals. This physical mechanism for the direct stiffness was first explained by Lomakin (1958). The actual pressure distributions shown in Figs 6.53, a and b are modified by shaft rotation and geometric variations within the seal, but the underlying principles governing the pressure buildup within a seal remain the same. The Cr R ratio for seals is generally on the order of 0.003 versus 0.001 for bearings. These enlarged clearances, combined with a high pressure drop across the seal, mean that flow is highly turbulent. Most analyses for pump seals have used "bulk-flow" models for the fluid within the seal. In such models, the variation of fluid velocity components across the clearance is neglected. Average (across the clearance) velocity components are used hence the bulk-flow designation. Bulk-flow models neglect the shear stress variation for the fluid within the clearance and only account for shear stresses at the boundaries of the model [6.14].
6.10.2 Rotordynamic coefficients

Dynamic seal forces are normally represented by a reaction force / seal motion model of the form [6.17]: Fy K = Fz k
& k y C c y +M + & K z c C z & y & , & z &
(6.77)
140
where y and z are fixed coordinate displacements of the shaft and the dots are derivatives with respect to time. This model is valid for small motion about a centered position. Short annular seals are usually considered isotropic. The diagonal terms of their stiffness and damping matrices are equal, while the off-diagonal terms are equal, but with reversed sign. The cross-coupled damping and the mass term arise primarily from inertial effects. Short seal solutions are obtained by neglecting the pressure-induced circumferential flow, while including the shear-induced flow.
Cross-coupled stiffnesses arise from fluid rotation, in the same way as in uncavitated plain journal bearings. For example, moving the seal rotor horizontally rightwards, a converging section is created in the lower half and a diverging section in the upper half of the seal (shaft rotates counterclockwise). The pressure is elevated in the converging section and depressed in the diverging section, yielding a reaction force upwards. This force acts normal to the radial displacement, having a destabilizing effect.
For long annular clearance seals, like those used to break down large pressure differences in multi-stage pumps, angular dynamic coefficients are introduced, because the rotor is acted upon by couples, and forces give rise to tilting shaft motions, and moments produce linear displacements. The dynamic coefficients of equation (6.77) can be expressed in the following form:
Stiffness coefficients: K = 3
2 2T 2 4 ,
k = 3 1 T / 2 ;
(6.78)
Damping coefficients: C = 3 1 T , Inertia coefficients:
c = 3 2 T 2 ;
(6.79)
M = 3 2 T 2 ;
where
(6.80)
3 =
Rp
(6.81)
and the mean passage time is

T= L . V
(6.82)
141
The quantities 0 , 1 and 2 depend on the friction loss coefficient , and the entry loss coefficient . Their formulae are given below for three slightly different annular seal models.
1. Black's model [6.20], [6.18]
0 = 1 =
2 =
( 1 + ) 2 ( 1 + + 2 ) 2
'
( 1 + ) 2 + ( 1 + )(2.33 + 2 ) 2 + 3.33 ( 1 + ) 3 + 1.33 4 , ( 1 + + 2 )3

0.33 ( 1 + ) 2(2 1) + ( 1 +
(6.83)
)( 1 + 2 ) 2 + 2 ( 1 + ) 3 + 1.33 4 . ( 1 + + 2 ) 4
For finite-length seals, Jensen developed the following formulae to account for finite L R ratios:
2 L 0 = 0 1 + 0.28 R 1
2 L 1 = 1 1 + 0.23 R 1
2 L 2 = 2 1 + 0.06 R
where 0 , 1 and 2 are given by (6.83), and the "bar" quantities have to be used in equations (6.78) to (6.80) instead of those without bar over letter.
2. Jensen's model [6.21]
In a further development, the coefficients 0 , 1 and 2 were defined in terms of the following additional parameter 7 Rc = 8R a
2
7R 1 + c 8R a
(6.84)
which accounts for a circumferential variation in due to a radial displacement perturbation from a centered position. They can be calculated as
0 =
1.25 A 2 , BD
142
1 =
A2 + 2 A C + 2 + E + F 6 3 8 6 6
1 5 B D2
(6.85)
1 1 1 1 2 A + A ( 1.25 + C ) 2 + E + F 6 2 6 3 2 = , 3 BD where
A =1+ , B = 1 + + 2 , E = A C2 3 , C = 1.75 0.75 , F = C3 4 .
D = A+C ,
Fig. 6.54 (from [6.21])
143
Plots of 0 , 1 and 2 from equations (6.85) are provided in Fig. 6.54 as a function of and , for = 0.5 . These coefficients are comparatively insensible to anticipated variations of the entrance loss factor . Black also examined the influence of inlet swirl on seal coefficients. In previous analyses, a fluid element entering a seal was assumed to instantaneously achieve the half-speed tangential velocity R 2 . In a later paper he demonstrates that a fluid element must travel a substantial distance axially along the seal before asymptotically approaching this limiting velocity. The practical consequence of this swirl effect is that predictions of the cross-coupling terms, k and c, may be substantially reduced. This applies especially for interstage seals, in which the inlet tangential velocity is negligible.
3. Childs' model [6.22]
Childs completed a seal analysis based on Hirs' turbulent lubrication model [6.23]. In his analysis he assumed completely developed turbulent flow in both the circumferential and axial directions. The short seal dynamic coefficients derived by Childs are given by equations (6.78)-(6.80) but the coefficients 0 , 1 and 2 are
0 = 1 =
2 =
where
E=
2 2 E ( 1 m0 ) , 1 + + 2 2 2 1 + + 2 E B 1 + 2 6 + E , (6.86)
1 + E , 1 + + 2 6
B = 1 + 4 b 2 ( 1 + m0 ) , b= Ra V = , 2 Rc R
1+ , 2 ( 1 + + B ) 1 1 + 4b
2
(6.87)
and m0 is an exponent entering the loss factor expression (mo 0.75)
n0 Ram0 1 +
1 4b 2
1+ m0 2
(6.88)
144
The short seal dynamic coefficients derived by Childs are similar to the previously mentioned solution by Black when a one-half shaft speed swirl velocity is assumed. Childs considered the influence of the initial swirl V0 . In this case, the seal dynamic coefficients are written in the following form:
Stiffness coefficients:
2T 2 ( 2 + a1 ) , K = 3 0 4
Damping coefficients:
k = 3 ( 1 + a2 )
T ; 2
(6.89)
C = 3 1 T ,
Inertia coefficients:
c = 3 ( 2 + a3 ) T 2 ;
(6.90)
M = 3 2 T 2 ,
(6.91)
where
a = [ 1 + (1 + m0 )] ,
a1 = a2 =
(6.92)
2V0 2 2 a 1 + + 2 2V0 2 2 a 1 + + 2 2V0 a 1 + + 2
1 1 1 a a E + 2 1 e 2 + a e , a 1 1 1 B a EB + 1 e E + + 1 , 2 a a 1 e a 1 1 a + E + + 2 1 e 2 2 a a
a3 =
For V0 =0, a1 = a2 = a3 =0, and one obtains equations (6.78)-(6.80). From equation (6.77), a circular shaft orbit of radius A and precession frequency yields radial and circumferential force components
Fr = K + c M 2 A ,
F = ( k C ) A .
The circumferential force can produce unstable operating regimes.
Example 6.2
An annular neck ring seal has the following geometric properties: length L =50 mm, radius R =75 mm (L D = 1 3) , radial clearance Cr =0.25 mm.
145
At operating conditions N =1200 rpm, the pressure difference across the axial length of the seal is p =1.38106 Pa. The water properties are: dynamic viscosity =4.1410-4 Ns/m2, mass density =979 kg/m3. Assume initially that the flow velocity is V =25 m/s. The initial axial Reynolds number is
Ra =
2 V Cr
= 29,559 .
The circumferential Reynolds number is

Rc =
R Cr = 5 572 .
3
Solving for gives

2 8 L 0.079 7 Rc = 1.217. 1 + = 1/ 4 8Ra Cr Ra
Considering =0.1, from the equation for p a new value of V is obtained

2 p / V = 1 + + 2 2 =26.77 m/s.
1
The process is repeated iteratively until the mean flow velocity is obtained as V = 28.592 m/s for which = 1.174 . The final value of the axial Reynolds number is Ra = 33,807 showing that the flow is fully turbulent. Substituting =0.1 and =1.193 into the equations (6.83) gives
0 =0.1414,
and also
1 =0.35,
2 =0.05287,
3 =5.6107.
The mean passage time is
T= L =1.74910-3 s. V
The dynamic coefficients (6.78)-(6.80) have the following values

K =7.1139106 N/m, k =2.0604106 N/m,
146
C =32793 Ns/m,
Note that
c =1153 Ns/m,
M =9.176 Ns2/m.
M =1, so that M and c tend to cancel for synchronous c precession at the running speed and the radial reaction force component Fr is given only by the direct stiffness K.
k =0.5 corresponding to the average C tangential velocity within the seal R 2 . The result is the same as for a plain journal bearing. For = n , the rotor's first critical speed, the tangential reaction
The whirl frequency ratio is w =
force component F becomes destabilizing at the speed th =
= 2n . This 0.5 shows that, theoretically, a seal becomes destabilizing at running speeds in excess of about twice the rotor's first critical speed. Because in seals w can be decreased, the on-set speed of instability th is increased.
6.10.3 Final remarks on seals

The above seal analyses are restricted to a) plain seal surfaces; b) constant inlet loss coefficients, and c) fully developed inlet circumferential flow velocity. The axial flow velocity is assumed to be much greater than the velocity in the circumferential direction. The predominant effect of variation in the inlet swirl velocity is to alter the cross-coupled stiffness coefficients. These coefficients are reduced in magnitude as the swirl velocity is reduced. In fact, under proper values of seal geometry, axial pressure drops and inlet swirl velocity, the signs of the cross-coupled stiffness coefficients are reversed; cross-coupling promotes stability rather than instability. Experiments with spiral serrated seals, for axial Reynolds numbers ranging from 20,000 up to 40,000, revealed that long serrated seals possess negative direct stiffness. In calculations, correction factors on the seal length in a plain seal have to be used to account for grooving in serrated seals. Dynamic coefficients in equation (6.77) are almost skew-symmetric, even for volute pumps, although such results would be expected only for rotationally symmetric configurations. In contrast to bearings, seals develop significant direct stiffness in the centered position, independent of the rotation speed, from a hydrostatic mechanism. Damping coefficients of seals are relatively large. As a result, the majority of centrifugal pumps operate with no evidence of rotor dynamic problems. For close clearance rings, hydrodynamic mass effects have been seen to be strong.
147
For synchronous rotor excitation and whirling, it was shown that the seal restoring force increases with the square of the speed. It has been thought of as being generated by a fictitious mass, called the Lomakin mass. However, a geometric mass correction is correct only for pure synchronous whirling. When the virtual mass is greater than the actual geometric mass, the critical speed is suppressed. Such an approach would be correct only for systems in which the Lomakin mass and the geometric mass occur at the same locations. For systems in which this does not occur, the Lomakin effect should be treated as a stiffness and hence as a function of speed.
6.11 Annular gas seals

Annular gas seals, or labyrinth seals, have a leakage-control role in centrifugal compressors. Forces developed by gas seals are roughly proportional to the pressure drop across the seals and the fluid density within the seal. Because of the density dependence, gas seals have had a great impact on high-pressure compressors for gases with high molecular weight.
Fig. 6.55 (from [6.14])
Several usual forms of labyrinths are shown in Fig. 6.55. Labyrinths can have a negative effect on the smooth running of compressor rotors. Two basically different mechanisms are possible. 1. The first case occurs if unbalance induced precession causes contact between labyrinth tabs and the shaft. Wherever a possible contact of the labyrinth
148
leads to a marked heating and hence deformation of the rotor, it is advantageous to have the labyrinth strips located on the shaft. The cooling induced by the strips protects the shaft against a local rise in temperature. 2. The second case is less apparent but more important. Any eccentricity causes a change of flow in the labyrinth and induces reaction forces that act back on the shaft and thus may diminish the rotor system damping. If this damping is sufficiently reduced, self-induced rotor vibrations occur that render the operation of the compressor impossible.
Fig. 6.56 (from [6.17])
Figure 6.56 illustrates a typical sealing arrangement for a last stage compressor impeller. The eye-packing seal limits return-flow leakages down the front of the impeller. The shaft seal restricts leakage along the shaft to the preceding stage. The eye packing and shaft seal configurations are called "seethrough" or half-labyrinth seals (see Fig. 6.55). In a throughflow compressor, leakage flow through the balance drum is returned to the inlet. The balance drum is a long seal which absorbs the full pressure differential of the compressor. The balance drum seal configuration is called an interlocking or full labyrinth (see Fig. 6.55). For a back-to-back compressor, the balance drum absorbs the pressure drop between the last stage of the compressor and the last stage of the initial series of impellers, i.e., about one half of pressure drop per machine. For the same inlet and discharge pressures, the average density is higher in the center labyrinth of a back-to-back machine than in the balance drum labyrinth of a series machine. Back-to-back compressors are more sensitive to the forces from the central labyrinth than are series machines to forces from the balance-drum labyrinth, and
149
this is mainly because of the larger deflections at the middle span for the first critical mode. The forces developed by compressor seal labyrinths are at least one order of magnitude lower than their liquid seal counterparts. They have negligible addedmass terms and are typically modeled by the following reaction force/motion model:
Fy K = Fz k
& k y C c y + . & K z c C z
(6.93)
The diagonal terms of their stiffness and damping matrices are equal, while the off-diagonal terms are equal, but with reversed sign, which denotes isotropy. Unlike the pump seal model of equation (6.77), the direct stiffness term is typically negligible and is negative in many cases. The coefficients in equation (6.93) are dependent on the gas density, giving rise to "load dependent" instabilities. This leads to an "onset power level of instability" instead of the known "onset speed of instability" encountered for rotors in hydrodynamic bearings, which exhibit speed-dependent instabilities.
Fig. 6.57 (from [6.9])
Tests have shown that for "see-through" labyrinth seals: a) cross-coupled stiffness coefficients increase directly with increasing inlet tangential velocities; b) seal damping is small but must be accounted for to obtain reasonable rotordynamic predictions; and c) as clearances decrease, teeth-on-rotor seals become less stable and teeth-on-stator seals become more stable. Stabilizing solutions include the swirl brakes or swirl webs at the entrance to labyrinth seals (Fig. 6.57). This device uses an array of axial vanes at the seal inlet to destroy or reduce the inlet tangential velocity. Another approach for stabilizing compressors involves a "shunt line" and entails rerouting gas from the compressor discharge and injection flow into one of
150
the early cavities in the balance-piston labyrinth (Fig. 6.58). Usually, the flow from the back side of the last stage impeller enters the labyrinth with a high tangential velocity which yields a high destabilizing force. The shunt flow forces the gas outward along the back face of the impeller, reducing the seal tangential velocity.
Fig. 6.58 (from [6.9])
Of all the labyrinth seals, the destabilizing forces are maximum for the balance piston labyrinth because it is the longest seal and has the highest pressure drop. For a throughflow machine it also has the highest gas density. Some success has been obtained in improving the stability of compressors by replacing labyrinth seals with honeycomb seals. The latter develop larger destabilizing coefficients but much higher direct damping values. Brush seals, using wires in contact with a ceramic coating on the shaft, have a sharply reduced leakage flow as compared to a labyrinth or honeycomb seal.
6.12 Floating contact seals

Oil seals are frequently used in high speed multistage compressors to minimize the leakage of the process gas into the atmosphere. For hazardous gases, tightness is an important requirement. This is achieved with liquid seals in conjunction with floating ring seals.
151
Fig. 6.59 (from [6.9])
6.12.1 Design characteristics

If the desired pressure drop across the seal ring is adequate, only one seal ring is employed. Otherwise, a double floating ring combination is used, with sealing oil feed in the middle, as illustrated in Fig. 6.59. The seal segments consist of outer and inner seals which are springloaded against each other. The preload of the spring causes the lapped external faces of the seal segments to be in contact with the seal cartridge housing. An antirotation pin is provided to prevent rotation of the seal segments. Oil is introduced between the rings at a pressure slightly above compressor suction pressure and then leaks axially along the shaft. The inner ring seals against process gas leakage, while the outer ring limits seal oil leakage to the outside. Most of the oil is recovered. Heat dissipation is critical to the operational security of oil seals. This is ensured by the liquid flow in the relatively wide seal ring clearances, typically 50 to 80 m, which entails high leak rates. These can be lowered providing the floating ring with a return thread.
152
Fig. 6.60 (from [6.24])
A single ring seal arrangement, used in a 5-stage compressor for refinery service, is depicted in Fig. 6.60, where: 1 oil bushing, 2 diaphragm, 3 sleeve (impeller), and 4 stator.
Fig. 6.61 (from [6.24])
153
The "trapped bushing seal", shown in Fig. 6.61, is a "dual" bushing which encompasses both the inner and outer seal in one ring. One can distinguish: 1 shaft, 2 - sleeve with interference fit under bushing, 3 stator, 4 stepped dual bushing, 5 bushing cage, 6 nut, 7 shear ring, 8 spacer ring. The compact design allows shorter bearing spans for higher critical speeds. The sleeve 2 under bushing protects the shaft and simplifies assembly and disassembly. It requires only a jack/puller bolt ring. The spacer ring 8 is fitted at initial assembly, so there is no field fitting of parts. This arrangement allows the bushing to freely track the shaft motion. Seal induced hydrodynamic forces are dissipated in seal motion and not applied to the shaft. The seal is also insensitive to wear on the axial faces [6.24].
a
Fig. 6.62 (from [6.25])
Figure 6.62 shows a typical cone seal arrangement. Fig. 6.62, a is the original design which experienced severe vibrations, and Fig. 6.62, b is the improved design, obtained by reducing the sealing length and the width of the lapped face, and increasing the steady holding force of the springs. An arrangement intended to solve the seal centering problem is shown in Fig. 6.63, where a tilting-pad bearing element is incorporated in the high-pressure seal ring. Possible design improvements to the high-pressure ring are: a) decreasing the seal face area, in order to decrease the unbalanced axial force, hence the friction force; b) cutting circumferential grooves in the seal land region, to reduce the effective axial hydrodynamic length of the ring; and c) increasing the seal clearance.
154
Fig. 6.63 (from [6.9])
6.12.2 Seal ring lockup

