International Journal of Modern Manufacturing Technologies

ISSN 2067–3604 1

DYNAMIC ANALYSIS OF ROTOR-BEARING SYSTEMS BY OPTIMIZATION

METHOD
Dumitru Nicoara
“Transilvania” University of Brasov-Romania, Department of Mechanics
[email protected]

Abstract: The paper deals with the optimization of the This paper has two principal goals. The first one is to
rotor–bearing systems in static and dynamic range. The find out a code computer program for the calculation
goal of this study is to find out the position of the bearings, of the static and dynamic response of the rotor-
the diameters of the shaft (different diameters for several bearing systems. The program is based on the finite
segments of the shaft) in order to minimize certain cost
element method [1], [5], [3]. The second goal is to
functionals as: static stiffness, amplitudes and “energy” of
the dynamic response, critical rotating speeds and
find out an optimization model of the rotor-bearing
receptance, i.e. the diminishing of the vibrations. Some systems. Based on the modal analysis of the rotor-
constraints are imposed: the maximum stress, the bearing systems, we introduce several objective
minimum diameter, distances between bearings, constant functions, which are a measure of the dynamic
volume (weight) of the shaft, etc. Unbalanced response, i.e. stiffness.
synchronous harmonic excitation as well as asynchronous The goal is to diminish the vibrations and is
harmonic response is analyzed and therefore several implemented by maximizing the dynamic stiffness,
conclusions resulted. i.e., minimizing the receptance function. Therefore,
Key words: rotor system, nonlinear optimization, finite the code computer optimization program is obtained
element, vibrations. by the coupling of the FEM with the non-linear
optimization methods with constraints [4], [11].
1. INTRODUCTION
2. FINITE ELEMENT MODEL OF ROTOR-
The dynamic performances of the rotor-bearing
systems are strongly influenced by the design
BEARING SYSTEMS
parameters such as: distances between the bearings,
2.1 Uniform Shaft Elements
diameters of the different portions of the shaft,
The model consists of a rotor treated as a continuous
bearing preload, bearing spacing etc. In most papers
elastic shaft with several rigid disks, supported on
this influence is studied by varying the parameters
anisotropic elastic bearings. Consider that the
and analyzing of its effect on the system [8], [12], [6].
dynamic equilibrium configuration of the rotor-
The designer needs to have some new techniques for
bearing system the undeformed shaft is along the x-
the prediction of critical speeds, unbalanced
direction of an inertial x, y, z coordinate system. In
responses, and threshold speeds of instability for
the study of the lateral motion of the rotor, the
synchronous and nonsynchronous whirling. Thus, the
displacement of any point is defined by two
translations (v , w) and two rotations (ϕy ,ϕz )
machining performance can be raised remarkably by
improving dynamic stiffness and static stiffness of the
rotor-bearing systems. It is well known that most [10], [5], [2], [3].
vibrations in rotating machinery are induced by The model could use one of the following three beam
rotating related sources. For example, rotating finite element types:
unbalance is the major source of vibration • beam C1 finite element type based on Euler-
synchronous to the rotational speed Ω; misalignment Bernoulli beam model;
and cracks in shafts cause the vibration iΩ (i is an • beam C1 finite element type based on Timoshenko
integer); ball bearings defects cause the vibration nΩ beam model;
(n is a real number); and so forth [8], [3]. As a • beam C0 izoparametric finite element type based
consequence, forced vibration analysis of a rotating on Timoshenko beam model.
equipment subject to synchronous or asynchronous The beam finite element has two nodes. For the static
harmonic excitation becomes essential to identify the analysis, a 2D problem, there are two degrees of
vibration source or to ensure the proper design in freedom (DOF) per node, one displacement
considering vibration problems. perpendicularly on the beam axis and the slope of the
deformed beam. In the case of the dynamic analysis
2
four degrees of freedom (DOF) per node are 2.3 Equations of motion
considered: two displacements and two slopes The equations of motions of anisotropic rotor-bearing
measured in two perpendicular planes containing the systems which consist of a flexible nonuniform
beam. The nodal displacements in the x-y and x-z axisymmetric shaft, rigid disk and anisotropic
planes, and corresponding vector of nodal forces are bearings are first obtained in second order form, by
defined by two 4x1 sub-vector, respectively, assembling the element matrices and may be written
as
Mq
T +(C +ΩG ) q +K q = FT (4)

