You are on page 1of 14

Journal of Sound and Vibration 351 (2015) 43–56

Contents lists available at ScienceDirect

Journal of Sound and Vibration


journal homepage: www.elsevier.com/locate/jsvi

Passive vibration control in rotor dynamics: Optimization


of composed support using viscoelastic materials
Eduardo Afonso Ribeiro a, Jucélio Tomás Pereira b, Carlos Alberto Bavastri b,n
a
Federal University of Paraná, Laboratory of Sound and Vibrations, Curitiba, Paraná, Brazil
b
Federal University of Paraná, Sound and Vibration Group, Curitiba, Paraná, Brazil

a r t i c l e i n f o abstract

Article history: One of the major reasons for inserting damping into bearings is that rotating machines are
Received 5 January 2015 often requested in critical functioning conditions having sometimes to function under dynamic
Received in revised form instability or close to critical speeds. Hydrodynamic and magnetic bearings have usually been
9 April 2015
used for this purpose, but they present limitations regarding costs and operation, rendering
Accepted 10 April 2015
the use of viscoelastic supports a feasible solution for vibration control in rotating machines.
Handling Editor: H. Ouyang
Available online 8 May 2015 Most papers in the area use simple analytic or single degree of freedom models for the rotor as
well as classic mechanical models of linear viscoelasticity for the support – like Maxwell,
Kelvin Voigt, Zenner, four-element, GHM models and even frequency independent models –
but they lack the accuracy of fractional models in a large range of frequency and temperature
regarding the same number of coefficients. Even in those works, the need to consider the
addition of degrees of freedom to the support is evident. However, so far no paper has been
published focusing on a methodology to determine the optimal constructive form for any
viscoelastic support in which the rotor is discretized by finite elements associated to an
accurate model for characterizing the viscoelastic material. In general, the support is meant to
be a simple isolation system, and the fact the stiffness matrix is complex and frequency-
temperature dependent – due to its viscoelastic properties – forces the traditional methods to
require an extremely long computing time, thus rendering them too time consuming in an
optimization environment. The present work presents a robust methodology based mainly on
generalized equivalent parameters (GEP) – for an optimal design of viscoelastic supports for
rotating machinery - aiming at minimizing the unbalance frequency response of the system
using a hybrid optimization technique (genetic algorithms and Nelder–Mead method). The
rotor is modeled based on the finite element method using Timoshenko’s thick beam
formulation, and the viscoelastic material is modeled based on four-parameter fractional
derivatives. The insertion of supports – a two-degree-of-freedom isolation system - into rotor’s
motion equations is performed in two different ways: (1) by adding degrees of freedom and
(2) by using the GEP technique. The results show that both techniques are consistent, but the
GEP technique proves to be less time consuming, regarding computing time. In the presented
simulations it is possible to observe the reduction in vibration amplitudes and transmissibility
in a system using optimized viscoelastic supports when compared to ball and hydrodynamic
bearings. One concludes that the methodology presented is robust and allows obtaining an
optimal design of any viscoelastic support – using GEP – in an efficient and viable way for
vibration passive control in rotors of rotating machines.
& 2015 Elsevier Ltd. All rights reserved.

n
Corresponding author. Postal Address: Av. Cel Francisco H. dos Santos, 210, Bloco IV, CEP 81531-970, Curitiba, Paraná, Brazil. Tel.: þ 55 41 3361 3430.
E-mail addresses: eduardo_a_ribeiro@yahoo.com.br (E.A. Ribeiro), jucelio.tomas@ufpr.br (J.T. Pereira), bavastri@ufpr.br (C. Alberto Bavastri).

http://dx.doi.org/10.1016/j.jsv.2015.04.007
0022-460X/& 2015 Elsevier Ltd. All rights reserved.
44 E.A. Ribeiro et al. / Journal of Sound and Vibration 351 (2015) 43–56