Compressors with oil seal rings are often subject to subsynchronous vibrations. The seal rings are designed to "float", in the sense that they orbit with the shaft as it vibrates (rotation is prevented by some pin or other device which still allows radial ring motion). If the seal ring orbits with the shaft, the oil film force, and thus the dynamic coefficients, between the shaft and ring are very small. In certain circumstances, the seal rings may be prevented from moving. This is known as the "seal ring lockup". It causes large oil film forces to act between the shaft and ring, which may produce large stiffness and damping coefficients. The building up of the cross-coupled stiffnesses by the seal ring lockup, in a machine with tilting pad bearings, can be described as follows [6.26]: 1. For a non-rotating shaft, the oil ring is resting on the top of the shaft. As the rotational speed increases, the oil seal fluid film supports the weight of the oil seal ring. 2. The pressure drop across the seal ring increases with the machine speed. This in turn increases the axial force due to unbalanced pressure, and correspondingly, the radial friction force acting on the seal ring (given by the coefficient of friction times the axial force). When the radial friction force acting on the seal exceeds the oil film radial force, the seal will lock up. 3. As the shaft speed continues to increase, the (tilting pad) bearing oil film forces tend to force the shaft upward in the bearing (the bearings are unloaded). The seal ring oil film forces tend to oppose this because the shaft is now taking an eccentric position within the fixed seal ring.
155
4. Unbalance forces inherent in rotor make the shaft undergo dynamic orbits. These dynamic forces in turn produce dynamic oil film forces in the lubricant between the shaft and fixed seal ring. If the combination of static and dynamic forces is larger than the radial friction force, the oil seal ring will move. If the dynamic forces are not large enough to overcome the available radial friction force at a given speed, the seal ring is likely to lockup. This can happen at some speed well below running speed. 5. As the shaft speed increases, the oil film forces in the bearings will lift the shaft in the bearing. This will create an eccentric position of the shaft relative to the locked up oil seal ring. This in turn will produce large cross-coupling stiffness terms acting on the shaft due to the oil film in the seal ring.
6.12.3 Locked-up oil seal rotordynamic coefficients

In approximate calculations, it is assumed that the seal ring is locked up with the shaft centered and behaves as a non-cavitated concentric plain sleeve bearing, and there is no axial pressure gradient. Oil seal rings are short (0.05<L/D<0.2) and their flow is generally laminar, non-cavitating. For this centric case, the cross-coupled stiffnesses are [6.24]
k yz = k zy =
3
D C = 2 L C , 4 r r D
R L 2
(6.94)
where - shaft angular speed of rotation, rad/sec, D shaft diameter, R shaft radius, L axial seal length, Cr - oil seal ring radial clearance, - fluid dynamic viscosity. As coefficients are proportional to L3 , a considerable reduction in destabilization can be achieved by providing the seal with circumferential grooves, as shown in Fig. 6.63. For a concentric seal, the direct stiffnesses k yy and k zz are negligible. The direct damping coefficients are
c yy = 2 2
k yz ,
c zz =
k yz .
(6.95)
3 , hence highly Note that these properties are inversely proportional to C r sensitive to variations in clearance. Thus, enlarged clearances at the seals would allow operation with low vibrations even when the seals are locked up.
The eccentric values of cross-coupled terms are known to be several times the value for the centered position.
156
There are more accurate calculation schemes to determine the seal dynamic coefficients, for various assumptions about the seal lockup eccentricity [6.17]. However, the results are sensitive to the assumptions about whether lockup results in increased or decreased journal loading in rotor bearings. Stable operation of a rotor is possible even if the seal locks up and the seal clearances are abnormally large. Due to eccentric lock-up of the seal ring, direct stiffness terms are developed. Reduced journal loading will raise the natural frequencies of the rotor because the effective bearing span is reduced transferring the load to the seals. In this case, the frequency of the non-synchronous instability is higher than the calculated rigid bearing critical speed. A quasi-static method of determining the seal ring lockup conditions has been used with good results [6.26]. It involves matching the radial friction force to the seal ring weight, wherefrom the ring initial lockup speed is calculated first. As the shaft speed increases, it lifts off of the bottom pad of the tilting pad bearings. Based on known relationships for tilting pad bearings, the shaft position relative to the bearing center line is determined, wherefrom the oil seal ring eccentricity and attitude angle can be calculated for a load capacity equal to the oil seal ring weight. Otherwise, because the static attitude angle depends on the friction force, it is, strictly speaking, indeterminate. Selection of the friction coefficient is a difficult task. The subsequent calculation of rotordynamic coefficients can be done using the Ocvirk short-bearing model [6.17]. The relationship of the static eccentricity ratio 0 to the modified Sommerfeld number
Ss =
2 L D N R L W* = W 60 W Cr D 2
(6.96)
for a short bearing is illustrated in Fig. 6.64, a. Figure 6.64, b shows the rotordynamic coefficients for a short seal with an assumed attitude angle of 90o [6.9]. The seal has no direct stiffness for this attitude angle, and the crosscoupled stiffness and direct damping terms rise sharply with decreasing S s , i.e. increasing static eccentricity ratio. The increase of c yy and c zz with 0 can be advantageous at low speeds, when the rotor is traversing a critical speed; however, the increase in k yz and k zy can lead to rotor instability at running speeds of the order of twice the critical speed. For five values of the static eccentricity ratio 0 , the dimensionless oil seal coefficients are given in Table 6.10 [6.27]. The physical (dimensional) coefficients are then calculated as
157
W* W* , c i, j = C i, j , Cr Cr
k i, j = K i, j
(i , j = y , z )
2 2
(6.97)
where
W* =
R L3 N
60 Cr2
R = L D C r
L D
(6.98)
and
= 2 N 60 .
a
Fig. 6.64 (from [6.9])
Table 6.10
0
0.01 0.1 0.2 0.5 0.8
K yz
3.14 3.28 3.76 9.67 92.1
K zy
-3.14 -3.19 -3.33 -4.84 -14.54
C yy
6.28 6.38 6.67 9.67 29.1
C zz
6.28 6.57 7.51 19.35 184.2
Note that K yz K zy and C yy C zz so that the locked-up eccentric seal is anisotropic.
158
The direct stiffness and damping coefficients are not as important as the cross-coupled coefficients because the seals are close to the bearings. Bearing direct stiffness and damping coefficients are much larger and tend to dominate the seal coefficients. Tilting pad bearings have no cross-coupled coefficients when symmetric loading occurs, so the oil seal cross-coupled coefficients are very significant. Dt p. A simple breakdown seal can have an axial force as high as 2 For a seal pressure drop p = 6.89106 Pa, a diameter D=165 mm, and a face width t = 6.3510-3 m, the axial locking force is approximately 1.11104 N. For three seals and assuming a coefficient of friction of 0.1, the radial load capacity could be as high as 3.34103 N, a value equal to a typical bearing load. It is easy to see that, unless a special pressure-balanced ring seal is employed in medium-weight rotor, high-pressure machines, the bearings and seals will interact as rotor supports. In heavy rotor, lower pressure machines this interaction is less likely. Thus, the high pressure can indirectly cause instability in otherwise stable machines, apart from influencing the gas density, by locking oil breakdown seals axially such as to cause interaction with the stable rotor / tilting pad bearing system. In conclusion, oil-buffered seals should be designed to act as sealing elements only and should not be part of the dynamic system of rotor and bearings. A pressure-balanced seal should be a design requirement, and means for causing seal lock-up should be eliminated from the compressor design.
References
6.1 Someya T. (ed), Journal-Bearing Databook, Springer, Berlin, 1988. 6.2 Constantinescu, V. N., Nica, Al., Pascovici, M. D., Ceptureanu, Gh., Nedelcu, t., Lagre cu alunecare, Editura tehnic, Bucureti, 1980. 6.3 Rieger, N. F., Vibrations of Rotating Machinery, Vibration Institute, Illinois, 1977. 6.4 Lund, J. W., Rotor-Bearing Dynamics, Ossolineum, 1979. 6.5 Lee, C.-W., Vibration Analysis of Rotors, Kluwer Acad. Publ., Dordrecht, 1993. 6.6 Childs, D., Moes, H., and van Leeuwen, H., Journal bearing impedance descriptions for rotordynamic applications, Journal of Lubrication Technology, pp.198-219, 1977.
159
6.7 Adams, M. L., Insights into linearized rotor dynamics, Part 2, Journal of Sound and Vibration, vol.112, nr.1, pp.97-110, 1987. 6.8 6.9 Cameron, A., Basic Lubrication Theory, 3rd ed., Ellis Horwood Ltd., Chichester, 1981. Ehrich F.F. (ed), Handbook of Rotordynamics, Mc Graw Hill, New York, 1992.
6.10 Bhushan, G., Rattan, S. S., and Mehta, N. P., Effect of pressure dams and relief-tracks on the performance of a four-lobe bearing, IE (I) Journal-MC, vol.85, pp.194-198, 2005. 6.11 Chen, W. J. and Gunter, E. J., Introduction to Dynamics of Rotor-Bearing Systems, Trafford, Victoria, 2005. 6.12 Lund, J. W., Spring and damping coefficients for the tilting-pad journal bearing, ASLE Transactions, vol.7, no.4, pp.342-352, 1964. 6.13 Hamrock, B. J., and Anderson, W. J., Incompressibly lubricated Rayleigh step journal bearing, NASA TN D-4873, 1968. 6.14 Gasch, R., Nordmann, R., and Pftzner, H., Rotordynamik, 2. Auflage, Springer, Berlin, 2002.
6.15 El-Shafei A., Squeeze film dampers: effective damping devices for rotating machinery, Vibrations, vol.5, nr.3, pp.8-11, 1989. 6.16 Gunter E. J., Barrett L. E., and Allaire P. E., Stabilization of turbomachinery with squeeze film dampers Theory and applications, Proc. Conf. "Vibrations in Rotating Machinery", I. Mech. Eng., pp.291-300, 1976. 6.17 Childs D., Turbomachinery Rotordynamics: Phenomena, Modelling and Analysis, Wiley, 1993.
6.18 Barrett L. E., Turbulent flow annular pump seals: A literature review, Shock and Vibration Digest, vol.16, nr.2, pp.3-13, 1984. 6.19 Yamada Y., Resistance of flow through an annulus with an inner rotating cylinder, Bull. J.S.M.E., vol.5, nr.18, pp.302-310, 1962. 6.20 Black H. F., Effects of hydraulic forces in annular pressure seals on the vibrations of centrifugal pump rotors, J. Mech. Eng. Sci., vol.11, nr.2, pp.206-213, 1969. 6.21 Black H. F. and Jensen D. N., Dynamic hybrid properties of annular pressure seals, J. Mech. Eng. Sci., Proc. Inst. Mech. Eng., vol.184, pp.92-100, 1970. 6.22 Childs D. W., Dynamic Analysis of Turbulent Annular Seals Based on Hirs' Lubrication Equation, J. of Lubrication Technology, Trans. ASME, vol.105, pp.437-444, 1983.
160
6.23 Hirs G. G., A Bulk-Flow Theory for Turbulence in Lubricant Films, J. of Lubrication Technology, Trans. ASME, pp.137-146, 1973. 6.24 Emerick M. F., Vibration and Destabilizing Effects of Floating Ring Seals in Compressors, NASA CP 2250, pp.187-204, 1982. 6.25 Kirk R. G. and Simpson M., Full Load Shop Testing of 18000 HP Gas Turbine Driven Centrifugal Compressor for Offshore Platform Service: Evaluation of Rotor Dynamic Performance, NASA CP 2409, pp.1-13, 1985. 6.26 Allaire P. E. and Kocur J. A. Jr., Oil Seal Effects and Synchronous Vibrations in High-Speed Compressors, NASA CP 2409, pp.205-223, 1985. 6.27 Kirk R. G., The Impact of Rotor Dynamics Analysis on Advanced Turbo Compressor Design, Proc. Conf. "Vibrations in Rotating Machinery", I. Mech. Eng., University of Cambridge, 15-17 Sept. 1976, pp.139-150, 1977.
7.
INSTABILITY OF ROTORS
Under certain operational conditions, the rotor precession orbit grows suddenly, accompanied by a change in the whirl frequency to a nonsynchronous value. With rare exceptions, this frequency is that of the lowest natural frequency of the rotor system, i.e. the frequency of the first critical speed. In most cases it proves impossible to pass through this threshold of instability without endangering the mechanical integrity of the machine. Limiting the analysis to the stability of the static equilibrium position of the rotor axis, the system can be assumed to be linear. At the threshold of instability, called the onset speed of instability, the real part of an eigenvalue becomes zero. Nonlinearities have little or no influence on this speed limit. The linear model contains the governing parameters which explain the physical mechanism of the instability. Only when it is of interest to explore the behavior beyond the onset speed of instability does it become necessary to include nonlinearities, which is beyond the aim of this presentation. This chapter is concerned with instabilities of the flexural precession motion produced by hydrodynamic forces in bearings and seals, interaction of bladed disks and impellers with fluid film forces, dry friction, internal shaft damping and rotor asymmetry. Some of them are self-excited instabilities, in which the whirling is always at a subsynchronous frequency equal to an eigenfrequency of the system. Others are produced by parametric excitation and can be synchronous, sub- or supersynchronous. They are more like forced vibrations and some can even be driven through the speed range of instability (i.e., a higher speed can be found where the rotor regains its stability).
7.1 Whirling of rotating shafts

In the case of whirling, rotors are destabilized by tangential reaction forces which are normal to a radial displacement and in the direction of shaft rotation. The destabilizing forces are in the direction of the whirling motion, and in opposition to
162
the external damping force, which tends to inhibit the whirling motion. Often these forces are approximated as being proportional to the radial deflection (Fig. 7.1). The proportionality factor is a cross-coupled stiffness coefficient, since it relates the force magnitude to a deflection normal to the force. In matrix form
0 fy = fz k yz
k yz y , 0 z
(7.1)
where the cross-coupling terms are skew-symmetric k zy = k yz . (7.2)
As shown in Section 6.6, as the tangential force fT acts in the direction of the whirling, it supplies energy to the motion, hence provides negative damping.
Fig. 7.1
If the free motion is decomposed into two whirls, each in an Archimedean spiral (Fig. 5.26), one with the direction of rotation and the other against it, at speeds below the onset speed of instability both spirals are of decreasing amplitude. At speeds above the onset speed of instability, in almost all cases, the whirl in the direction of rotation is of exponentially increasing amplitude (hence unstable) while that in opposite direction is of decreasing amplitude. It is the forward whirl component which is responsible for the negative damping. For an idealized rotor with a single lumped mass m and shaft stiffness k, whirling with a speed at an instantaneous radius r, the forces acting on the mass are (Fig. 7.2) the centrifugal force m r 2 , the elastic restoring force k r , the radial damping force c (d r dt ) , the radial inertia force m d 2 r dt 2 , the tangential damping force c r , the Coriolis force 2m (d r dt ) and the destabilizing force fT . The equations of motion are written using dAlemberts principle [7.1].
The radial force balance yields
7. INSTABILITY OF ROTORS
163
d 2r dr +c + k r m r 2 = 0 , 2 dt dt
(7.3)
and the tangential force balance gives

2m dr + c r fT = 0 , dt (7.4)
where the destabilizing force is proportional to the radial deflection fT = k yz r , and k yz is the cross-coupled stiffness. (7.5)
Fig. 7.2
The solution takes the form r = r0 e t . For the system to be stable, the coefficient of the exponent k yz c 2m (7.6)
(7.7)
must be negative, giving the requirement for stable operation as k yz < c or, in dimensionless form, (7.8)
164
k yz m
2
<
c = 2 m
(7.9)
where is the damping ratio. At the onset of instability = 0 , so that the whirl speed at onset is found, from equation (7.3), to be
k . m
(7.10)
The whirling speed at the onset of instability is the rotor natural frequency, irrespective of the shaft rotational speed. Even if the rotational speed increases, the whirl speed remains equal to the rotor natural frequency so that the whirling is a subsynchronous motion. Beyond the onset speed of instability, > 0 and the solution (7.6) describes an exponential spiral, hence the name of rotor whirling. Actually, nonlinearities in the rotor system dissipate energy more rapidly with increasing amplitudes than predicted by a linear model. Hence, whirl amplitudes increase sharply with time but are generally finite, achieving a steady-state limit cycle with a closed orbit. Equation (7.9) shows that, increasing the stator system damping ratio, the onset speed of instability is pushed to a higher speed. Rotor stability is enhanced by increasing the external damping, or equivalently, by introduction of anisotropy into the bearing supports (see Section 3.1.3). The sensitivity of rotors to destabilizing forces increases with rotor flexibility. Avoidance can be achieved by increasing the system stiffness (and critical speeds).
7.2 Instability due to rotating damping