{} { } { } { }
vi wi yi z i
ϕ − ϕ where q is the
M global displacement vector,− Mwhose
{ qay }= vz  i  ,     { qaz }= wy i  ,  { f aycontains
upper half }= T z ithe  , 
nodal { f az }= inTtheyi
displacements y-x
j j plane, while they  j
lower half contains those inz z-y j
ϕz j − ϕ y j plane, and where M z thej positive definite matrix−MMy isj
(1) mass (inertia) matrix, the skew symmetric matrix G is
gyroscopic matrix, and the nonsymmetric matrices C
We do a comparative study of the three proposed and K are called the damping and the stiffness
models and on its basis we adopt the optimal model matrices, respectively. The matrices of M, C, G, K, q,
of the goal. Thimoshenko beam model is finally and F consist of element matrices given as
adopted as the beam might be short and therefore the
effect of the shear force must be considered [10]. In m 0 c yy c yz 
M =  , C =  ,
the following, only axisymmetric rotors are 0 m  N ×N c zy c zz 
N ×N
considered. The gyroscopic effect and damping in
bearings may be taken into account.
 0 g k yy k yz 
G =  , K =  ,
2.2 Bearings − g 0  N ×N k zy k zz 
N ×N
In the rotor system, the fluid film bearings play an (5)
important role in the dynamic behaviour of the
 f y (t) q ( t ) 
system. Since the squeezed thin film between the F =  , q( t ) =  y 
journal and the ring acts to provide the effects of  f z ( t )  N ×1 q z ( t )  N ×1
spring and damping, the dynamic properties of the
fluid film bearings will dominate the critical speeds, where N = 4n , n is the number of nodes.
the unbalance responses, and threshold speeds of The equation of motion (4) can be rewritten in
instability, which is significantly, and totally different state space form as
from those of the rigid supports. However, these fluid
films are very complicated in operation. When the  +B X =Q
AX (6)
fluid film bearings is analyzed, the physical model of
fluid film bearing may be simplified as a linear where
element [7], [5], [8], [6]. The linearized bearing are
commonly modelled as four spring coefficients and C + ΩG M K 0  F  q 
four damping coefficients, i.e., A =  , B =  , Q =  , X = 
 M 0 0 −M 0  q 

 k yy
b b
k yz  q by  c byy c byz  q by   f yb  The 2 N × 2 N matrices A and B are real but in
 b   +  b   =   general indefinite, nonsymmetric. The resulting
 k zy k zzb  q zb   c zy c zzb  q zb   f zb 
system of equation (6) gives nonself-adjoint
(2) eigenvalue problem. The eigenvalue problem
associated with Eq. (6) and its adjoint are given by

[ ]  0 A ui = B u
[ ] T 
k A vk = B vk , (i , k i =1,2 ,.., 2 N )
b b T
b k λ ib cj
i , λi
0
k i
j = 
i
j
 , c i
j =   , j =
0  0   0 0 
(3) where ui and v i are right and left eigenvectors,
respectively. The eigenvalue problem has 2N
b b
where the superscript b denotes the bearing, cij , k ij eigensolutions, where N is the order of the system
are the linearized directional bearing damping and global matrices. They are purely real for overdamped
stiffness due to j directional motion, respectively, and modes and appear in complex conjugate pairs for
f yb , f zb are the bearing forces in the y-x and z-x underdamped or undamped modes of precession. In
general, the eigenvalues are a function of the rotating
planes, respectively. assembly spin speed and are of the form
 3
λ i = α i ± j ω i , ( i = 1,2 ,.., N ) where u*r and v *r are the upper halves of the
corresponding modal vectors.
The real part αi of the eigenvalue is the damping Equation (8), (9) and (10) describe a modal
constant, and the imaginary part ωi is the damped superposition method that can be used to calculate
natural frequency of whirl (whirl speed or precession receptance functions. From these equations, one
speed). element of system receptance matrix can be written
Next, let us introduce the 2 N × 2 N matrices of right as
and left eigenvectors
2N
u ir v kr Aq i br
Rik = ∑ = , λr = −
U = [ u1 u2 ..... u2 N ] , V = [ v1 v 2 ..... v 2 N ] r =1 a r ( jΩ − λ r ) AFk ar
(12)

The ( 2 N ×1 ) right and left eigenvectors satisfy the The author elaborated several computer codes in
bi-ortogonality relations, written in compact matrix MATLAB programming language. Optimization
form toolbox is used to perform the computer programs,
which have very good performances in order to study
   
complex rotor with two or more bearings and with
V AV =   V BV = 
T T 
 a  ,  b  several rigid disks.