Nomenclature keq equivalent dynamic stiffness of the support


Keq global equivalent dynamic stiffness matrix
Ω circular excitation frequency c related to compound system, indicating
Ωrpm circular rotation speed matrix with increased dimension
E complex elasticity modulus M global mass matrix
G complex shear modulus G global gyroscopic matrix
T0 reference temperature K global complex stiffness matrix
T temperature of interest K global stiffness matrix
φ1 ; φ2 material constants y state variable in time domain
α shift function (temperature effect) Y state variable in frequency domain
Ωr reduced circular frequency q generalized coordinates
E0 ; E1 asymptotic storage modulus (fractional deri- Q generalized coordinates in frequency domain
vative parameter, experimentally determined t time
in (N/m2)) f excitation force in time domain
b1 ; β fractional derivative parameters, experimen- F excitation force in frequency domain
tally determined f vector of excitation force in time domain
k complex viscoelastic sheet stiffness of support F vector of excitation force in frequency domain
kb roller bearing stiffness 0 matrix of zeros
mbs bearing shell mass Λ spectral matrix
m support mass/tuning mass Θ right eigenvectors, modal matrix
I; J nodes of support insertion Ψ left eigenvectors, modal matrix
u translation degree of freedom along x I identity matrix
w translation degree of freedom along z min minimum
θ rotation degree of freedom around x f ðxÞ cost function
ψ rotation degree of freedom around z GEP generalized equivalent parameters
Kad stiffness support additional matrix VEM viscoelastic materiaL
Mad mass support additional matrix

1. Introduction

The demand for ever higher power pushes the component and factor dimensions in designing rotating machines to their
limits, which may result in an undesirable dynamic behavior. In many cases, the usual measures regarding geometrical
changes aiming at an increase in stiffness prove to be impracticable, evidencing the real need for an effective and
comprehensive solution for controlling rotor dynamics.
The use of viscoelastic supports seems to be a feasible and adequate solution for controlling rotor dynamics due to its
strong ability to dissipate vibration energy of the rotating system besides generating low initial and operational costs when
compared to more traditional solutions like those using hydrodynamic and active magnetic bearings. Hydrodynamic
bearings present major limitations, according to [1] and [2]; and the magnetic active bearings are very complex in their
practical implementation, according to [3].
The state of the art presents some research regarding the application of viscoelastic supports and bearings in rotor
dynamics.
First of all, it seems important to make a distinction between the concepts of ‘support’ and ‘bearing’ as used in the
present paper. ‘Supports’ add degrees of freedom to the system, whereas ‘bearings’ do no present such addition. Among
recent developments, one can quote [4], which is based on an experimental dynamic analysis of a rotor mounted on
bearings composed of viscoelastic material (VEM). His work evidences a better control of the system’s vibration response
when compared to the conventional bearing used by the author. Also considering viscoelastic bearings, [5] used a frequency
dependent VEM model with four elements associated to a Jeffcot rotor model to optimize the geometry of viscoelastic sheets
in order to reduce the vibration amplitudes of the system. From [6] its presented an approach to vibration control in rotors
using VEM corrugated sheets in ball bearings. In that work, a finite element method is used for the rotor and a four-
parameter fractional derivative model for modeling VEM. The author shows the reduction in vibration amplitudes in
numerical tests, but no optimization is performed in the parameters of the sheets.
In a pioneer work, [7] present the use and optimization of viscoelastic supports as applied to rotors. Through a Jeffcot
rotor model and a frequency constant Voigt model for VEM they concluded that it is possible to optimize the damping factor
aiming to reduce the response amplitudes of the rotor. Similarly, [8] also used a single-degree-of-freedom rotor model, but
for the VEM the authors used a four element frequency dependent model and optimized stiffness and loss factor of the
support VEM to maximize the stability limit and minimize the vibration response of the rotor. Recently, [9] have presented a
E.A. Ribeiro et al. / Journal of Sound and Vibration 351 (2015) 43–56 45