The instability due to internal rotor friction is analyzed in Sections 2.3.2, 2.3.3 and 3.1.3 (Part I) for a single-disk rotor. The threshold speed marking the transition from stability to instability is given for viscous damping and is shown to be raised by anisotropic support stiffness. In 1980 appeared the remarkable paper of Stephen Crandall [7.2] giving a physical explanation of the fact that a mechanism of energy dissipation can cause instability. Based on a simple two-dimensional model, Crandalls paper gives a clear insight into the destabilizing mechanism of damping in rotating parts and permits an elementary physical determination of the stability limit. The following presentation reproduces most of the content of that paper.
165
7.2.1 Planar rotor model
Consider the planar model of a disk mounted on a flexible shaft (Fig. 7.3). The rigid ring is forced to turn at an angular speed by an external source. A mass particle is suspended elastically from the ring by four massless linear springs. The mass m represents the central disk of the Laval-Jeffcott model and the springs represent the massless flexible shaft of stiffness k, independent of the rotational speed. Gravity in the plane of the figure is neglected. The elastic force system for small displacements is circularly isotropic. The equilibrium position is with the mass at the origin. When the mass has a radial displacement r, in any direction within the plane, the elastic restoring force is k r , directed back toward the origin.
Fig. 7.3 (from [7.2])
Fig. 7.4 (from [7.2])
The motion is conveniently expressed in terms of either stationary coordinates y, z , or rotating ring-fixed coordinates , , as in Fig. 2.14. In stationary coordinates, the equations of motion for the mass particle have the form (2.1). They are decoupled and independent of . All natural motions are linear combinations of two independent modes which have the same natural frequency n = k m . The two independent modes may be taken as rectilinear oscillations along two diameters or as a pair of circular precession modes, one counter-clockwise and the other clockwise. The equations of motion with respect to the rotating axes , are
&& 0 m 0 0 m && + 2m & k m 2 2m & + 0 0 0 = . k m 0 (7.11) 0
2
166 7.2.2 Qualitative effect of damping
Consider now the introduction of isotropic linear viscous damping to the rotor model of Fig. 7.3. Damping with respect to the stationary axes (external damping) is represented in Fig. 7.4 by the four dashpots whose outer extremities are fixed in the stationary reference frame. For small motions of m, the system of dashpots develops a circularly isotropic damping force. The resultant dashpot force acts in the opposite direction of the absolute velocity of m. If m is forced to travel in a circular orbit of radius r with counter-clockwise angular speed n , the damping force cs rn acts tangent to the circle in a clockwise sense. Damping of the relative motion with respect to the rotating system is represented in Fig. 7.5 by the four dashpots whose outer extremities are fixed in the rotating ring.
Fig. 7.5 (from [7.2])
For small motions of m, this set of dashpots develops a circularly isotropic damping force which acts in opposition to the velocity of m relative to the rotating system. In particular, if m is forced to travel in a circular orbit of radius r with counter-clockwise angular speed n as viewed from the stationary axes, it will appear to have a counter-clockwise angular speed n in the rotating frame. The damping force generated by the rotating dashpots is cr r n tangentially to the circle in a clockwise sense. The effects of these damping mechanisms on the free motion of the rotor can be argued qualitatively as follows. Suppose that the damping forces are very small in comparison with the spring forces. This implies that a free orbit will undergo only a small change during one period 2 n as a result of the damping. If initially the orbit is a counter-clockwise circle of radius r, the preceding discussion shows that when the rotor rotation angular speed is subcritical, i.e. less than the natural frequency of the free motion n , both the stationary and rotating dashpots act to retard the motion of m. The rotor does work on the dashpots and the total
acting
167
energy of the rotor orbit is diminished. If initially the orbit is a superposition of a counter-clockwise whirl and a clockwise whirl, a similar argument shows that the radii of both components are diminished after a period. Any rotor orbit generated by an accidental disturbance will thus be damped out and the system is stable. When the system rotates supercritically, the rotating dashpots do work on the rotor and add energy to the rotor orbit. This destabilizing action can be seen by returning to Fig. 7.5 and considering the case of a counter-clockwise whirl of radius r at the absolute angular speed n when > n . The relative motion now is a backward (i.e., clockwise) whirl with angular speed n . The resultant dashpot force acts tangentially to the circular path in a forward (i.e., counter-clockwise) sense with magnitude cr r n . The supercritical rotation of the system thus acts to
drag the rotor forward around its orbit through the rotating dashpots. This action may be seen in an even clearer fashion if, instead of the four dashpots in Fig. 7.5, it is imagined that the interior of the ring is filled with a massless viscous fluid which rotates with the ring and acts to retard the relative motion of the mass m with respect to the ring. When the mass has a circular orbit in the same absolute sense as the ring rotation, but at a slower speed, the viscous drag pulls the mass forward and adds energy to its orbit.
Fig. 7.6 (from [7.2])
The model in Fig. 7.6 (Pippard, 1978) is essentially equivalent. The rotor is modeled by a conical pendulum with a heavy bob suspended in the gravity field by a string of length l . The natural modes are forward and backward conical whirls with the same natural frequency n = g l (for small cone angles). Damping is modeled by allowing the pendulum bob to dip into a glass of water. If the glass is on
168
a variable speed turntable, damping with respect to a rotating frame can be demonstrated. If the fluid is caused to rotate more slowly than the pendulum bob in its circular orbit, the bob experiences a retarding force and sinks toward the vertical. But if the fluid rotates faster than the bob, it urges the bob onwards and causes its orbit radius to increase. Returning to the planar rotor model with both stationary and rotating damping, a forward whirl at absolute angular speed n will be retarded by the stationary damping force and will be urged on by the rotating damping force when the rotation is supercritical, > n . Whether the net effect is to remove or to add energy to the whirl orbit, depends on which force is larger. Neutral stability occurs when the forces are equal. There is then no change in the orbit energy and the free motion persists indefinitely. The quantitative determination of the stability borderline is discussed in the following section, taking account of the frequency dependence of the damping mechanisms.
7.2.3 Whirl speeds of rotor with rotating damping
For some of the systems considered later in this chapter, is useful to study the variation of whirl speeds as a function of the rotational speed . 7.2.3.1 In rotating coordinates In the rotating coordinates, the equations of motion for the planar rotor model of Fig. 7.3 (perfectly balanced shaft) with the rotating damping elements of Fig. 7.5 are
&& cr m 0 0 m && + 2m
& k m 2 2m & + cr 0
0 = . k m 0 0
2
(7.12)
Note that the rotating damping is represented by both a normal damping term and by a quasi-gyroscopic term. Trying solutions of the form
0 t = e , 0
we obtain
( m
+ cr + k m 2 0 2m 0 = 0,
2m 0 + m2 + cr + k m 2 0 = 0. The condition to have nontrivial solutions is
(7.13)
169
( m
or
+ cr + k m 2 + cr + k m 2
) )
+ (2m ) 2 = 0 , ( i 2m ) 2 = 0 , (7.14)
( m
[ m
m2 + cr + k m 2 = i 2m ,
+ ( cr + i 2m ) + k m 2 Denoting
] [ m
,
+ ( cr i 2m ) + k m 2 = 0 .
r =
cr 2 m n
n =
k , m
(7.15)
the characteristic equation becomes
[ + (2
2
2 + i 2 + n 2
] [ + (2
2
2 i 2 + n 2 =0.
From the first factor it may be shown that, if r << 1 , the roots are
2 = r ( n + ) i ( n + ) ,
and from the second factor
1 = r ( n ) + i ( n ) ,
(7.16, a)
4 = r ( n ) i ( n ) ,
The whirl radius is
r n t
3 = r ( n + ) + i ( n + ),
(7.16, b)
=e
C1 e
i n t
+ C2 e
i n t
+e
r n + t
C3
( e
i n + t
+ C4 e
i n + t
(7.17) .
The amplitudes of the second pair of whirls obviously decay with time, whereas the time variation of the amplitudes of the first pair of whirls depends upon the value of the imposed rotational speed of the shaft relative to the value of the natural frequency of the system when not rotating, n . When is less than n the whirls decay and when is greater than n they grow exponentially with time; the system is unstable. From the first equation (7.13) the mode shapes are defined by
170
2 2 0 2 + 2 rn + n = , 0 2
which for the two pairs of roots yields 0 = i , 0 1,2 7.2.3.2 In stationary coordinates In the stationary coordinates, the equations for the same damped system are
& cr y m 0 & + 0 m & & z 0 & k 0 y 0 0 y + + & 0 k cr z z c r
0 = +i . 0 3,4
(7.18)
c r y 0 = . 0 z 0
(7.19)
In (7.19) the rotating damping is also represented by both a normal damping term and a circulatory term defined by a skew-symmetric matrix. Trying solutions of the form
y y0 t = e , z z0 the condition to have nontrivial solutions is (7.20)
( m
or
+ cr + k
+ ( cr ) 2 = 0 ,
( m
( m
2
+ cr + k
( i cr ) 2 = 0 ,
which can be written + cr + k i cr
)( m
+ cr + k + i c r = 0 .
(7.21)
For c r << 2 k m , the roots are
1,2 = r ( n + ) i n ,
with the notations (7.15).
3,4 = r ( n ) i n ,
(7.22)
This analysis masks the importance of the direction of . For in the positive direction some of these roots do not apply. As shown in Section 2.3.2.4 when = n ,
= r ( n ) + i n ,
(7.23)
171
and when = n , therefore the whirl radius is

r = Ae
= r ( n + ) i n ,
(7.24)
r n t i n t
)e
+ Be
r n + t i n t
)e
(7.25)
which is exactly the same motion as found previously in the rotating coordinate system. 7.2.3.3 Whirl speeds In equation (7.17), the first pair of whirls consists of a co-rotating and a counter-rotating whirl of speeds n relative to the moving coordinate
system. Relative to the fixed coordinate system, they are forward whirls of speed n and 2 n respectively. The second pair consists of co-rotating and counterrotating whirls respectively, of speeds n + relative to the moving coordinate system, and backward whirls of speeds 2 + n and n relative to the fixed coordinate system.
a
Fig. 7.7 (after [7.3])
The whirl speeds, that is the imaginary parts of the eigenvalues, are shown as functions of the rotational speed in Fig. 7.7. They are the natural angular frequencies of precession. The values R in Fig. 7.7, a are relative to the rotating coordinate system and the values S in Fig. 7.7, b are relative to the stationary coordinate system. Note that S = R + for positive only. At first sight this is a surprising result. As we have assumed light damping, the natural frequencies do not depend on the magnitude of damping. For zero
172
damping we expect the result obtained for undamped rotors with whirl speeds n . Figure 7.7, b indicates two extra natural frequencies which are speed dependent. They result from the second factor in equation (7.21). This can be obtained from the first factor, replacing by , implying that it is associated with the rotor rotating in a reverse direction. In this case S = R . This confusion is due to the formulation of system equations in real notations. When real notations are used, the information over is folded into that over + . Using complex notation instead of real notation elucidates the character of the additional eigensolutions [7.4]. The complex notation can distinguish a rotor rotating in the positive direction from one in the negative direction. The real notation always gives complex conjugate solutions (the equations of motion have real coefficients), not making any difference in the rotating directions. On the other hand, it is difficult to interpret physically modal vectors with complex vector components such as (7.18). Introducing the complex notation, the complex modal vectors are of the form 0 + i 0 , and become zero for the modes associated with 3 and 4 . This implies that those modes exist merely as a pure mathematical consequence, but they do not physically exist. In other words, they are never excited by any realistic excitation forces. This can be simply shown calculating the forced response, for example, to a constant force rotating with a constant angular speed with respect to the moving coordinate system [7.3].
7.2.4 Quantitative effects of damping
s ( ) by taking the ratio of the tangential drag force, due to stationary damping, to
the radial spring force when the mass particle executes a circular orbit with radius r at the speed . When = n , the backward drag force Ds due to the stationary damping is D s = s n k r .
For the planar model of the rotor, we can define a stationary loss factor
Similarly, a rotating loss factor r ( ) is defined as the ratio of the tangential drag force, due to rotating damping, to the radial spring force when the mass particle executes a circular orbit with radius r at a speed , with respect to the rotating system. In a forward whirl at absolute speed n , the speed with respect to the system rotating at speed is n . Thus, when the system rotates supercritically, the drag force Dr due to the rotating damping is forward and has the magnitude D r = r n k r
( )
(7.26)
(7.27)
173
for an orbit of radius r. If Dr = Ds , we have a steady orbit of fixed radius r. If Dr < Ds , there is a net retardation and the orbit energy decreases. If Dr > Ds , there is net acceleration and the orbit energy increases. Alternatively, the stability is decided by the relative magnitudes of the rotating and stationary loss factors as indicated in Fig. 7.8. The stability borderline condition occurs when the loss factors are equal
r ( n )= s ( n ) .
(7.28)
The forward whirl is unstable for supercritical rotation speeds at which the loss factor of rotating damping is greater than the loss factor of stationary damping.
Fig. 7.8 (from [7.2])
The running speeds which satisfy (7.28) depend crucially on the frequency dependence of the rotating damping mechanism and on the magnitude of the stationary damping.
7.3 Whirl in hydrodynamic bearings