 
 
 
 The FEM model has been theoretically validated by
the recalculation of some models recognized in the
In the case of the synchronous excitation, due to specialized literature, [6], presented in the next
unbalance section, and by means of experimental measurements
F = AF e j Ωt , q = Aq e j Ωt which were performed in the dynamic range [10].

transforming Eq. (6) into frequency domain, we 3. OPTIMIZATION MODEL OF ROTOR-

obtain BEARINGS SYSTEMS
 Aq   Aq  In this section, based on the modal analysis we
A X = Rd AQ , AX =   , AQ =  
 jΩ Aq  0  propose an “external” (passive) optimization model
(7) for rotor-bearing systems. The goal is the diminishing
the vibrations by the maximizing of the dynamic
where the matrix Rd is recepance matrix stiffness. To do this we need to find out the design
parameters: the position of the bearings, the
diameters of the shaft (different diameters for several
R d = ( jΩ A+ B )
−1
, ( j = −1 ) segment of the shaft) in order to minimize certain
(8) coast functionals. In the dynamic case, the objective
functions may be chosen depending on the shaft
By matrix operational transform, the receptance excitation: synchronous or asynchronous excitation.
becomes In the case of rotating unbalance (synchronous
excitation) we introduce three types of the objective
Rd ( Ω) = U ( jΩ a + b ) V T
−1
function:
(9) • the receptance for a given rotating speed
Au AF ;
Next, let us introduce the dynamic stiffness matrix • the average receptance for an interval of rotating
K d , defined as the inverse of receptance matrix Ω2
1 Au
speeds:
Ω1 − Ω2 ∫A dΩ ;
K d = Rd
−1
( Ω) . Ω1 F

(10) • the lowest critical whirl speed or precession

speed (with or without the gyroscopic effect):
From the Eq. (7) and (9) we obtain ωcr .
The optimization problems obtained in these three
T
2N
u*r v *r cases are represented in Eq. (13, a,b,c).
Aq = ∑ AF
r =1 jΩa r + br
(11)
v rT A ur = a r , v rT Bur = br
4
(a) become 3.6Ω and 5.4Ω . In this case the objective
Au function is the “energy” E of the response
min
sk ,dk AF Tmax
1 v 
∫ {u} {u} dt , {u} = w
T
s ki ≤ sk ≤ s ks E=
Tmax 0
d ki ≤ d k ≤ d ks
(15)
Ω = Ω0
V = const . where Tmax is the integration total time corresponding
(b ) to 50…100 shaft rotations, {u} is displacement
Ω2 vector. In this case it is necessary to compute the
1 Au dynamic response of the rotor–bearing system and the
min
sk ,d k Ω1 − Ω2 ∫A
Ω1 F
dΩ
θ -Wilson step by step integration method was used
(13) [2]. The corresponding optimization problem is
s ki ≤ s k ≤ s ks
d ki ≤ d k ≤ d ks 1
Tmax
v 
∫ {u} { u} dt , { u} = w
T
Ω∈( Ω1 , Ω2 )
min
s Tmax 0
V = const . si ≤ s ≤ s s ;
and di ≤d ≤ds
(c ) (16)
max ω cr
sk , dk Some constraints are imposed:
s ki ≤ sk ≤ s ks (13) • the maximum stress;
• the minimum diameter;
d ki ≤ d k ≤ d ks
• distances between bearings, constant
V = const . volume (weight) of the shaft, etc.
Unbalanced response (synchronous harmonic
In the above equations Au is the amplitude of the excitation) as well as asynchronous harmonic
displacement, AF is the force amplitude, Ω is the response are analyzed and therefore several
rotor spin speed and ω is the whirl speed. The first conclusions resulted.
two objective functions (a) and (b) lead to a pseudo-
static problem, the response is found out by solving a 4. NUMERICAL EXAMPLES
linear system of equation, in complex if damping In this section we shall consider two numerical
and/or gyroscopic affect are considered. The last examples, for two rotor-bearings systems.
objective function introduces an eigenvalue problem. The Campbell diagrams and logarithmic decrement
All objective functions are the measure of dynamic obtained by the computer code written by the authors
stiffness defined by relation (10). is further presented.
The asynchronous harmonic excitation is mainly due Example 1 [6] We consider the system from Fig. 1
to defects in bearing rollers and/or rings. For the with rotor data from table 1. In Fig. 2 the Campbell
asynchronous excitation the outer ring defect was diagram and logarithmic decrement of the simple
approximated, [4], by the relation (14a) rotor supported by ball bearings are given.