more robust model using beam finite elements with Timoshenko’s formulation for the rotor, but the VEM model is constant
regarding frequency. In that work, the authors conclude that the characteristics of stiffness and damping of the supports
have an influence on the critical speeds of the rotor, and also that those very characteristics considerably reduce the
response amplitudes in critical rotations. In [10], an equivalent formulation for hysteretic damping in rotor dynamics is
presented. In that work, shaft and supports are considered a non-viscous damping model with constant dynamic behavior
for the storage modulus and loss factor. The non-viscous damped model produces an increase in the degrees of freedom of a
rotor system.
However, the classic viscoelastic models used in the above-mentioned works are inaccurate in large ranges of frequency
and temperature variation when compared to fractional models, since their comparison is based on a small number of
coefficients [11–13].
Recently, [14] presented a methodology for modeling rotors based on viscoelastic supports using the transfer matrix
method associated to a four-element fractional derivative. However, that work does not present a methodology for
designing and optimizing the viscoelastic supports.
The authors present a robust methodology based mainly on generalized equivalent parameters (GEP) for an optimal
design of viscoelastic supports aiming at minimizing the frequency unbalance response using a hybrid optimization
technique. The VEM model is based on four-parameter fractional derivatives. The insertion of the supports into the dynamic
model is performed both by adding degrees of freedom and by using the GEP methodology. The numerical results show the
reduction of vibration amplitude and transmissibility in a system that uses optimized viscoelastic supports when compared
to ball and hydrodynamic bearings. The methodology using the GEP allows obtaining an optimal design of any viscoelastic
support in an efficient and viable way for vibration passive control in rotors.
The support proposed in the present work (Fig. 1) is characterized by adding a discrete tuning mass to the system
attached by means of two viscoelastic sheets characterizing double isolation in the system.

2. Viscoelastic support

After Fig. 1 the support is configured symmetrically leading to equal stiffness in the orthogonal directions to the
rotor’s shaft.
The mechanical behavior of the VEM sheets is based on the linear viscoelasticity theory and on a four-parameter
fractional derivative model, which provides it with great accuracy, according to [15].

2.1. Viscoelasticity

From [16], a linear viscoelasticity model that adequately represents a VEM uses a differential equation with fractional
order derivatives and four material parameters. When taken to the frequency domain through the Fourier transform, it
results in the following model for the complex storage modulus:
a0 þ a1 ðiΩÞβ
EðΩÞ ¼ : (1)
b0 þ b1 ðiΩÞβ
From Eq. (1) it is possible to notice its dependence on frequency.
VEMs are also clearly dependent on temperature T. After [17], for simple thermorheologic VEMs, the temperature
influence on dynamic properties (complex modulus) is a horizontal shift in the frequency domain. Thus, the model used to
simulate that displacement is presented by [18], and written as
ðT  T 0 Þ
log αðT Þ ¼ φ1  : (2)
T  T 0 þ φ2

Infinite stiffness Bearing stiffness and


damping

Shaft

Support tuning
mass
Bearing shell
VEM sheets (un-
der shear or com-
pression)

Fig. 1. Viscoelastic support.


46 E.A. Ribeiro et al. / Journal of Sound and Vibration 351 (2015) 43–56

In Eq. (2), φ1 and φ2 are experimentally determined. Now it is possible to introduce the reduced frequency variable, which
concentrates temperature and frequency effects as in
Ωr ¼ ΩαðTÞ: (3)
Finally, in order to obtain the dynamic storage modulus expression versus frequency and temperature, Eq. (1) is considered
with its terms divided by the material parameter b1 . Defining
a0 a1
b0 ¼ 1; ¼ E0 ; ¼ E 1 ;
b0 b1
the final expression of viscoelastic four-parameter model for the complex storage modulus is

E0 þE1 b1 ðiΩαðT ÞÞβ


EðΩ; T Þ ¼ : (4)
1 þ b1 ðiΩαðT ÞÞβ
Likewise, one deduces the complex shear modulus as follows:

G0 þ G1 b1 ðiΩαðT ÞÞβ
GðΩ; T Þ ¼ ; (5)
1 þ b1 ðiΩαðT ÞÞβ
were parameters G0 ; G1 ; β, and b1 are the four parameters also to be determined experimentally, since φ1 , φ2 , and T 0
are known.