A well-known form of rotor instability, also referred to as oil-whirl or fractional frequency whirl, is a self-excited whirl induced by the nonlinear forces generated in the oil wedge of hydrodynamic journal bearings. Oil whip occurs at the frequency corresponding to the lowest critical speed and maintains that frequency while the rotation speed increases. Both oil whirl and oil whip are superimposed on the synchronous precession due to mass unbalance. Oil whirl amplitudes are usually small. Oil whip amplitudes are large and usually exceed bearing clearance. The onset of oil whip should trigger emergency shutdown of the machine.
174 7.3.1 Oil-whirl and oil-whip phenomena
The following description is adapted from [7.5]. When the shaft starts rotating with a slowly increasing rotating speed, small amplitude synchronous precession (denoted 1X) is observed all along the rotor axis. This is produced by inherent rotor residual unbalance. At low rotating speeds, this precession is stable, an impulse perturbation of the rotor causes a short time transient whirling motion, and the same precession pattern is re-established (Fig. 7.9). At higher rotation speeds (usually below the first critical speed), the forced synchronous precession is not the only regime of vibration. Along with 1X precession, oil whirl appears. Oil whirl is the rotor lateral forward subharmonic precession around the bearing center at a frequency close to half the rotation speed. In this range of rotating speeds the rotor behaves as a rigid body. The amplitudes of oil whirl are usually much higher than those of the synchronous precession. They are, however, limited by the bearing clearance and the fluid non-linear forces.
Fig. 7.9 (after [7.5])
With increasing rotation speed, the pattern of precession remains stable. The oil whirl half frequency follows the increasing rotating speed, maintaining the 1 2 ratio with it. The magnitude of precession radii remains nearly constant and usually high. At the bearing, the precession radius may cover nearly all bearing clearance. In the considered range of rotating speed, the bearing fluid dynamic effects clearly dominate. The forced synchronous precession represents a small fraction of dynamic response, as the half-spectrum indicates (Fig. 7.10).
175
When the increasing rotation speed approaches the first critical speed c1 , i.e. the first natural frequency of the rotor, oil whirl suddenly becomes unstable and disappears, being suppressed and replaced by increasing synchronous precession. The forced precession dominates, reaching the largest radius at the resonance frequency corresponding to the mass/stiffness/damping properties of the rotor. The bearing fluid dynamic effects now yield priority to the elastic rotor mechanical effects. Above the first critical speed, the synchronous precession decays, and the bearing-fluid forces come back into action. With increasing rotating speed, shortly after the first critical speed, oil whirl occurs again. The previously described pattern repeats. The width of the rotating speed range in which the synchronous precession magnitude dominates depends directly on the amount of rotor unbalance: the higher the unbalance, the wider this range [7.4].
Fig. 7.10 (from [7.5])
When the rotor speed reaches about twice the first synchronous critical speed 2 c1 , the half-speed oil whirl frequency reaches the value of the first natural frequency of the rotor. The oil whirl pattern is replaced by oil whip a subharmonic forward precession motion of the rotor. The rotor continues whirling violently with the frequency and the mode shape corresponding to the first critical speed, even if the speed increases further. The rotor precession locks on the frequency of the first synchronous critical speed even if the rotating speed is increased.
176
In this range of high rotating speed, the shaft cannot be considered rigid. Its flexibility causes a strong coupling of the rotor/bearing system. The rotor mass and stiffness parameters become the dominant dynamic factors. The amplitude of oil whip journal precession is limited by the bearing clearance, but the shaft orbit radius may become very high, as the shaft whirls at its natural frequency, i.e. in resonant conditions. The above described phenomena may take slightly different forms in various machines provided with fluid-film bearings and/or seals. Since its discovery (B. L. Newkirk and H. D. Taylor, 1925) much progress has been made in explaining oil whirl and oil whip but the phenomenon is still imperfectly understood. What is clear is that when a rotor is whirling at less than the average velocity of the oil film, the oil drags on the rotor and generates a force opposing the rotor motion. This is positive damping. When the rotor whirl exceeds the average velocity of the oil film, the oil pushes on the rotor. This is negative damping. At the onset of negative damping the rotor becomes unstable and a perturbation will cause the bearing journal to initiate a subsynchronous whirl of ever-increasing amplitude. As the journal motion grows large, the damping becomes positive, the amplitude decreases, and the motion settles into a stable limit cycle. This is oil whirl. This scenario is consistent with oil whirl being a nonlinear phenomenon of the Van der Pol type [7.6]. What is less clear is how oil whirl is transformed into oil whip. Apparently, as the lowest critical speed is approached, the limit cycle frequency synchronizes with c1 , and stays locked to c1 , as long as is greater than approximately 2 c1 . As continues to increase, it is possible for oil whip to be suppressed and for oil whirl to reappear [7.6].
7.3.2 Half-frequency whirl
Consider a lightly loaded journal bearing, i.e., the pressure developed in the film is insignificant, the journal center operates close to the bearing center and the eccentricity is very small compared to the radial clearance. Since the pressure induced flow is assumed to be negligible, the velocity profile of the film in the clearance space is linear, with a maximum value R at the journal surface, as shown in Fig. 7.11. The flow into the wedge of the journal bearing is
qi = 1 L R ( C + e ) . 2
(7.29)
The flow out of the wedge is
qo =
1 L R ( C e ) . 2
(7.30)
177
If pressure is developed in the film, when the bearing is operating under steady conditions, the flow-in is reduced and the flow-out is increased by the pressure induced flow, which balances qi and qo to maintain the flow continuity. However, if the load is small, in the absence of pressure a small whirl velocity is induced to maintain the flow balance.
Fig. 7.11 (after [7.7])
If the instantaneous whirl velocity is for the journal center OJ , then the induced velocity is e as shown in Fig. 7.11. By lifting off the journal from its steady state position, the film volume increases by V = 2 L R e , where 2 L R is the projected area of the bearing. Therefore 1 1 L R ( C + e ) = L R ( C e ) + 2 L R e 2 2 and (7.32) (7.31)
= .
1 2
(7.33)
Hence the rotor tries to whirl at a frequency of half the speed of rotation to maintain the flow balance.
1 , the outward flow is more and therefore pressure is developed in 2 1 the film and the bearing becomes stable. If however < , the flow-in is more, 2 the bearing loses its load carrying capacity and continues to whirl, in order to create
If >
178
more space for the excessive oil coming into the wedge. The rotor thus loses the load carrying capacity and becomes unstable. The frequency of whirl in the rotors under such conditions is observed to be around 0.46 to 0.48 rotation speed.
7.3.3 Onset speed of instability
It can be shown that, for describing the onset of instability, a journal bearing can be represented by a single effective spring coefficient and a single effective damping coefficient. These coefficients can be calculated from the eight dynamic bearing coefficients. However, the two coefficients derived in this manner depend on the precession frequency, such that the effective damping coefficient is negative for small frequencies and becomes positive for higher frequencies. The frequency at which the damping becomes zero is called the instability frequency. This can be found as a function of speed as shown by the corresponding curve in Fig. 7.12. The synchronous precession line crosses the natural frequency curve at the first critical speed (where the rotation speed is equal to the system natural frequency). The half frequency whirl line crosses the curve at the instability threshold known as oil whip.
Fig. 7.12 (from [7.8])
For frequencies less than the instability frequency (i.e. in the region below the curve), the effective damping is negative, and for frequencies greater than the instability frequency the damping is positive. The effective spring coefficient of the bearing together with the flexibility of the rotor determines the resonant frequencies of the rotor-bearing system. Since the bearing stiffness is a function of speed, the resonant frequencies become speed dependent. The lowest of these resonant frequencies is shown by the curve labelled system eigenfrequency. This curve intersects the curve for the instability frequency at a speed denoted as instability threshold speed.
179
Assume that the rotor is subjected to a small disturbance. It will then tend to precess at its lowest natural frequency. However, if the rotor is running below the instability threshold speed, the bearings provide positive damping and the precession motion dies out. As the speed is increased the damping available for the precession diminishes until it becomes equal to zero at the onset speed of instability. Attempting to increase the speed beyond the threshold speed causes the damping to become negative, so that any imposed disturbance is amplified and the system is unstable. From Fig. 7.12 it is seen that if the rotor mass is increased or the shaft is made more flexible, the system natural frequency is lowered, whereby the intersection between the two frequency curves moves to the left and the threshold speed is reduced. Conversely, if external damping is present in the system (for instance, in the supports) the curve of the film instability frequency is lowered, thereby raising the threshold speed. The latter characteristic suggests means by which an otherwise unstable system may be stabilized and also explains why some machinery, notably liquid pumps and high-pressure gas compressors with liquid seals, may operate stable well above the theoretical threshold speed. The damping capacity of the liquid passing through the impeller or seals acts as external damping to the system and stabilizes the rotor.
7.3.4 Crandalls explanation of journal bearing instability
In a second remarkable paper of Stephen Crandall [7.9] a simple explanation of the journal-bearing instability is based on a heuristic hypothesis. A fluid-filled journal bearing is viewed as a pump circulating fluid around the annular space between the journal and the bearing. A small whirling motion of the journal generates a wave of thickness variation progressing around the channel. It is hypothesized that the fluid flow drives the whirl whenever the mean of the pumped fluid velocity is greater than the peripheral speed of the thickness-variation wave. For non-cavitating long bearings, the hypothesis predicts the instability onset correctly for unloaded bearings but gradually overpredicts the onset speed as the load is increased. The following presentation reproduces most of the content of Crandalls paper to make it available to our students. 7.3.4.1 Sommerfelds analysis For simplicity, the discussion is centered on the case of a full circular bearing with two-dimensional non-cavitating flow. The idealized case treated in 1904 by Sommerfeld [7.10] is sketched in Fig. 7.13. A journal of radius R rotates at constant speed within a full circular bearing of radius R + C , where the radial clearance C is very small in comparison to R. The annular space between journal and bearing is filled with an incompressible fluid lubricant with uniform viscosity . The fluid flow is taken to be two-dimensional,
180
i.e. axial flow is assumed to be negligible. In addition it is assumed that cavitation does not occur. The width of the bearing normal to the plane of the figure is L. Sommerfelds principal result is that in the equilibrium position under a vertical load W the journal is eccentrically displaced by a distance e in the horizontal direction.
Fig. 7.13 (from [7.9])
Because of the eccentricity e, the film thickness h varies with position around the annular space. For C << R , the approximate relation is h = C + e cos = C ( 1 + cos ) , where = e C is the eccentricity ratio. Under the assumptions of laminar viscous flow with no pressure variation across the thickness of the film, the only possible flow patterns that satisfy the requirements of Fluid Mechanics in a uniform channel of thickness h are combinations of the two basic patterns shown in Fig. 7.14. Here z is distance along the channel and y is distance across the channel, with 0 < y < h . (7.34)
Fig. 7.14 (from [7.9])
The fluid velocity (in the z-direction at position y) is denoted by w . The volume flow rate across a section of the channel of width L normal to the plane of the figure is denoted by Q and p z is the pressure gradient along the channel. For the linear profile at the left of Fig. 7.14, the pattern depends on the parameter w0 ,
181
which is the velocity of the upper channel wall (the velocity of the lower channel wall is taken to be zero). It can be shown that w = w0 y , h Q = w0 Lh , 2 p = 0. z (7.35)
For the parabolic profile at the right of Fig. 7.14, the pattern depends on the parameter A which is a velocity whose magnitude is four times the peak velocity in the profile or six times the average velocity. y y2 w = A 2 , h h
Q=A Lh , 6
p 2 = A 2 . z h
(7.36)
While the relations (7.35) and (7.36) are strictly correct for steady state flows with constant values of the parameters h, w0 , and A, Reynolds theory of lubrication extends them to apply to slow variations, both in time t and space z, of these parameters. In application to the bearing of Fig. 7.13, the component flows of Fig. 7.14 are superposed, with z = R and w0 = R , with h given by (7.34), and the undetermined parameter A to be fixed by the requirements of continuity, Q = 0 , and uniqueness of pressure, p ( ) = p ( + 2 ) . The value of A so determined is 6 1 2 1 3 A1 = R 2 2 + 1 + cos and the total volume flow rate is
(7.37)
Q1 = R L C
1 2 2+2
(7.38)
The pressure gradients of the component flows in Fig. 7.14 are superposed and integrated to obtain the pressure distribution p ( ) acting on the journal. The resultant of the pressures acting on width L of the journal is a force, acting vertically upwards through the journal center OJ , of magnitude
R W1 = 12 L R C
2 2
( 2 + )(1 )
2 12
(7.39)
The viscous shearing stresses acting on the boundary of the journal also produce a resultant force on the journal, but its magnitude is smaller than the pressure resultant (7.39) by a factor of order C R , and thus may be neglected. The force (7.39) must then be equal and opposite to the applied load W for the journal to be in equilibrium.
182
The remarkable property of the Sommerfeld bearing model is that the equilibrium displacement is at right angles to the applied load. This characteristic implies that an unloaded bearing is unstable with respect to slow forward whirling motions of the journal. To see this, imagine that an external agent moves the center of the journal OJ of Fig. 7.13 in a circular path of radius e in the counter-clockwise sense about the bearing center OB . When the journal passes through the position shown in Fig. 7.13, if the motion is slow enough, the lubricant flow and pressure distribution will be very the same as for the equilibrium configuration shown there. This means that the fluid will be exerting a force very nearly equal and opposite to W on the journal. This force, in the same direction as the velocity of the journal center OJ in its circular path, does positive work on the whirling motion. The Sommerfeld bearing thus promotes whirling instability for very slow forward whirling speeds. To consider more rapid whirling speeds it is necessary to extend the Sommerfeld analysis to include motion of the journal center OJ . 7.3.4.2 Whirling stability of unloaded bearing
( W = 0)
According to (7.39), the equilibrium position for an unloaded bearing has zero eccentricity ( = 0 , e = 0) . In this configuration, the parameter A1
of equation (7.37) vanishes and the lubricant flow pattern is simply the linear profile shown on the left of Fig. 7.14.
Fig. 7.15 (from [7.9])
The bearing kinematics are displayed in Fig. 7.15 for the case where the journal center OJ whirls at the steady angular speed in a circle of radius r0 about the equilibrium position in which OJ and OB coincide. Note that the diametrally
183
opposite sections of maximum and minimum film thickness advance around the bearing at the speed . At the location , the film thickness is h ( ,t ) = C + r0 cos ( t ) . (7.40)
Note that the dependence of h on space, , and time, t, is that of a progressive wave circling the annular channel with an angular phase speed or a linear phase velocity R . The explanation of Newkirk and Taylor [7.11] is based on an application of the continuity requirement to the flow in a channel whose thickness varies according to (7.40). The continuity requirement applied to a differential arc of length R d is Q h +L =0. R t (7.41)
Newkirk and Taylor assumed that the lubricant flow retains the linear profile of Fig. 7.14 so that Q = R L h 2 which reduces (7.41) to
h h + =0. 2 t
(7.42)
When h ( ,t ) from (7.40) is substituted in (7.42), the result is that continuity cannot be satisfied unless
(7.43)
Newkirk and Taylor took this to provide analytical verification of the halffrequency whirl phenomenon which they observed in a vertical shaft running in a bearing with plentiful oil supply. It also appeared to explain the oil resonance peak in response when the rotation speed was twice the natural frequency of the system. The simple result (7.43) was less satisfactory in explaining the oil-whip phenomenon where the whirling frequency remains at the natural frequency as the rotation speed is increased, although it was noted that the onset of whipping always occurred at speeds equal to or greater than twice the natural frequency. In Crandalls heuristic hypothesis [7.2] the rotating journal is considered to be a pump impeller which maintains the lubricant flow pattern with linear profile when the journal is unloaded and centered. The fluid velocity varies linearly from w = 0 at the bearing to w = R at the journal. This flow can be decomposed into a mean flow with uniform velocity w = R 2 and no vorticity plus a residual flow with zero mean velocity and large vorticity. It is assumed that the mean flow is available to encourage (or discourage) small whirling perturbations of the centered configuration. If a whirl involving a thickness variation wave of the form (7.40) is imposed, Crandalls hypothesis postulates that energy will be pumped into the whirl
184
if the mean flow velocity R 2 is greater than the phase velocity R of the whirl around the periphery of oil film. Conversely, energy will be removed from the whirl if the phase velocity of the whirl is greater than the mean velocity of the lubricant. Neutral stability occurs when these velocities are equal, i.e. when = 2 . This hypothesis thus explains half-frequency whirls and oil whip for a system which has an unloaded bearing with plentiful oil supply. If the system provides little restraint on the journal, its motion will be primarily determined by the fluid film forces acting on it. When a low frequency whirl ( < 2) is accidentally started, energy will be pumped into the whirl, accelerating the whirl until gets sufficiently close to 2 that the energy pumped into the whirl just balances the system energy losses in a steady half-frequency whirl. If the system provides considerable restraint on the journal and only permits appreciable whirling motion at a natural frequency, the fluid film in the bearing will remove energy from accidental whirls at a natural frequency whenever < 2 . However, when > 2 the fluid film will pump energy into any such whirl. Whether or not oil whip occurs depends on whether the energy supplied by the fluid film is sufficient to overcome the system losses. Both of the previous explanations are incomplete in the sense that they do not make use of all the physical requirements involved. Both arguments use the flow pattern of the underlying steady centered flow and both use the kinematics of a small whirling perturbation. The Newkirk and Taylor argument makes explicit use of the continuity requirement for the perturbation but neither explanation explicitly invokes any consideration of the pressure developed in the oil film. Historically, the first complete dynamic analysis of the whirling stability of an unloaded bearing was given in 1933 by Robertson [7.12]. The development which follows is essentially just a linearized version of Robertsons analysis. We consider the whirl defined by the eccentricity r0 r0 << C and the frequency to be a small perturbation of the underlying centered rotation in which the flow distribution is simply
w = R z C ,
0< z<C.
(7.44)
When the journal is whirling, the flow pattern will be a superposition of the two profiles of Fig. 7.14. To fit the conditions of Fig. 7.15 we take
w = R
y y2 y + A 2 , h h h
(7.45)
where h ( ,t ) is given by (7.40) and A is to be determined from the continuity requirement (7.41) and the requirement of single-valued pressure, p ( ) = p ( + 2 ) .
185
Using a linear perturbation analysis, we neglect terms of order comparison with terms of order r0 C and find A = 6R r0 cos ( t ) . C 2
Lh r0 R L cos ( t ) , 2 2
( r0 C ) 2
(7.46)
in
The corresponding total volume flow rate for the lubricant film is
Q = R
Q = R
LC r0 R L cos ( t ) . 2
(7.47)
The pressure distribution is obtained by integrating the pressure gradient given in (7.36). To first order in r0 C p ( ,t ) p0 = 12 R 2 r0 sin ( t ) . 3 2 C (7.48)
The resultant force acting on the journal due to these pressures is directed at right angles to the journal displacement r0 and has the magnitude F = 12 r0 L R3 . 3 2 C (7.49)
The sense is such that when 2 > , F has the same direction as the instantaneous velocity VJ of the journal center. The rate at which the fluid film forces do work on the journal (i.e., the power flow into the whirl) is F VJ = 12 L R3 2 r . 3 0 2 C (7.50)
This analysis shows that the amplitude A of the component flow with parabolic profile, the pressure in the fluid film, the resultant force on the journal, and the power flow into the whirl all are proportional to the factor ( 2 ) . Low speed whirls are encouraged and high speed whirls are discouraged. The whirling frequency of neutral stability is = 2 . These results can be compared with the two simplified arguments considered previously. The assumption in the Newkirk and Taylor argument that the velocity profile remains linear is equivalent to assuming that the parameter A in (7.36) vanishes, which according to (7.46) implies that must equal 2 . Furthermore the vanishing of A implies an absence of pressure gradient and consequently an
186
absence of resulting force so that at the particular whirl frequency = 2 the fluid film neither retards nor advances the whirl. This is the neutral stability condition. Crandalls hypothesis that the mean flow drives the whirl whenever the mean fluid velocity is greater than the phase velocity of the whirl is essentially a qualitative statement equivalent to the quantitative statement represented by equation (7.50). It happens that for an unloaded bearing the frequency of neutral whirl is given correctly by the heuristic hypothesis. For loaded bearings it is difficult to see how the Newkirk and Taylor argument can be extended to predict any other frequency of neutral whirl than = 2 . Crandalls hypothesis was however easily extended. It no longer predicts the exact frequency of neutral whirl but it provides useful approximations for moderate loads. A linear perturbation analysis [7.9] shows that the neutral stability whirl frequency
2 ( ) = 2 9 (1 2 )
3 + 1 2
(7.51)
varies from = 2 at = 0 for an unloaded bearing, to 3 when the load approaches infinity and the eccentricity ratio approaches unity. The variation of according to (7.51) is represented by the curve labeled A in Fig. 7.16.
Fig. 7.16 (from [7.9])
A computation based on Crandalls hypothesis [7.9] yields the curve B in Fig. 7.16. It provides a useful approximation of the whirl frequency at the neutral stability for values of the steady state eccentricity ratio that are less than = 0.5 .
187
7.3.5 Stability of linear systems
For equations of motion with real coefficients, the characteristic equation
f ( ) = a 0 n + a 1 n 1 + a2 n 2 + ...... + an 1 + an = 0 ,
(7.52)
has n roots of the general form i = i + i i . Repeated roots are not considered here. If any i > 0 , the system will be unstable. The types of motion associated with the position of a root in the complex plane (Fig. 7.17) are summarized in Table 7.1.
Fig. 7.17 (after [7.3]) Table 7.1
Position of root
in Fig. 7.17 1 2 3 4 5
Type of motion
stable; damped oscillatory stable; damped aperiodic oscillatory unstable; divergent (distance increases exponentially with time) unstable; oscillatory with increasing amplitude, referred to as dynamically unstable
It is often of interest to decide whether or not a system is stable without having to solve its characteristic equation. In 1877, E. J. Routh established an
188
algorithm for determining the number of roots of a real polynomial which lie in the right half-plane. Independently, in 1895, A. Hurwitz gave a second solution. The determinant inequalities obtained by him are known as the Routh-Hurwitz stability conditions. Consider the matrix with real elements
| a0 a1 a3 a2 a5 a4 L L L L a 2n 2 a 2 n 1
| | |
0 a1 a3 L L a 2n 3
| 0 L 0 L | | a0 L 0 L | | a2 L a 0 L L L
0 0 0 . an
(7.53)
Construct the following n determinants indicated by broken lines D1 = a1 , D2 = a1 a3 a0 = a1a 2 a 3a 0 , a2 (7.54) (7.55)
a1 D3 = a 3 a5
a1 a3 a5 M
a0 a2 a4
a1 = a1 a1 a 4 a 5 a 0 + a 3 a1 a 2 a 3 a 0 , a3
a0 a2 a4 0 a1 a3 . 0 a0 a2 . L L L . . . . M L an
(7.56)
Dn =
(7.57)
a 2 n 1 a 2 n 2
If r > n or r < 0 then ar = 0 . A necessary and sufficient condition for all roots of f ( ) = 0 to lie in the left half-plane is that all partial determinants be positive D i > 0 . If an > 0 , it is only necessary to test the determinants from D1 to D n 1 as D n = an D n 1 .
189
The coefficients ar may be expressed in terms of the roots i of equation (7.52), for example a1 = a0
i ,
i =1
a 2 = a0
i j (i j )
i =1 j =1
and so on. By considering the forms of these expressions it is easy to show that all a i have the same sign as a 0 if all the roots have negative real parts. The only way in which any of the coefficients may be zero or take the opposite sign to a 0 is for one or more of the roots to have positive real parts. Hence, a necessary, but not sufficient, condition for all roots of f ( ) = 0 to lie in the left half-plane is that all a r ( r = 1, 2 ,..., n ) be positive.
7.3.6 Instability of a simple rigid rotor
Consider a mass 2m attached at the middle of a rigid shaft running in two identical fluid film bearings. Due to symmetry, the bearings share the rotor load equally, having the same operating eccentricity and dynamic coefficients. Only the translatory whirl is of interest. 7.3.6.1 Threshold speed and unstable whirl frequency For constant load and rotation speed, the journal center maintains a steady state equilibrium position in the bearing clearance, defined uniquely by the Sommerfeld number (6.3)
DL R S= 2 W C
(7.58)
which represents the operating conditions. The equation of motion with respect to the equilibrium state is written, considering each journal having a mass m, as
& c yy y m 0 & + 0 m & & czy z
c yz y & k yy + czz z & k zy
k yz y f y = . k zz z fz
(7.59)
Substituting solutions of the form
y = y0 e t ,
z = z0 e t ,
(7.60)
into the homogeneous part of equation (7.59) we obtain
190
m2 + c yy + k yy c zy + k zy
y0 0 = . m2 + c zz + k zz z0 0
c yz + k yz
(7.61)
For nontrivial values of y0 and z0 , we obtain the characteristic equation
( m
+ c yy + k yy
)( m
+ czz + k zz ( czy + k zy )( c yz + k yz ) = 0 .
(7.62)
This is a fourth order algebraic equation of the form
a 0 4 + a 1 3 + a2 2 + a3 + a4 = 0 ,
where
(7.63)
a 0 = m2 ,
a1 = m ( c yy + czz ) , a 2 = m ( k yy + k zz ) + ( c yy czz c yz czy ) , a3 = c yy k zz + czz k yy ( c yz k zy + czy k yz ) , a4 = k yy k zz k yz k zy .