fy cos ψ 0  Table 1. Rotor data

  = fe   cos ( 3.6 Ω t ) (a) Shaft Disk Bearings
 fz  sin ψ 0  E = 2e15 N/m2 m = 30 Kg kyy = 0.50e7 N/m
(B1)
(14) ρ = 103 Kg/m3 JT = 1.2 Kg m2 kzz = 0.70e7 N/m
fy  cos Ω t  (B1)
  = fi   cos ( 5.4 Ω t ) (b) d = 10 mm JP = 1.8 Kg m2 kyy = 0.50e7 N/m
fz  sin Ω t  (B2)
L = s+a kzz = 0.70e7 N/m
=(0.3+0.4) m (B2)
where Ψ0 is the angular position of the defect and
Ω is the rotational speed of the shaft. The inner ring
defect was approximated by the relation (14b) where
y
fi denotes the excitation force amplitude. In the case 2
1
of nr = 9, nr being the number of rolling elements in d x
the bearing, the fundamental excitation frequencies

s a
L
 5
minimized. In this case we have: Ω∈( 0,9000 )
[rpm], L = a + s = const . , V = const .
(OP3) In this problem the design parameter are the
Fig. 1: Rotor configuration distance between bearings and the diameters of the
4
x 10
3.5

3
two shaft segments. The objective function is average
receptance.
2.5
Natural Frequency [RPM]

4
x 10
2 16
14
1.5
12

Natural Frequency[RPM]
1 10
0.5 8
0 6
0 0.5 1 1.5 2
4
Rotational Speed [RPM] x 104
2
0
Fig. 2.(a) Natural frequency (Campbell diagram) 0 0.5 1 1.5 2 2.5 3 3.5
4
x 10
Rotational Speed [RPM]