2.2. Adding degrees of freedom to support model

A common way of inserting the support effect into the discretized rotor is by adding degrees of freedom to it. Such
increase must be conducted adequately in order to deal only with the degrees of freedom of interest, in this case, just
translations.
Now, let us consider the rotor-support sketch presented in Fig. 2, the rotor discretized into finite elements with n nodes.
Applying Newton’s second law to nodes I and J, grouping the related terms in a matrix shape, the matrices inserted into
the equations of motion can be written as follows:
uI un þ 1 un þ 2
2 3
kb  kb 0
uI
I
¼ 6 7
un þ 1 6 7
Kad kb þkðΩÞI1  kðΩÞI1 (6)
6  kb 7
4 5
un þ 2 0 kðΩÞI1 kðΩÞI1 þ kðΩÞI2

uJ un þ 3 un þ 4
2 3
kb  kb 0
uJ
Kad J
¼ 6 7 (7)
un þ 3 6
6  kb kb þkðΩÞJ1  kðΩÞJ1 7
7
4 5
un þ 4 0 kðΩÞJ1 kðΩÞJ1 þ kðΩÞJ2

where u is the displacement on x direction. The support stiffness symmetry leads to obtaining, analogously, the support
coefficient matrices for direction z. The mass matrices, on the other hand, are presented in Eqs. (8) and (9):
uI un þ 1 un þ 2
2 3
uI 0 0 0
Mad ¼ I 6 0 7 (8)
un þ 1 6
4
0 mbs 7
5
un þ 2 0 0 m1

Rotor

Node I Node J
Node 1 Node

b
bs Node Node bs
J
I
1 Node Node 2
J
I

Fig. 2. Compound system.


E.A. Ribeiro et al. / Journal of Sound and Vibration 351 (2015) 43–56 47

and
uJ un þ 3 un þ 4
2 3
uJ 0 0 0
Mad ¼J
6 0 7 (9)
un þ 3 6
4
0 mbs 7:
5
un þ 4 0 0 m2

This model of support insertion in the system is very useful for it allows determining the transmissibility of the system.
However, when placed in an optimization environment it demands an extremely long computing time. That happens
because supports and rotor are parts of a unique system. Thus, any change in the support characteristics requires a new
survey of the modal parameters, this can be avoided using GEP.

2.3. Support model using GEP

GEP methodology uses the equivalent dynamic stiffness coefficient (Eq. 10) to replace the matrices of Eqs. (6), (7), (8),
and (9). This is performed as in [19] and [20], in a study of viscoelastic dynamic absorbers, and constitutes a very agile
computing technique. In that work, the motion equation of the compound system (Fig. 6b) can be written in terms of the
generalized coordinates of primary system (Fig. 6a) and the use of GEP. Fig. 3 shows the procedures followed to obtain the
dynamic equivalence between the models.
Applying Newton’s second law to the three nodes of Fig. 3, calculating the corresponding Fourier transform, and
evaluating the relation between FðΩÞ=XðΩÞ the equivalent dynamic stiffness, keq ðΩÞ is written as
2
kb
keq ðΩÞ ¼ kb   : (10)
kðΩÞ1 2
kb þ kðΩÞ1  Ω2 mbs þ
kðΩÞ1 þ kðΩÞ2  Ω2 m

Finally, matrix Keq ðΩÞ will be assembled by the coefficients of Eq. (10) and sum to the primary system stiffness matrix (Eq.
(30)) in the appropriate degrees of freedom linked to the support.

3. Solving the global system of equations

The two approaches to support modeling produces two different forms of resolution of the global system of equations.
Because these two forms of resolution are quite different, it is necessary to deal with them separately.
As mentioned in Section 2.2, the system is discretized by the finite element method.
The beam finite element used (Fig. 4) is characterized by showing 3 nodes and 4 degrees of freedom per node. Besides, it
has additional class C0 quadratic interpolation functions. The Timoshenko’s beam formulation takes into account the shear
deformation energy. The stiffness matrix is obtained by the deformation potential energy functional with sub-integration in
order to avoid locking problems. The remaining mass and gyroscopic matrices are obtained through the kinetic energy of
the system.

3.1. Resolution of a composed system from adding degrees of freedom modeling

Starting from the discretized model of the rotor and adding the matrices of the Eqs. (6), (7), (8), and (9), one obtains the
global system of Eq. (11).

Mc q€ c ðtÞ þ Gc ðΩrpm Þq_ c ðtÞ þKc ðΩÞqc ðtÞ ¼ f c ðtÞ (11)

Due to increase in degrees of freedom, the matrices presented in Eq. (11) have a dimension equal to ðd þ 2sÞ  ðd þ 2sÞ, where
d is the initial dimension of the system and s is the number of supports inserted. Vectors q and f, on the other hand, present
dimension ðd þ 2sÞ.