The Routh-Hurwitz conditions of stable precession are (7.64)
a1 > 0 , a 2 > 0 , a 3 > 0 , a 4 > 0 , a1 a 2 a 0 a 3 > 0 ,

2 2 a 1 a 2 a 3 a 0 a3 a1 a4 > 0 .
(7.65) (7.66) (7.67)
At the onset of instability, the root of equation (7.63) is purely imaginary
= i ,
where is the whirl frequency. Substituting (7.68) into equation (7.63) yields
(7.68)
a 0 4 a2 2 + a4 + i a 1 3 + a3 = 0 .
Equating the real and imaginary parts to zero, we find
(7.69) (7.70) (7.71)
a 0 4 a2 2 + a4 = 0 ,
( a3 a 1 2 ) = 0 .
From equation (7.71) we obtain the whirl frequency at the stability limit
191
2 =
a3 . a1 = th
(7.72)
Substituting (7.72) into equation (7.70) yields

2 2 a1 a 2 a 3 a 0 a3 a1 a4 = 0 .
(7.73)
This means that at the rotation speed = th the stability boundary is encountered and the unstable whirl frequency is given by equation (7.72). Usually, the threshold speed and the whirl frequency are expressed in terms of the eight dimensionless bearing coefficients Ki j = C ki j , W Ci j = C ci j , W
i, j = y, z
(7.74)
where C is the radial clearance, W = m g is the static load on the bearing and is the rotational angular speed. Denoting
A1 = C yy + C zz , A2 = K yy + K zz , A3 = C yy C zz C yz C zy ,
(7.75)
A4 = C yy K zz + C zz K yy C yz K zy + C zy K yz , A5 = K yy K zz K yz K zy ,
and substituting (7.74) and (7.75) into (7.64), we obtain

W W2 W2 W a0 = , a = A , A2 + a = A3 , 1 1 2 g g C gC C W2 W a3 = 2 A4 , a 4 = A5 . C C
2
(7.76)
Substituting now (7.76) into equations (7.73) and (7.72) we obtain the dimensionless onset speed of instability
2 th
g C
A1 A3 A 4
2 A2 4 + A1 A5 A1 A 2 A 4
(7.77)
and the dimensionless unstable whirl frequency
A4 = g C A 1 = th
2 th
(7.78)
192
In practical calculations, the Sommerfeld number is computed at the operating speed. If the threshold speed is higher than the operating speed, the system is stable; otherwise, it is unstable.
7.3.6.2 Critical rotor mass

The threshold of instability can be defined by means of general design charts in which both the rotor and the bearing parameters are included. If the rotor is assumed to be rigid, then the rotor mass is the only rotor parameter. In the case of laminar incompressible fluid-film bearings, the dimensionless unstable whirl frequency is only a function of the Sommerfeld number S (and also of the eccentricity ratio ). Hence it is possible at any Sommerfeld number (or eccentricity ratio) to calculate the journal mass required for the whirl motion to exist. In dimensionless form, the journal mass for a rigid rotor may be expressed as
C m W . Equation (7.78) can be written C yy K zz + C zz K yy C yz K zy + C zy K yz C m 2 =K= , W C yy + C zz

where K is the dimensionless rotor mass per bearing.
(7.79)
Fig. 7.18 (from [7.8])
193
For a rotor operating in incompressible plain cylindrical bearings, a curve giving the relationship between K and (Fig. 7.18) defines the threshold of instability. This curve can be used to determine the onset speed of instability once it has been found how the eccentricity ratio depends on the rotor speed. Equation (7.77) can be written (dropping the index th)
C m 2 W
A1 A3 A4
2 A2 4 + A1 A5 A1 A 2 A4
= K+
A3 A5
= A2
K A3 A5 A 2 K + K 2
or using (7.58)
Cm = W
C mW
K K A3
A5 A 2 K + K 2
K A3
. (7.80) A5 A 2 K + K 2 R DL C The dimensionless critical rotor mass can be expressed as a function of the Sommerfeld number [7.8] as
2
2 S
. (7.81) 2 S A5 A 2 K + K 2 R DL C Figure 7.19 shows a stability design chart for a plain journal bearing [7.8]. The dimensionless critical mass C mW D L (R C ) 2 is plotted as a function of
2
C mW
K A3
the Sommerfeld number S for two values of the ratio L D . A rotor is stable for operating conditions at the left of the curves.
Fig. 7.19 (from [7.8])
194
Similar design charts are shown in Figs. 7.20 giving the dimensionless critical rotor mass at onset of instability as a function of S. Figure 7.20, a applies to elliptical bearings operating with an incompressible lubricant in a laminar regime, while Fig. 7.20, b applies to partial arc, centrally loaded bearings.
a
Fig. 7.20 (from [7.8])
The procedure for determining the threshold speed is the following. Calculate the dimensionless rotor mass CmW D L (R C )2 . Enter the appropriate design chart and read off the corresponding value of S. The speed corresponding to this Sommerfeld number is the threshold speed, i.e. the rotor speed at the onset of instability. If the operating speed is higher than the threshold speed the design must be revised. The charts assume the rotor and the pedestals to be rigid. However, the threshold speed with a flexible rotor and flexible pedestals can be calculated on the basis of the given design charts.
7.3.7 Instability of a simple flexible rotor
Consider a symmetric rotor consisting of a shaft with a stiffness 2k and a lumped central mass 2m . The shaft is supported at the ends in fluid film bearings. The angular speed of rotation is . External damping is provided at the rotor mass and at the bearings, with damping coefficients 2ce and c s , respectively. Although practical rotors are considerably more complex, this simplified model contains the essential features of the problem and admits closed form solutions [7.13].
195
7.3.7.1 Rotor with external damping For small amplitudes, with y, z, the coordinates of the rotor mass and y 1 , z 1 the coordinates of the journals, the equations of the free motion can be written as
& ce 0 y & y k 0 y k m 0 & = + + 0 m & & 0 ce z & 0 k z 0 z &1 k yy k yz y1 cs c yy c yz y = &1 k zy k zz z1 0 c zy c zz z
0 y1 = k z1
&1 0 y . &1 cs z
(7.82)
The eight bearing coefficients depend on the particular bearing geometry and on the operating conditions, expressed through the Sommerfeld number S. Considering the plain cylindrical journal bearing as a representative example, and assuming it to be short (L D < 1 2 ) and operating at sufficiently high speed such that L 2 S >1, D then the dynamic bearing coefficients are approximately given by (6.30)
k zz 2k yy
2
(7.83)
8 W , C
2
c yz c zy
8 W , C
3 2
W L R L c yy c zz 2 2 S = 2 L , C D C D L W k yz k zy 2 S = c yy . C 2 D
2
(7.84)
, whereby equations (7.82) can be written as

k yy + i c yy X k + i c zy zy where
On the threshold of instability, the motion is purely harmonic with frequency k yz + i c yz y1 0 = , k zz + i c zz X z1 0
(7.85)
196
X = k m 2 + i ce k m + i ce
2
) i c ,
s
(7.86)
X =
k k m 2 m 2 ( ce )2
( k m )
2 2
+ ( ce ) 2
k 2 ce c i + s . k m 2 2 + ( c ) 2 e
This value should be equal to the root of the determinant of equation (7.85) which, by means of (7.84) becomes
+ i c yy ( ( c yz ) 2 1 k yy k zz ) 2 4
2 k yz
X=
1 k yy + k zz 2
1+
. (7.87)
As is equal to
1 or less, the second term inside the square root is of 2
the order of ( 1 4 ) 2 and can be ignored. Thus, with k yz = approximately equal to
c yy from (7.84), X is
X = K + i c yy , 2
where the average bearing stiffness K= 1 6W . k yy + k zz = 2 C
(7.88)
(7.89)
disregarding the term ( ce ) 2 , the whirl frequency is determined as
Equating the real and imaginary parts of equation (7.86) and (7.88) and
=0 =
kK , m (k + K)
(7.90)
which is the natural frequency of the system when the shaft is supported in bearings with stiffness K. The condition for stability is
1 K +k 0 cs + 0 c e > 0 c yy , 2 k
2
197
0 cs +
R L K +k 0 c e > 2 0 L . C D k
(7.91)
The whirl motion is a forward precession. In the absence of external damping, the relation (7.91) shows that the rotor is unstable when the speed exceeds twice the first critical speed. This is true only for sufficiently large values of the Sommerfeld number, just as the inequality altogether is based on several assumptions. Hence, (7.91) primarily serves the purpose of estimating the magnitude of external damping required for stabilization. It should be added that, instead of providing external damping which acts directly on the journal, a much more effective method of stabilizing hydrodynamic bearing whirl is to mount the bearing in a flexible, damped support. By proper tuning it is possible to eliminate the instability completely [7.13].
Fig. 7.21 (after [7.14])
198
7.3.7.2 Rotor without external damping If the external damping in the simple flexible rotor is neglected, introducing ce = 0 and cs = 0 in (7.86), then substituting X in (7.85) and not limiting the analysis to plain short bearings, we obtain
2 th
g C
2 A3 A1
A1 A3 A4
2 C A1 A2 4 + A1 A5 A1 A2 A 4 1+ 2 S st A4
(7.92)
where st =
mg is the static deflection of the shaft. k
Figure 7.21 shows stability charts computed based on equation (7.92) for a circular and an elliptical bearing, with L D = 0.8 and C R = 2.1 10 3 , the latter having a horizontal radial clearance three times the vertical radial clearance. Practical measures taken to enhance stability above th are, amongst others, increasing bearing clearance, decreasing lubricant viscosity, shortening bearing length or, equivalently, making a central circumferential oil groove, increasing shaft diameter or shortening shaft length/bearing span, increasing bearing damping and bearing stiffness anisotropy, adding squeeze film dampers and changing to multi lobe or tilting pad bearings.
Fig. 7.22 (from [7.15])
199
7.3.8 Instability of complex flexible rotors
Complex flexible rotors can be modeled by finite elements. The equations of motion contain the system matrices which are assembled from the element matrices, as shown in Chapter 5. The study of the free precession leads to an eigenproblem with speed dependent eigenvalues. The Campbell diagram (Fig, 7.22, a) displays the natural frequencies of precession as a function of the rotational speed. The stability diagram (Fig. 7.22, b) shows the real part of the eigenvalues versus the rotational speed. The onset speed of instability is determined at the intersection of such a curve with the speed axis. The problem is fully illustrated in Chapter 4.
7.4 Interaction with fluid flow forces