0 .1 2
Fig. 4(a) Natural frequency
0 .1 1

0 .1
0.7
0 .0 9
Logaritmic Dec

0.6
0 .0 8
0.5
Logaritmic Decrement

0 .0 7
0.4
0 .0 6
0.3
0 .0 5
0 0 .5 1 1 .5 2 0.2
4
R otational S p eed [R P Mx] 1 0
0.1
Fig. 2. (b) Logarithmic decrement 0
0 0.5 1 1.5 2 2.5 3.5 3
4
Rotational Speed [RPM] x 10
Example 2
We consider the rotor configuration from Fig. 3. For
the case of the linearized bearings model and Fig. 4 (b) Logarithmic decrement
Timoshenko beam model, we shall define and resolve
three optimization problems: Table 2 shows the rotor data. Tables 3, 4 and 5 show
y the optimal value of the parameter s, for the first
2 3 optimization problem, obtained with the three
1 d1 x d2 objective functions: receptance, average receptance
and the lowest whirl speed. It can be noticed that the
values of the resulting design parameters are very
close for the objective functions used in this work.
s a For the second problem, the optimization result had
Fig. 3. Rotor configuration
been calculated with the cost functional the average
receptance for s = sopt, determined in the first problem
(OP1) Find out the optimal value sopt of the distance s and under the following conditions: L = sopt+ a =
between the bearings so that the dynamic stiffness const., Ω∈ (0, 9000) [rpm]. The results are shown in
(calculated in disk station) is maximum. Table 6.
(OP2) Given s = sopt, already determined in the For the third problem the results obtained with the
previous problem, find out the optimum diameters d1 average receptance as objective function are shown in
and d2 so that the average receptance to be Table 7.
6
We notice that dynamic stiffness has considerably maximization of the static and dynamics stiffness of
increased with over 100% for an average increasing the systems, i.e. the diminishing of the vibrations.
of the diameters with 11%. The method is very useful for the design engineers
The ratio between the optimal distance between the from the very beginning of the design, offering to the
bearings and the length a of the console is sopt / a = designer the optimal values of the parameters.
2,6. For a given volume of the shaft, the maximization of
the dynamic stiffness was done, considering two
Table 2. Rotor data – numerical example cases: (I) the diameters d1, d2, given and the design
Shaft Disk Bearings
parameter is s, the distance between the bearings, and
E = 2.068e11 m = 75 Kg Stations 1 and 2
N/m2 JT = 0.190 Kg m2 kyy = 5e8 N/m; (II), the design parameters are in the same time the
ρ = 7833 Kg/m3 JP = 0.368 Kg m2 kyz = kzy = 0 distance between the bearings as well as the
d1 = d2 = 0.08 m e = 0.01 m kzz = 3e8 N/m diameters d1, d2. In the second case, by distributing
a = 0.1 m cyy = czz = 1e4 the diameters by optimization program, the dynamic
Ns/m
czy = cyz = 0
stiffness increased by 75 % comparing to the first
case. The results of optimizations are strongly
Table 3. Optimal value of the distance s influenced by the bearing stiffness. The beam model
objective function – receptance is very important, too. Correct results are obtained
Objective function – receptance using Timoshenko beam model as the distance
Ω [rpm] sopt [m] Receptance between bearings may become small and the
[m/N] influence of the shear force may not be neglected.
1 2000 0,3374 1,26*10 – 8
This aspect is more important for statically
2 4000 0,3404 1,37*10 – 8
3 6000 0,3467 1,64*10 - 8 indeterminate systems because in this case the values
4 8000 0,3591 2,30*10 - 8 of the reactions, and therefore the values of efforts
5 9000 0,3615 3,32*10 - 8 and displacements, depends on the stiffness of the
beam. The dynamic performances of the rotor-
Table 4. Optimal value of the distance s bearing systems are strongly influenced by the design
objective function – average receptance parameters such as: distances between the bearings,
Objective function - average receptance, d1 =d2 = 0,08
[m]
diameters of the different portions of the shaft,
Ω [rot/min] sopt [m] Average receptance bearing preload, bearing spacing etc. In most papers
[m/N] this influence is studied by varying the parameters
1 Ω∈ (0, 9000) 0,3507 1,65*10 – 8 and analysing of its effect on the system. In this paper
2 Ω∈ (0, 15000) 0,3 3,57*10 – 8 we propose a new method for optimal determination
of these parameters using the optimization principles.
Table 5. Objective function – the lowest whirl speed In order to do this we introduce four types of
Objective function – the lowest whirl speed – no
gyroscopic effect objective functions. The choice of these objective
sopt [m] ω 1cr [rpm] functions depends on the type of the excitation:
0,3523 10.553 synchronous or asynchronous. The goal of the
Objective function – the lowest whirl speed – whit optimization is the determination of the design
gyroscopic effect parameters so as the dynamic stiffness to be
0,3193 10.057 maximized.
Table 6. Optimization results – second problem
Objective function - average receptance, L = sopt+a = 11. REFERENCES
const., sopt = 0,3507
Ω d1[m] d2[m] Average receptance [m/N] 1. Akella, S., (1986). Modification to a Timoshenko
[rpm] Beam-shaft Finite Element to Include Internal Disk
Ω∈ ( 0,0796 0,0811 1,65*10 – 8 and Change in Cross- section, Journal of Sound and
0,
Vibration, 106, p.227-239.
9000)
2. Brandon, J.A., (1992). AL-Sharreef, K.J.H.,:
Table 7. Optimization results for the third problem Optimization Strategies for Machine Tool Spindle-
Objective function - average receptance, Ω∈ (0, 15.000) Bearing Systems: A Critical Review, ASME Journal
[rpm], of Engineering for Industry, Vol.114, p. 244-253.
s [m] d1[m] d2[m] Average receptance 3. Childs, D., (1993). Turbomachinery
[m/N]
Rotordynamics, John Wiley & Sons, New York.
0,2616 0,0889 0,0903 1,45*10 - 8
4. Gill, P.E., Murray, W., Wright, M.M., (1991)
Numerical Linear Algebra and Optimization, Addison
5. CONCLUSIONS Wesley.
5. Hughes, J.R.T., (1987) The Finite Element
In this paper we proposed several optimization model Method, Prentice –Hall Inc..
for rotor-bearing systems. These models allows the
 7
6. Jei, Y. Kim., J., (1993). Modal Testing Theory
of Rotor-Bearing Systems, Journal of Vibration
and Acoustic, 115: 165-176.
7. Krämer, E., (1993) Dynamics of Rotors and
Foundations. Springer Verlag, Berlin.
8. Lee, C.W., S.W. Hong, S.W., (1990)
Asynchronous Harmonic Response Analysis of
Rotor Bearing Systems, The International Journal
of Analytical and Experimental Modal Analysis
5(2): 51-65.
9. MATLAB: Reference Guide, The MathWorks Inc.,
Natick, Mass., USA, 1993.
10. Nicoară, D.D.,Munteanu, Gh.M., (1999).
Contributions on the Optimization of Rotor –
Bearing Systems. Anspruch und Tendenzen in
der experimentellen Strukturmechanik, VDI-
Gesellschaft Mess-und Automatisierungstechnik,
369-375, VDI Verl., Düsseldorf.
11. Vanderplaats, S.V., (1984) Numerical
Optimization Tehniques for Engineering Design:
with Applications , McGraw-Hill, New York.
12. Wang, W.R., Chang., C.N., (1994) Dynamic
Analysis and Design of a Machine Tool Spindle-
Bearing System, ASME Journal of Vibration and
Acoustic, 116: 280-285.