Any whirl motion of the shaft in rotating machinery (turbines, compressors, pumps) affects the flow field of the working fluid, thereby setting up differential forces and moments. Thus, a coupling exists between the motion of the shaft and the aerodynamic/hydrodynamic reaction forces which could be destabilizing. One mechanism for such a coupling, postulated by Thomas [7.16] and Alford [7.17], applies to axial flow machines where a radial displacement of the wheel center in a stage generates a transverse force, proportional to the displacement. It is known as steam whirl or blade-tip-clearance effect.
b
Fig. 7.23 (after [7.14])
7.4.1 Steam whirl
In the stage of an axial flow machine (Fig. 7.23, a), a radial deflection (downwards) will close the radial clearance on one side (bottom) and open the
200
clearance on the opposite side (top). The closer clearance zone is expected to operate more efficiently and to be more highly loaded than the open clearance zone. There will be an increased horizontal force F1 at the bottom and an opposite decreased horizontal force F 3 at the top, with vertical forces F 2 and F 4 of intermediate magnitudes (Fig. 7.23, b). A resultant net tangential force F is generated (Fig. 7.23, c) which, if larger than the external damping force, can induce a whirl in the direction of rotor rotation (i.e., forward whirl). The transverse force is taken proportional to the radial displacement of the wheel center. The coefficient of proportionality is of the form [7.13]
d T = 0 0 , h 2r H d H
(7.93)
where is the stage efficiency, T is the stage torque, r is the pitch (mean blade) radius, H is the vane height, h is the tip clearance, and subscript o refers to the concentric condition. Consider a symmetric rotor consisting of a shaft with a stiffness 2k and a lumped central mass 2m . The shaft is supported at the ends in bearings with stiffness k B such that the undamped natural frequency of the system is obtained from (7.90) substituting K = k B . The angular speed of rotation is . External damping is provided at the rotor mass and at the bearings, with damping coefficients 2ce and cB , respectively. Assuming the central rotor mass to be a turbine stage, with y, z, the coordinates of the rotor mass and y 1 , z 1 the coordinates of the shaft at the supports, the equations of the free motion are
& ce 0 y & 0 y y m 0 & k 0 y k 0 y1 + + = + = 0 m & & 0 ce z & 0 z z 0 k z 0 k z1 &1 k B 0 y1 0 y c = B . &1 0 k B z1 0 cB z (7.94) Neglecting second order terms in , ce and cB , the characteristic equation becomes
2 k k kB i = 0. m + ce + c B + k + kB k + kB 2
201
(7.95)
Setting equal to i at the threshold of instability, the whirl frequency is found to be equal to (7.90)
=0 =
while the stability condition becomes
k kB , m ( k + kB )
k 0 ce + 0 c B > . k +k B
(7.96)
At the onset of instability, the whirl motion is forward precession when is positive, and backward precession when is negative. If the aerodynamic load torque T of the single stage increases with speed, then equation (7.96) shows that there may be an onset speed of instability above which the inequality is no longer satisfied. This threshold speed is load dependent, so that the instability starts at a certain level of power. It can be raised by either stiffening the shaft or increasing the damping. It is known that increasing the bearing support stiffness usually reduces the effective damping coefficient and often has a very small effect on the critical speed. Thus, in practice, the parameter most feasible to modify for improved stability is the bearing support damping. Indications are that the proposed form of rotor-fluid flow interaction is in qualitative agreement with practical experience, but the particular expression for , equation (7.93), only covers one special mechanism for the interaction. Although the equation seems to yield values of the right order of magnitude, there is evidence to suggest that, in addition to torque, such parameters as flow and pressure level may be of equal importance [7.13].
7.4.2 Impeller-diffuser interaction
Hydraulic forces acting on the rotor of a centrifugal pump arise in two different ways. First, when the shaft center is fixed, radial forces occur as a result of distribution of static pressure and fluid momenta. These forces are termed steady excitation forces. Second, when the shaft center also moves as a result of precession, additional forces are generated due to the interaction of impeller and surrounding fluid. These are dynamic forces and are described by dynamic coefficients.
202
As the rotor moves eccentrically within the pump, significant forces are developed a) in the annular seals; b) between the impeller shroud and the stator; c) between the impeller back disc and the stator; and d) between impeller and diffuser. When a shaft is displaced laterally in a ring seal, a strong restoring force is set up in the clearance space. This hydrostatic force increases with the pressure drop and the amplitude of shaft lateral motion. In effect, the clearance acts as a spring. The phenomenon is known as the Lomakin effect (Section 6.10.1). For a whirling shaft, a destabilizing tangential force appears normal to the shaft radial displacement, yielding a cross-coupled stiffness. This is approximately proportional to the average tangential velocity of the fluid. Wide variations occur in the inlet swirl velocity, depending upon where annular seals are located in the pump. Swirl brakes are commonly used upstream of seals to sharply reduce this velocity. A deliberately roughened stator also reduces the average tangential velocity. Recent research revealed that impeller shrouds develop significant destabilizing forces. Conservation of momentum considerations for leakage flow that proceeds radially inward, along the impeller shroud, can generate high tangential velocities for the flow entering the neck ring and balance piston seals. Enlarged clearances of the exit seal, due to wear or damage, increase the tangential velocity over the shroud and into the seal. Lateral forces on pump impellers can also have a significant dynamic effect. Forces are exerted on the impeller as a result of the asymmetry of flow caused by the volute or diffuser and the motion of the impeller. Supersynchronous forces are generated in volute pumps at the vane passing frequency. In turn, the diffuser pumps can generate greater forces in the subsynchronous range, depending on the casing tip clearance. Subsynchronous forces with frequency components from 30% to 80% of running speed can be attributed to partial flows or associated with destabilizing forces. The impeller/casing interaction can be described with dynamic force coefficients.
Self-excited vibrations are common in centrifugal compressors, apparently with greater frequency in back-to-back designs and in high density and high-speed industrial units. Actually, a large percentage of high speed compressors will sustain some low-level subharmonic vibrations even during normal operation. Some of these units develop instability problems. The vibration frequency observed is normally the fundamental bending natural frequency of the rotor.
Self-excited vibrations due to aerodynamic forces are twofold: a) vibrations where the destabilizing forces are function of the whirling motion of the rotor; they are produced by labyrinth seals and asymmetric blade loading of the impeller; and b) vibrations where the main flow is destabilized by hydrodynamic or viscous forces which are independent of the resulting mechanical vibration, such as those produced by rotating stall and/or surge.
203
Aerodynamic forces arising from the interaction of the impeller with the driven fluid introduce both direct and cross-coupled stiffnesses. In a radial flow machine, the change in force due to the clearance variations is negligible compared to that generated by the change in fluid momentum in an eccentric impeller, since the former is perpendicular to the main flow path. Forced vibrations have also been recorded in all high-density compressors. They have the following typical behaviors: a) they appear relatively near to surge and are very stable in amplitude; b) asynchronous frequency is very low (10 percent of rpm); and c) asynchronous amplitude depends on the tip speed and gas density.
7.5 Dry friction backward whirl

Dry friction backward whirl is a self-excited vibration in systems with rotorto-stator contact. The rotor is in continuous contact with the stator, slipping continuously on the contact surface and whirling backward at a non-synchronous frequency which depends on damping and the ratio between rotor and stator stiffnesses.
7.5.1 Rotor-stator rubbing
If the rotor contacts the stator, a wide variety of different motion types are possible, from synchronous forward whirl, sub- and super-harmonic precession, to backward whirl and chaotic motion. The contact problem is highly non-linear even if the rotor and the stator are linear. The phenomenon of backward whirl was recognized the first time by Newkirk [7.18] investigating radial shaft rubbing. In an experiment, the rubbing became severe and due to the large friction force the rotor started rolling violently in the inner surface of a rigid stator ring. The rolling did not take place until the shaft rubbed very hard on the stator ring. Despite the strong non-linearity due to contact, rotor unbalance can cause purely synchronous precession. However, in some circumstances, the synchronous motion may become unstable and the rotor motion may turn into a non-synchronous precession which can be very distructive. The backward whirl may emerge even in speed ranges where the synchronous forward precession is stable. A rub occurrence in a 600 MW turbo generator which was totally destroyed during backward whirl is described in [7.19]. Damages resulting from rotor-to-stator contact are reported by insurance companies to be the most frequently failures in steam turbines and to cover up to 22% of all indemnification payments in the field.
204
Dry friction whirl is mainly due to inadequate or improper lubrication in close clearance machinery. Den Hartog gave a simplified description of the mechanism causing this effect in his well-known book [7.21]. The same explanation was presented in [7.22]. An analytical solution taking into account rotor unbalance, slip and friction between rotor and stator, the mass, damping and stiffness properties of the rotor is presented in [7.19].
Fig. 7.24 (from [7.20])
A recent investigation supported by experimental work is reported in [7.20]. Figure 7.24 illustrates the displacement of a rotor without slip at the inner surface of its housing, at several consecutive times. It can be seen that while the rotor rotation is only 60 degrees, the whirl motion completes one whole cycle.
7.5.2 Dry friction whirl
Basically, the dry whirl is the result of an unequalized Coulomb friction force appearing at a point on the periphery of a shaft or rotor when it is deflected for some reason against the side of a clearance annulus (Fig. 7.25). Since the friction force, F f , is approximately proportional to the radial component, N, of the contact force, we have the preconditions for instability. The friction force F f can be resolved into a couple F f C and a parallel force of equal magnitude through the (deflected) shaft center. The couple acts merely as a brake on the shaft, which is supposed to be driven at uniform speed , requiring some increase in the driving torque and is inconsequential. The force F f through the center of the shaft, however, drives it in a direction tangent to the bearing inner surface. The direction of F f changes with the position of the shaft in bearing or guide, so that the shaft will be driven around, as indicated by the dotted
205
circle. It will be noticed that the shaft is driven around the clearance in a direction opposite to that of its own rotation (counterwhirl). As the point of contact moves around the periphery of the housing or guide, so does the force remain tangential to the precession circle, thus perpetuating the whirl.
Fig. 7.25
Fig. 7.26
This backward precession can arise if the contact friction force is large enough to prevent slipping between the rotor and stator. A no-slip condition at the point of contact requires the rotor to precess at a speed R related to the shaft rotational speed through the kinematic requirement (Fig. 7.26)
R =
r , C
(7.97, a)
where r is the radius of the contacting shaft and C is the radial clearance. Since annular clearances even at non-seal locations are very small compared to the radius of the rotor, this results in extremely high frequencies of the backward precession which cannot be reached in reality, with slipping motion and flexible contact surfaces. However, still very high super-synchronous frequencies and high forces can occur which can cause serious damage. Note that backward precession induces alternating stresses in the shaft which can lead to fatigue failures. If the rotor is running with slip, a different expression can be established for the whirl speed. The normal force is equal to the difference between the centrifugal
2 k C . In the tangential direction, force and the elastic restoring force N = m C R the viscous damping force is Fd = ce C R and the friction force is F f = N , where
is the coefficient of friction.

To maintain continuous contact, the resulting radial force must be directed 2 outward. The condition m C R > k C yields
R >
k , m
206
i.e. the absolute value of the whirl frequency must be higher than the rotor undamped natural frequency. In order to sustain the whirl, the resulting tangential force must be in the direction of the friction force, F f + Fd 0 ,
2 m C R k C + ce C R 0 .
This condition leads to the following quadratic equation in the whirl frequency
2 R +
ce k R 0 m m
with the acceptable solution
R =
where e =
ce . 2 km
k e m
1+
e2 2
(7.97, b)
This simple explanation is based on a single-disk symmetrical rotor model in which the plane of rubbing coincides with the plane containing the center of the lumped mass. In the general case, however, rubbing in one plane may cause a large amplitude whirl in another, with the whirl amplitude not bounded by the rubbing clearance. Although Den Hartogs explanation seems to indicate that all cases of clearance rubbing would initiate uncontrolled rotor vibration, such is not the case in actual machinery, mainly due to stator flexibility and inherent damping in the system. It is considered that, since there must be some minimum value of the coefficient of friction at the rubbing surface, below which the vibration cannot be sustained, it is probable that, given a minimum damping level, vibration would decay regardless of the degree of stator stiffness present. Whether backward whirl with slip or pure rolling occurs, is determined by the friction angle. Below a minimum friction angle backward whirl is impossible.
7.6 Instability due to asymmetric factors

The purpose of this section is to present the peculiar theoretical characteristics of unsymmetrical rotors in relatively simple quantitative terms. Shaft orthotropy produces a peak of resonant vibration about half the regular critical speed and, for small damping, a range of possible unstable behavior between the two
207
critical speeds. Moreover, the unbalance response is a function of the angle at which the unbalance happens to lie in the rotor. Rotors having disks with unequal diametral moments of inertia (e.g., two-bladed small airplane propellers, wind turbines and fans) are dynamically unstable above a certain speed and some of these may return to a stable condition at a sufficiently high speed, depending on the particular magnitudes of the gyroscopic coupling and the inertia inequality.
7.6.1 Parametric excitation
A separate class of rotordynamic instabilities is characterized by conditions in the machine which are described by differential equations with variable coefficients. Since the coefficients are made up of the parameters of the rotorbearing system, this type of instability is referred to as parametric excitation. In contrast to the self-excited instabilities, in which the whirling is always at a subsynchronous frequency equal to an eigenvalue of the system, parametric excitations produce whirling which can be synchronous, subsynchronous, or supersynchronous, depending on how the parameters vary in each particular case. Some of these instabilities are more like a critical speed or forced vibration, than like a true instability, in both their mathematical form and their behavior. Indeed, some can even be driven through, i.e. a higher speed can be found where the rotor regains its stability. Sometimes, however, the speed range of instability is quite wide. In the following two simple cases are presented: a) a shaft with unequal stiffnesses in orthogonal directions, and b) a disk with unequal dimetral moments of inertia in orthogonal directions.
7.6.2 Shaft anisotropy
A shaft may have unsymmetrical flexibility due to slots cut in it, such as keyways and armature winding slots, or due to manufacturing imperfections. For shafts with dissimilar stiffnesses in two mutually perpendicular direction (stiffness orthotropy) a range of rotational speeds exists in which the shaft precession is unstable. For such systems with circularly isotropic supports it is more convenient to work in a coordinate system which rotates with the principal axes of the shaft cross section, thus avoiding time varying coefficients in the equations of motion.
7.6.2.1 Free precession of an undamped vertical shaft
Consider the Laval-Jeffcott rotor having a central mass and a cross-section with principal second moments of area I1 and I 2 and rigid bearings (Fig. 7.27, a). The analysis is carried out using a coordinate system , rotating at the constant angular speed of the shaft, , positive in the counter-clockwise direction. It is
208
assumed that the displacements and are parallel to the principal axes of inertia (Fig. 7.27, b). Let the bending stiffness in the direction be k and that in the
direction be k . For the shaft in Fig. 7.27, k < k . The axis along which
the stiffness is a minimum is referred to as the weak axis. Denoting 1 k + k , 2 the principal stiffnesses can be written
k= k = k + k ,
k =
1 k k , 2
(7.98)
k = k k ,
(7.99)
and a degree of dissymmetry can be defined as
k
k
k k k + k
(7.100)
a
Fig. 7.27
If the shaft were isotropic, the equations of motion in stationary coordinates would be (2.3) & + k y = me 2 cos ( t + 0 ) , m& y (7.101) & + k z = me 2 sin ( t + 0 ). m& z Using the coordinate transformation (2.35)
y = cos t sin t , z = sin t + cos t ,
(7.102)
equations (7.101) can be written in rotating coordinates as (2.47)
209
& + ( k m 2 ) = me 2cos , && 2 m m 0 2 2 & & & + ( k m ) = me sin0 . m + 2m
(7.103)
For an orthotropic shaft, equations (7.103) become

& + (k m 2 ) = me 2cos , && 2m m 0 && + 2m & + (k m 2 ) = me 2sin0 . m 7.6.2.1 Free precession. Whirl speeds For a vertical rotor having a central perfectly balanced mass, consider the homogeneous part of equations (7.104)
& + ( 2 2 ) = 0 , && 2
2 && + 2 & + ( 2 ) = 0,
(7.104)
(7.105)
where
2 =
k m
2 =
k m
(7.106)
are bending natural frequencies. With exponential solutions of the form
= 0 e t ,
the equations (7.105) transform into
= 0 e t ,
(7.107)
2 (2 + 2 ) 0 2 0 = 0 , 2 2 0 + (2 + 2 ) 0 = 0.
(7.108)
In order to have non-trivial solutions, the determinant of the displacement coefficients must vanish. This condition yields the characteristic equation
2 2 ( 2 + 2 ) ( 2 + 2 ) + (2 )2 = 0 ,
or
2 2 2 2 4 ( + + 2 2 ) 2 + ( 2 ) ( 2 ) = 0 . (7.109)
Thus
2 2 2 2 1 ,2 = ( + + 2 )
1 2 . 1 2 2 2 2 2 2 2 2 ( + + 2 ) ( ) ( ) 4
(7.110)
210
For the shaft in Fig. 7.27, < . The variation of 2 as a function of
is shown in Fig. 7.28, a.
a
Fig. 7.28 (from [7.23])
When < and < , both roots 2 are negative, so that is purely imaginary,
1 , 1 = i 1 , 1 =
and the rotor is always stable.
2 1 , 2 , 2 = i2 , 2 =
2 2 ,
(7.111)
2 is positive When < < , the root 1
1 , 1 = 1 , 1 =
2 1 ,
(7.112)
such that one of the values of is positive real and the rotor becomes unstable. The variation of 1 , 2 and 1 as a function of is illustrated in Fig. 7.28, b. On the threshold of instability, the real part of is zero and, as 2 is real, the imaginary part of must also be zero. Thus the threshold speeds may be determined by setting = 0 in equation (7.109). They are
c1 = ,
c2 = ,
(7.113)
211
and the rotor is unstable between these two speeds. The unstable motion in this range could have been predicted from equation (7.109) using the Routh-Hurwitz criterion by observing that the coefficient a4 is negative. 7.6.2.2 Whirl speed diagrams Figure 7.29, a gives the natural frequencies R as observed in the rotating coordinate system, as a function of . Figure 7.29, b shows a similar plot of the natural frequencies S relative to the stationary y , z coordinate system. Both figures are computed for a system with a degree of dissymmetry = 0.28 .
a
Fig. 7.29 (from [7.3])
An interesting phenomenon is associated with the point P in Fig. 7.29, b. It defines a whirl component which does not precess ( S = 0) at the rotational speed = n 2 . It could be expected that such a motion could be excited by a stationary excitation such as gravity acting on a horizontal rotor. In Fig. 7.29, a this point is on the line R = . Indeed, a force fixed in the stationary frame is seen as rotating backwards at the speed in the rotating reference frame. This is analyzed in Section 7.6.2.5. The asymptotes are like the whirl speed lines in Fig. 7.7, drawn for a rotor with a rotationally isotropic shaft. Although four quadrants are shown, only the right half-plane (where > 0 ) is required.
212
Fig. 7.30
Fig. 7.31
A whirl speed diagram represented in the right half plane only (for positive ) is shown in Fig. 7.30. In order to amplify the instability region, the diagram is computed for an irealistic degree of dissymmetry = 0.5283 [7.23]. The 45 degree
213
dotted line S = shows the two critical speeds corresponding to the synchronous precession. Often, the whirl speed diagram is represented only in the first quadrant, as in Fig. 7.31. The aliasing of positive and negative frequencies requires great care and some expertise in the interpretation of the diagram. 7.6.2.3 Stability chart Denoting
2 0
2 2 +
(7.114)
the precession is unstable in the region < < or at rotational speeds
0 1 < < 0 1 + .
(7.115)
Figure 7.32 is a stability chart showing the regions of stable and unstable precession as functions of the degree of dissymmetry and the rotational speed . The limit curves are parabolas. The unstable range increases with the degree of dissymmetry.
Fig. 7.32 (after [7.14])
7.6.2.4 Unbalance response In rotating coordinates, the unbalance response without damping is described by equations (7.104). The steady state solutions are
=e
2 cos 0 , 2 2
=e
2 sin 0 . 2 2
(7.116)
The unbalance gives rise to a displacement constant in time, of magnitude
214
= 2 + 2 . In the stationary coordinate system, the disk center has a synchronous precession along a circle of radius r = . This depends on both the speed and the orientation angle of the center of gravity 0 . A more detailed
presentation is given in Section 7.6.2.6 for rotors with external damping. 7.6.2.5 Gravity loading As can be seen from the natural frequency plots (Section 7.6.2.2), the precession of a perfectly balanced rotating asymmetrical shaft exhibits an additional peculiarity when its axis is horizontal. The gravitational force induces a secondary resonance. In stationary coordinates f y = mg ,
fz = 0 .
(7.117)
The transformation of forces from stationary to rotating coordinates is f = f y cos t + f z sin t , f = f y sin t + f z cos t . (7.118)
Substituting (7.117) into (7.118), the gravity loading can be expressed in rotating coordinates as f = m g cos t , f = m g sin t . (7.119)
Including in equations (7.104) the gravitational force m g instead of the unbalance forces we obtain && 2m & + (k m 2 ) = m g cos t , m && + 2m & + (k m 2 ) = m g sin t . m With solutions of the form (7.120)
= G cos t , = G sin t ,
we obtain
2 ( 2 2 ) G 2 2 G = g , 2 2 2G + ( 2 2 ) G = g ,
(7.121)
(7.122)
and G = g
2 4 2 2 2 2 2 2 2 +
(7.123, a)
2 4 2
215
G = g
2 +
(7.123, b)
A resonance will be observed when
1 = 2 2 2 +
2 2 2
= 2, s
(7.124)
that is, at approximately one half of the mean of the two critical speeds.
= ( 1 + )0 , s2
or
=
Indeed, when the two stiffnesses differ only slightly, i.e. = 0 ,

2 2 2 2 0 (1 + ) 2 2 0 ( 1 + 2 ) 0 (1 + ), 4 (1 + ) 4
2 2 2 +
2 1 + (1 + ) 2
1 + . 1 + = 2 2 2 2
(7.125)
Thus a sub-harmonic resonance, caused entirely by the gravitational force, occurs in rotors with asymmetric shafts, known as a secondary critical speed.
Fig. 7.33 (from [7.4])
In the neighborhood of the secondary critical speed, both the numerator and the denominator of the gravity forced precession are of the same order of magnitude. The steady state precession radius depends upon the ratio 2 2 s2 2 . Therefore, the rate of amplitude build-up at this speed is
)(
slow.
216
A diagram showing the whirl amplitude as a function of the rotational speed in a forced precession produced by both out-of-balance and gravity is shown schematically in Fig. 7.33.
7.6.2.6 Dissymmetric shaft with external damping
The following presentation is reproduced from the classical paper by H. D. Taylor [7.24].
Fig. 7.34
Figure 7.34 illustrates the variation of the static deflection of a horizontal shaft, if it is rolled slowly, the deflection being a maximum when the weak axis is vertical (position A) and a minimum when the weak axis is horizontal (position C). For intermediate positions (as at S) the shaft center traverses a circular path through the extreme points, and makes two complete circuits of this path per revolution of the shaft. Figure 7.35 shows the shaft section position at several consecutive times.
Fig. 7.35
217
When the shaft is running, a mechanical unbalance rotating at twice shaft speed will produce a double-frequency vibration of the shaft, with a resonant peak at a shaft speed of about half the average fundamental critical speeds. The unbalance response of a simple rotor with external viscous damping is described in rotating coordinates by equations of the form
&& 2 e 0 && + 2
2 2 & ( 1 ) 2 0 & + 2 e 0 2 e 0
cos 0 = e 2 2 2 ( 1 + ) 0 sin 0 (7.126) 2 e 0 (7.127)
where the damping ratio
e =
ce . 2m 0
The unbalance response curves have different shapes according to the relative values of the degree of dissymmetry with respect to the damping ratio. When > e , the precession radius of the shaft may become infinite, in spite of the presence of damping. For shafts with moderate dissymmetry ( < e ) , there is apparently no infinite precession radius but a number of other peculiarities.
Fig. 7.36
In the case of moderate dissymmetry, a set of unbalance response curves is given in Fig. 7.36. These are drawn for = 0.05 and e = 0.07. Note the importance of a new variable, the orientation angle 0 of the center of gravity with
218
respect to the week axis. The peak response amplitude apparently varies by five to one on a rotor running with the same amount of unbalance over the same speed range, with the same damping and the same degree of dissymmetry and with changes only in the angular position of unbalance. The highest peak appears when the unbalance is 450 ahead of the weak axis, in the direction of rotation. The lowest
peak appears when the unbalance is 450 behind this axis 0 = 450 .
Fig. 7.37
This suggests the use of a polar diagram, Fig. 7.37, to show the variation of the peak amplitude with the angle of unbalance. It appears that this amplitude may vary considerably with the direction of rotation since the most sensitive point for unbalance ( 450 ahead of the weak axis) becomes the least sensitive ( 450 trailing) if the rotation is reversed. In can be shown that the stability condition is
2 2 1 2 + 4 e2 0 . 2 2 0 0 2
(7.128)
219
Fig. 7.38 (after [7.14])
A stability chart is shown in Fig. 7.38, where the limit curves have equations
2 = 1 2 e2 m 2 0
2 1 2 e2
2 0
(7.129)
With appropriate damping levels and especially with a minimum of shaft dissymmetry it seems possible to avoid the unstable operation.
7.6.3 Asymmetric inertias
We shall restrict our study to a light symmetrical elastic shaft rotating with a constant angular speed in identical symmetrical bearings and carrying a centrally mounted rigid disk. As the disk is supported such that its angular motions are uncoupled from the translational motions and it spins about one of its principal axes of inertia, we shall examine the small amplitude oscillations about the other two principal axes.
7.6.3.1 Eulers equations

Assume a rigid disk with its center of mass pivotally mounted at point O, the origin of a stationary set of axes X, Y, Z (Fig. 7.39). It is convenient to use a coordinate system , , which rotates with the disk, in which the moments of inertia and the products of inertia are constant. If this system coincides with the principal axes of inertia of the disk, the products of inertia are zero and the moments of inertia are its principal moments of inertia J1 , J 2 , J 3 .
220
If we let the triad of unit vectors i , j , k of axes , , coincide and rotate with the disk principal axes of inertia 1, 2, 3, the angular momentum about its center of mass is given by
K = J1 1 i + J 2 2 j + J 3 3 k ,
along the principal axes of inertia.
(7.130)
where 1 , 2 , 3 are components of the absolute angular velocity of the disk The angular velocity, , of the triad of unit vectors i , j , k , relative to a stationary coordinate system must be
= 1 i + 2 j + 3 k .
(7.131)
According to the theorem of angular momentum, the external torque is equal to the time rate of change of K M = dK K = + K , dt t i j k (7.132)
& 1i + J 2 & 2 j + J 3 & 3k + 1 M = J1

J1 1
2 J2 2
3 . J3 3
(7.133)
If the torque acting on the disk has components M 1 , M 2 , M 3 , along the principal axes of inertia M =M 1i + M 2 j + M 3 k . (7.134)
Equating the corresponding components from (7.133) and (7.134) we obtain the scalar equations
& 1 ( J 2 J 3 ) 2 3 , M 1 = J1 & 2 ( J 3 J1 ) 3 1 , M 2 = J2 & 3 ( J1 J 2 ) 1 2 , M 3 = J3

known as Eulers dynamical equations. 7.6.3.2 Equations of angular motion
(7.135, a) (7.135, b) (7.135, c)
If the axis is along the axis of rotation (spin axis) and the and axis are along perpendicular diameters, we can introduce the following notations
221
J P = J 1 , J = J 2 , J = J 3 ,
(7.136) (7.137) (7.138)
= 1 = const . , = 2 , = 3 ,
M1 = 0 , M = M 2 , M = M 3 .
Equations (7.135, b) and (7.135, c) become
( ) & ( J P J ) = M . J
& J J P = M , J
(7.139)
Let the angular position of the disk relative to the stationary coordinate system be defined by
= j + k
(7.140)
where and are the components of the angle between the shaft axis and the bearing axis in the rotating system.
Fig. 7.39
The time rate of change of is
&, & j+ & & = j + & k + k

where
& = k = i k = j , & j= j=i j=k , k
(7.141)
so that But
& = ( ) j + ( + )k .
(7.142) (7.143)
& = j + k .
222
The components of the absolute angular velocity along the diametral axes are
& , =
= & + .
Substituting (7.144) into (7.139) yields
(7.144)
( ) ( ) ( & + ) = M , & ) ( J J ) ( & )= M , && + J ( P

&& & J J J P
or
(7.145)
&& + J J J & + 2 J J = M , J P P
( ) ( ) & + 2 ( J J ) = M . && ( J P J J ) J P
M = K , M = K ,
(7.146)
If the angular displacements and are opposed by moments (7.147)
where K is a torsional stiffness, equations (7.146) become
( ) [ ( )] = 0 , & + [K + 2 ( J J )] = 0. && ( J P J J ) J P
&& + J J J & + K + 2 J J J P P 7.6.3.3 Whirl speeds in rotating coordinates
(7.148)
The equations of motion (7.148) can now be rewritten in terms of dimensionless parameters
&& + ( 2 ) & + [ 2 2 ( 1 ) ] = 0 , (1 + ) 0 & + [ 2 2 ( 1 ) ] = 0 , && ( 2) (1 ) 0
(7.149)
where
JP 1 J + J 2
is a gyroscopic coupling factor,
(7.150)
J J J + J
is an inertia inequality factor,
(7.151)
2 0 =
223
K 1 J + J 2
is a reference (fictitious) natural frequency.
(7.152)
With exponential solutions of the form
= 0 e t ,
the equations (7.149) transform into
= 0 e t ,
(7.153)
[ ( 1 + )
2 +0 2 ( 1 ) 0 + ( 2 ) 0 = 0 ,
( 2 ) 0 +
[ ( 1 + )
2 +0
( 1 ) ] 0 = 0.
(7.154)
In order to have non-trivial solutions, the determinant of the angular displacement coefficients must vanish. This condition yields the characteristic equation a14 + 2a22 + a3 = 0 , where (7.155)
a1 = 1 2 ,
2 a2 = 0 + 2 1 2 + ( 2 ) , 2
2 a3 = 0 2 (1 )
(7.156)
] [
2 0
2 (1 + )
The eigenvalues are
1,2 ,3,4 =
a2
2 a1 a3 a2 . a1
(7.157)
The conditions for stability are a1 > 0 , a2 > 0 , a3 > 0 . (7.158)
The first two conditions are satisfied for 0 < <1 and the last condition requires 2 and 0 < 2 , (7.159)
2 0
2 (1 )
] [
2 0
2 (1 + )
> 0.
(7.160)
224
Let denote the critical speeds
1 =
0
1 +
2 =
0
1
(7.161)
Case I. When J P is the largest moment of inertia, 1 + < , the condition (7.160) is satisfied and the system is stable for all . The whirl speeds relative to a coordinate system which rotates with the rotor are represented as a function of the 2 a1 a 3 > 0 , rotational speed in Fig. 7.40, a. It is easy to show that for this case a 2 that is 2 is real and negative. The system is oscillatory.
b
Fig. 7.40
Case II. When J P is the intermediate moment of inertia, 1 < < 1 + , the system is stable at speeds < 1 and unstable for > 1 . If there is one critical speed, motion will be stable below this speed and unstable above it (Fig. 7.40, b). Case III. When J P is the smallest moment of inertia, < 1 , the system is unstable at speeds 1 < < 2 and stable for < 1 and 2 < . If there are two critical speeds, motion will be stable below the lower and above the higher speed, but it will be unstable between the two critical speeds (fig. 7.40, c). For a given , introducing an inertia asymmetry tends to lower 1 and to rise 2 , increasing the speed range of instability. This applies also to overhung rotors, with disks or propellers supported by rigid shafts, pivotally mounted at one end, for which the transverse moments of inertia with respect to the pivot (including a mass times length squared) are larger than the axial mass moment of inertia.
225
Fig. 7.41 (after [7.25])
Figure 7.41 gives a 3D diagram of the stability boundary of undamped rotors with non-circular shafts as a function of the parameters and , and the dimensionless speed 0 [7.25] for Case III. 7.6.3.4 Influence of external damping In the presence of external viscous damping, equations (7.49) become
&& + 2 & + ( 2 ) & + [ 2 2 ( 1 ) ] 2 (1 + ) e 0 e 0 0
= 0,
& + 2 & + 2 + 2 2 ( 1 ) = 0 , && ( 2 ) (1 ) e 0 e 0 0

where
(7.162)
e =
2 0 m
ce
(7.163)
The condition of stability (7.160) becomes
2 0
2 (1 )
] [
2 0
2 ( 1 + ) + (2 e )2 > 0
(7.164)
and the critical speeds (7.161) at the ends of the unstable region become
1,2 = 0
(1
2 e2
4 e2
e2
+ 1 +
.
2
(7.165)
226
As shown in the stability map of Fig. 7.42, with (stationary) external damping the unstable range 2 1 is narrower and shifted at higher rotational
speeds.
Fig. 7.42 (after [7.14])
7.6.4 Finite element analysis of asymmetric rotors
When the rotor is asymmetric and the bearings are axially anisotropic, the equations of motion have periodic coefficients with respect to both a rotating and a stationary reference frame. The latter is used in the following. A general approach based on a variant of Hills infinite determinant method [7.26] is presented for rotors modeled by finite elements. 7.6.4.1 Element matrices in the stationary reference frame In the reference frame rotating with the rotor, the mass matrix of an asymmetric element can be written in the form
[ Mr ]=
m1 [0 ]
[ ]
[m 2]
[0 ]
where [ m1 ] and [ m2 ] are the coherent mass submatrices corresponding to the coordinates in the mutually perpendicular planes 1 and 2, respectively [7.27]. In the stationary reference frame attached to the stator, the mass matrix is obtained by the congruence transformation
[ M s ] = [ R ] [ M r ] [ R ]T
(7.166)
227
in which the orthogonal rotation matrix
[ R ]=
sin t , sin t cos t
cos t
where is the rotational speed of the rotor, assumed to be constant. Matrix (7.166) has the form
[ md ] sin 2 t [ mm ] + [ md ]cos 2 t [M s ] = [ m ] sin 2 t [ mm ] [ md ] cos 2 t d

where
(7.167)
[ mm ] = 1 ( [m1 ] + [m2 ] ),
2 Matrix (7.167) can also be written
[ md ] = 1 ( [m1 ] [ m2 ] ) .
2
[ M s ] = [ M 0 ] + [ M 2 s ] sin 2 t + [ M 2c ] cos 2 t
or
(7.168)
[M s ] = [M 2 ]
where
e i 2 t + [M 0 ] + [M + 2 ] e + i 2 t
(7.169)
[ M 2 ] = 1 ( [ M 2c ] + i [ M 2 s ] ) , [ M + 2 ] = 1 ( [ M 2c ] i [ M 2 s ] ) .
2 2 Analogously, the combined gyroscopic and damping matrix is
[ C s ] = [ C 2 ] e i 2 t + [ C0 ] + [ C + 2 ] e + i 2 t
and the stiffness matrix for an asymmetric shaft is
(7.170)
[ K s ] = [ K 2 ] e i 2 t + [ K 0 ] + [ K + 2 ] e + i 2 t .
(7.171)
The submatrices entering the matrices (7.169)-(7.171) are presented Chapter 5. 7.6.4.2 Solution of the equations of motion with harmonic coefficients The equation of motion of the finite element model containing asymmetric components can be written in matrix form (dropping the index s)
&& } + [ C ( t ) ] { u & } + [ K ( t )] { u } = { F ( t ) } . [ M (t )] { u
(7.172)
In (7.172) matrices are real and periodic, with period T = 2 :
228
[ M ( t )] = [ M ( t + T )] , [ C ( t )] = [ C ( t + T )] , [K ( t ) ] = [ K ( t + T ) ] .
Equation (7.172) can be transformed in a set of differential equations of first degree [7.28] which, when { F ( t ) } = { 0 } , can be written
& ( t ) } [ A (t )] { x (t ) } = { 0 } , {x
where [ A ( t ) ] = [ A ( t + T ) ] .
(7.173)
The differential equations with periodic coefficients have solutions of the form
where r ( k ) (t ) = r ( k ) ( t + T ) eigenvalue
} {
{x
(k )
( t ) }= e k t { r ( k ) ( t ) }
(7.174)
is the eigenvector associated with the complex (7.175)
k = k + i k .
The eigenproblem is solved representing the matrix eigenvectors
{r
(k )
( t ) } by the Fourier series
[ A(t )]
.
and the
[ A ( t )] =
[A m ] ei m t ,
{r
i n t
(k )
(t ) =
m =
} {r }e
(k ) n n =
i n t
(7.176)
Substitution of (7.176) into equation (7.173) yields
n =
{r }e
(k ) n
i n t
in
{r } e
(k ) n
n =
m =
[A m ] ei m t { rn(k ) } ei n t = 0 ,

n =
equation which must be satisfied independently for all frequencies n . Canceling the coefficients of the terms ei n t , for each value n (harmonic balance), leads to a hyper eigenvalue problem (of infinite dimension) of the form
... ... .. . k [I ] . .. . .. .. . ...
[ ]
... A 0 i 2 [I ]
[A+1 ] [A+2 ]
... ... ...
[ ] [A ] i [I ] [A ] [A ] [A ]
0 1 0 +1
... A 1
... [A2 ]
... ...
... ...
[A+2 ]
... ...
[A ] [A ] [A ]+ i [I ]
1 +1 0
[A 2 ]
[A+2 ]
...
[A+1 ]
...
[A ] [A ] [A ]+ i 2 [I ]
2 1 0
...
...
... ... ... ... ... ... ...
{ {r {r {r {r
...
(k ) r 2 (k ) 1
(k ) 0 (k ) +1 (k ) +2
} } } = {0 } } }

...
229
or
( k [ I ] [ B ] ) { q ( k ) } = { 0 } .
(7.177)
Matrix [B ] , with constant elements (time invariant), has an infinite number of eigenvectors (and eigenvalues), which are not all linearly independent. In order to completely describe the homogeneous solution, only 2N eigenvectors (and eigenvalues) are required, where N is the dimension of the matrices of the original system. Indeed
{x
(k )
(t ) = e k t
n = -
{r }e
(k ) n
i n t
= e ( k + i l ) t
n = -
{ rn(k ) } ei (n l ) t .
It follows that
l = k + i l = k + i (k + l ) ,
( l = ..., 3, 2, 1, 0, + 1, + 2, + 3,...)
i l
is also an eigenvalue of the matrix [ B(t ) ] , and the corresponding eigenvector is
{ q }= { q } e
(l ) (k )
so that only 2N eigenvectors are linearly independent [7.28]. In practical applications, the infinite Fourier series is truncated to the first p harmonics, hence to (2 p + 1) terms. This leads to a finite eigenproblem ]) { q ( k ) }= { ( k [ I ] [ B 0}
(7.178)
of dimension 2 N (2 p + 1) , where 2N is the number of degrees of freedom. From the 2 N (2 p + 1) eigenvalues (and eigenvectors) of the matrix B only the basis 2N eigenvalues (and eigenvectors) are calculated [7.28]. Generally, the only major difficulty in the solution of the eigenproblem (7.178) is the inversion of the matrix [ M s ] to obtain the state space form (especially for systems with large dissymmetry).
[ ]
References
7.1 Ehrich, F. F. (ed.), Handbook of Rotordynamics, McGraw Hill, New York, 1992.
230
7.2 Crandall, S. H., Physical explanations of the destabilizing effect of damping in rotating parts, Rotordynamic Instability Problems in High-Performance Turbomachinery, NASA CP 2133, 1980, pp 369-382. 7.3 McCallion, H., Vibration of Linear Mechanical Systems, Longman, London, 1973. 7.4 Lee, C.-W., Vibration Analysis of Rotors, Kluwer Academic Publ., Dordrecht, 1993. 7.5 Muszynska, A., Whirl and whip rotor/bearing stability problems, Instability in Rotating machinery, NASA CP 2409, 1985, pp 155-177. 7.6 Nelson, F. C., A review of the origins and current status of rotor dynamics, Proc. 6th Int. Conference on Rotor Dynamics, Sydney, Australia, Sept 30 Oct 4, 2002, pp 745-751. 7.7 Rao, J. S., Rotor Dynamics, Wiley Eastern Ltd., New Delhi, 1983. 7.8 Sternlicht, B. and Rieger, N. F., Rotor stability, Proceedings of the Institution of Mechanical Engineers., vol.182, Pt.3A, 1967-1968, pp 82-99. 7.9 Crandall, S. H., Heuristic explanation of journal bearing instability, Rotordynamic Instability Problems in High-Performance Turbomachinery, NASA CP 2250, 1982, pp 274-283.
7.10 Sommerfeld, A., Zur hydrodynamischen Theorie der Schmiermittelreibung, Zeitschrift fr Mathematik und Physik, vol.50, 1904, pp 97-155. 7.11 Newkirk, B. L. and Taylor, H. D., Shaft whipping due to oil action in journal bearings, General Electric Review, vol.28, 1925, pp 559-568. 7.12 Robertson, D., Whirling of a journal in a sleeve bearing, Philosophical Magazine, S.7, vol.15, no.96, 1933, pp 113-130. 7.13 Lund, J. W., Some unstable whirl phenomena in rotating machinery, Shock and Vibration Digest, vol.7, no.6, 1975, pp 5-12. 7.14 Gasch, R., and Pftzner, H., Rotordynamik, Springer, Berlin, 1975. 7.15 Fritzen, C. P., and Nordmann, R., Influence of parameter changes to stability behavior of rotors, Rotordynamic Instability Problems in High-Performance Turbomachinery, NASA CP 2250, 1982, pp 284-306. 7.16 Thomas, H.-J., Instabile Eigenschwingungen von Turbinenlufern, angefacht durch die Spaltstrmungen in Stopfbuchsen und Beschaufelungen, AEGSonderdruck Z10/5729, 1958. 7.17 Alford, J. S., Protecting turbomachinery from selfexcited whirl, Journal of Engineering of Power, Trans. ASME, Series A, vol.87, no.10, 1965, pp 333-344. 7.18 Newkirk, B. L., Shaft rubbing, Mechanical Engineering, vol.48, 1926, pp 830832.
231
7.19 Ehehalt, J., Hochlenert, D., Markert, R. and H. I. Weber, Approximate description of backward whirl at rotor-stator-contact, Advances in Vibration Control and Diagnostics (N. Bachschmid and P. Pennacchi, eds), Polimetrica Int. Sci. Publ., Monza, Italy, 2006. 7.20 Bartha, A. R., Dry Friction Backward Whirl of Rotors, Dissertation ETH No. 13817, ETH Zrich, 2000. 7.21 DenHartog, J. P., Mechanical Vibrations, 4th ed., McGraw-Hill, 1956. 7.22 Ehrich, F. F., The dynamic stability of rotor/stator radial rubs in rotating machinery, Journal of Engineering for Industry, Trans.ASME, Series B, vol.91, no.4, 1969, pp 1025-1028. 7.23 Krmer, E., Maschinendynamik, Springer, Berlin, 1984. 7.24 Taylor, H. D., Critical-speed behavior of unsymmetrical shafts, Journal of Applied Mechanics, June 1940, pp A71-A79. 7.25 Brosens, P. J. and Crandall, S. H., Whirling of unsymmetrical rotors, Journal of Applied Mechanics, Trans.ASME, Series E, vol.28, no.3, 1961, pp 355-362. 7.26 Xu, J., Aeroelastik einer Windturbine mit drei gelenkig befestigten Flgeln, Fortschrittsberichte VDI, Reihe 11, No.185, VDI-Verlag, Dsseldorf, 1993. 7.27 Genta, G., Whirling of unsymmetrical rotors. A finite element approach based on complex coordinates, Journal of Sound and Vibration, vol.124, no.1, 1988, pp 27-33. 7.28 Gasch, R. Knothe, K., Strukturdynamik, Springer, Berlin, 1989.
Selected Bibliography
7.29 Biezeno, C. B., and Grammel, R., Technische Dynamik, Bd.3, Springer, Berlin, 1953. 7.30 Bolotin, V. V., Nonconservative Problems of the Theory of Elastic Stability, Pergamon Press, London, 1963. 7.31 Dimentberg, F. M., Flexural Vibrations of Rotating Shafts, Butterworth, London, 1961. 7.32 Gasch, R., Nordmann, R., and Pftzner, H., Rotordynamik, 2. Auflage, Springer, Berlin, 2002.
7.33 Krmer, E., Dynamics of Rotors and Foundations, Springer, Berlin, 1993. 7.34 Thomas, H. J., Thermische Kraftanlagen, 2.Aufl., Springer, Berlin, 1985. 7.35 Tondl, A., Some Problems of Rotor Dynamics, Publishing House of the Czechoslovak Academy of Sciences, Prague, 1965.
232
7.36 Vance, J. M., Rotordynamics of Turbomachinery, Wiley, New York, 1988. 7.37 Crandall, S. H. and Brosens, P. J., On the stability of rotation of a rotor with rotationally unsymmetric inertia and stiffness properties, Journal of Applied Mechanics, Trans.ASME, Series E, Dec 1961, pp 567-570. 7.38 Hori, Y., A theory of oil whip, Journal of Applied Mechanics, Trans.ASME, Series E, vol.26, no.2, 1959, pp 189-198. 7.39 Inagaki, T., Kanki, H., and Shiraki, K., Response analysis of a general asymmetric rotor-bearing system, Journal of Mechanical Design, vol.102, no.1, 1980, pp 147-157. 7.40 Iwatsubo, T., Stability problems on rotor systems, Shock and Vibration Digest, vol.11, no.3, 1979, pp17-26. 7.41 Jei, Y.-G., and Lee, C.-W., Modal characteristics of asymmetrical rotorbearing systems, Journal of Sound and Vibration, vol.162, no.2, 1993, pp 209229. 7.42 Lund, J. W., Stability and damped critical speeds of a flexible rotor in fluidfilm bearings, Journal of Engineering for Industry, Trans.ASME, Series B, vol.96, no.2, 1974, pp 509-517. 7.43 Smith, D. M., The motion of a rotor carried by a flexible shaft in flexible bearings, Proceedings of the Royal Society, London, Series A, vol.142, 1933, pp 92-118. 7.44 Vance, J. M. and Laudadio, F. J., Experimental measurement of Alfords force in axial flow turbomachinery, Rotordynamic Instability Problems in HighPerformance Turbomachinery, NASA CP 2250, 1982, pp 260-273. 7.45 Yamamoto, T., and Ota, H., On the unstable vibrations of a shaft carrying an unsymmetrical rotor, Journal of Applied Mechanics, Sept 1964, pp 515-522.
Index
Annular seals 34 gas 147 liquid 137 Approximate bearing temperature Attitude angle 79 axial groove 112 elliptical 118 floating ring 129 four-lobe 121 multilobe 117 offset halves 116 plain 112 pressure dam 114 multi-lobe 122 Rayleigh step 128 three-lobe 120 tilting pad 124 Fractional frequency whirl 173 Full-Sommerfeld boundary conditions Gravity loading 214 Guyan reduction 69 Gmbel boundary conditions Gyroscopic matrix of rigid disk 9 shaft element 24, 38
107
Bearing 77 clearance 79 dynamic coefficients 83 dimensionless 84 physical significance 103 preload 118 static characteristics 78 static equilibrium locus curve 80 temperature 107 Blacks seal model 141 Bulk flow seal models 139 Campbell diagram 44 Childs seal model 142 Component mode synthesis 62 Concentric circles method 54 Conservative gyroscopic system 46 Constrained normal modes 64 Constraint modes 65 Crandalls model 165, 179 Critical rotor mass 192 Cross-coupled stiffness 162 Damping matrix 34 Dissymmetric shaft 216 Double-frequency vibration Dry friction backward whirl Eccentricity 79 Effective shear area Eigenvalue analysis Elliptic orbit 49 14 40
89
89
Half-frequency whirl 176 Half-Sommerfeld boundary conditions 89 Impeller-diffuser interaction Improved reduction 10 Instability of rotors 161 Internal rotor friction 164 Iterated improved reduction Jenssens seal model Kinetic energy of axi-symmetric disk shaft element 21 Lagrange equations 1 Left eigenvectors 42 Lomakin effect 138 141 7 201
71
217 203
Finite-length cavitated bearing Flexible couplings 35 disk 12 Fluid film bearings 77
99
Macroelements 59 Mass matrix of rigid disk 9 shaft element 23, 38
234
DYNAMICS OF MACHINERY Short bearing 90 Squeeze film dampers 130 design 134 stiffness and damping coefficients 133 Sommerfeld bearing 99, 179 number 80 modified 156 Spectral analysis 48 Stability chart 135, 197, 213, 225 criterion 187 Steam whirl 199 Stepwise model reduction 68 Stiffness matrix of fluid film bearing 34 radial bearing 31 shaft element 26, 38 Strain energy of shaft element 25 Substructure coupling 62 Unbalance response 47 Unsymmetric inertias 219 Viscosity-temperature relationship chart 110 Walthers equation 109 Whirl speed diagrams 171, 211 109
unsymmetric element 226 Mass unbalance vectors 39 of rigid disk 10 of shaft element 28 Modal analysis 47 Model condensation 56 combined static and modal 66 Guyan-Irons 57 modal 60 static 58 Model order reduction 56, 68 Moes solution 99
Nutation 3 Ocvirk bearing 90 Oil whip 174 whirl 174 Onset of instability speed Permanent disk skew vector Planar rotor model 165 Precession backward 3 forward 3
43, 178 10
Reference frames 3 Reynolds boundary conditions 90 equation 87 Right eigenvectors 41 Rotating damping 164 Rotor-stator rubbing 203 Routh-Hurwitz stability conditions 188 Seals 77 annular gas 147 annular liquid 137 bushing 152 floating contact 150 labyrinth 147 locked-up 154 Seal ring lockup 154 Secondary critical speed 215 Shaft anisotropy 207 finite elements 16, 18 Shape functions 19, 21 for rotations 20 shear-modified 20 Shear coefficient 14 energy 25

M. Rades - Dynamics of Machinery 2

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

M. Rades - Dynamics of Machinery 2

Uploaded by

Copyright:

Available Formats

Mircea Rade

5.3 Equations of motion for rotor components

5.4 System equations of motion

5.5 Eigenvalue analysis

5.6 Unbalance response

5.7 Kinematics of elliptic motion

5.8 Model order reduction

6. Fluid film bearings and seals

6.5 Analytical solutions of Reynoldsequation

6.6 Physical significance of the bearing coefficients 6.7 Bearing temperature

6.8 Common fluid film journal bearings

6.9 Squeeze film dampers

6.10 Annular liquid seals

6.11 Annular gas seals 6.12 Floating contact seals

7.3 Whirl in hydrodynamic bearings

7.4 Interaction with fluid flow forces

DYNAMICS OF MACHINERY 203

7.6 Instability due to asymmetric factors

5.1 Rotor component models

5. FEA OF ROTOR-BEARING SYSTEMS

5.2. Kinematics of rigid body precession

5.2.2 Rigid-body precession

5. FEA OF ROTOR-BEARING SYSTEMS

The relation between the unit vectors is

& around Oz1 ) (Fig. 5.5, b) b) Rotation (angular velocity

& = around Ox) (Fig. 5.5, c) c) Rotation (angular velocity

The relationship between the unit vectors is

In expanded form 0 i 1 j1 = 0 cos k 0 sin 1 0 sin

where the transformation matrix

1 sin cos cos + sin [ T ] = cos sin sin cos

The vector of the instantaneous angular velocity is

In terms of the unit vectors of the x,y,z system of coordinates

5. FEA OF ROTOR-BEARING SYSTEMS

& + & cos + & sin = & sin & cos

These will be used in the expression of the disk kinetic energy.

5.3 Equations of motion for rotor components

1 1 1 &2 + w & + & 2 2 + J T & 2 + &2 . & 2 + J P 2 + 2 md v 2 2 2

Neglecting the term

1 & 2 2 , the kinetic energy becomes JP 2

1 1 1 &2 + w & 2 + & 2 + J P & + J P 2 . & 2 + JT md v 2 2 2

In equation (5.4), the term translation,

accounts for rectilinear - for gyroscopic

= md & & = Ty , v d d T T d && J P & = Mz , = JT & dt d d T = md w && = Tz , & d t w

5. FEA OF ROTOR-BEARING SYSTEMS

equation (5.5) becomes

10 Mass unbalance vectors

md as 2 , and along the (inertial) fixed reference system are

The vector of unbalance forces (neglecting disk skewness) can be written

a s 2 ac 2 0 0 cos t + md sin t . (5.9) = md 2 2 ac as 0 0

Permanent disk skew vector

5. FEA OF ROTOR-BEARING SYSTEMS

The components of angular displacements (Fig. 5.8, b) become

The corresponding angular velocities and accelerations are

&&y = && + 2 sin ( t + ) , &&z = && 2 cos ( t + ) .

From equations (5.5, b) and (5.5, d), replacing by y and by z , we obtain

Substitution of (5.10) yields

The vector of gyroscopic moments acting on the disk is

The vector of unbalance forces and moments is

5. FEA OF ROTOR-BEARING SYSTEMS

5.3.2 Uniform shaft elements

and the corresponding vectors of nodal forces

5. FEA OF ROTOR-BEARING SYSTEMS

u = y . x The bending moment is