Newton’s second law


b Fourier transform

bs

Fig. 3. Equivalence between support models.


48 E.A. Ribeiro et al. / Journal of Sound and Vibration 351 (2015) 43–56

1 2 2 3 3

1 1 3

Fig. 4. Beam finite element.

The mass global matrix corresponds to M; the gyroscopic global matrix is represented by GðΩrpm Þ; the global matrix of
stiffness – complex and frequency dependent – is represented by KðΩÞ; and vectors q and f represent the generalized
coordinates and the generalized loads of the composed system, respectively.
For the modal parameters calculation, the problem presented in Eq. (11) is taken to a state space. Thus defining the
following state variable
( )
qc ðt Þ
y ðt Þ ¼ _ ; (12)
qc ðt Þ

Eq. (11) results in


"   # " #  
Gc Ωrpm Mc Kc ðΩÞ 0 f c ðt Þ
y_ ðt Þ þ y ðt Þ ¼ : (13)
Mc 0 0 Mc 0

From Eq. (13), taking the homogeneous system, one obtains the generalized eigenvalue and eigenvector problem
together with its adjunct, which is written as follows:

ΛAΘ ¼ BΘ (14)

and
T
ΛAT Ψ ¼ B Ψ ; (15)

where,
" #
Gc ðΩrpm Þ Mc
A¼ (16)
Mc 0

and
" #
Kc ðΩÞ 0
B¼ : (17)
0  Mc

One notices that the dependence of matrices A and B (Eq. (16) and Eq. (17)) regarding rotation frequency Ωrpm and any
excitation frequency Ω renders the eigenvalue problem iterative. Therefore, for each rotation frequency, an auxiliary
Campbell diagram (Fig. 5b) Ωi vs: Ω – where Ωi is the undamped natural frequency – must be performed as presented in [6].
Fig. 5c shows the 3D-Campbell diagram for rotating systems with double dependency matrices (Eq. 16 and Eq. 17). In that
diagram it is possible to note that dependencies on rotation frequency and any excitation frequency – due to the gyroscopic
effect and the dynamic characteristics introduced by the VEM, respectively – act simultaneously in the system.
Having determined the modal parameters, one proceeds to calculate the rotor response in the frequency domain. For that
purpose, a coordinate transformation is applied so that
yðt Þ ¼ Θpðt Þ; (18)

where Θ refers to the matrix of eigenvectors orthonormalized by matrix A: Taking the Fourier transform of Eq. (18), and pre-
multiplying it by Ψ T , one obtains:
 
iΩΨ T AΘ þΨ T BΘ PðΩÞ ¼ Ψ T NðΩÞ; (19)
 
f c ðtÞ
where NðΩÞ is the Fourier transform of vector . Thus, by the orthogonality properties, according to [21], one deduces
0
that

Ψ T AΘ ¼ I (20)

and

Ψ T BΘ ¼ Λ: (21)
E.A. Ribeiro et al. / Journal of Sound and Vibration 351 (2015) 43–56 49

80
80

70
70

60
60
NF [Hz]
NF [Hz]

50
50

40 40

30 30

20 20

6 5 0 0 80 100
4 20 1 2 3 60
Spin S 3 40 4 5 20 40
peed [r 2 1 80 60 Spin Speed [rp 0 z]
pm] x 100 0 cy [Hz] m] x 1000 Frequency [H
0 Frequen

Fig. 5. (a) Campbell diagram, (b) Auxiliary Campbell diagram, (c) 3D Campbell diagram.

Defining DðΩÞ ¼ ðiΩIþ ΛÞ, the frequency response of the global composed system in state space is written as
Y ðΩÞ ¼ ΘDðΩÞ  1 Ψ T NðΩÞ: (22)
From Eq. (12) one concludes that:
( )  
Q ðΩÞ Fc ðΩÞ
¼ ΘDðΩÞ  1 Ψ T : (23)
iΩQ ðΩÞ 0

From Eq. (11) to Eq. (23) one notes the extremely long computing time involved, which renders the use of this foundation
model unfeasible in an optimization environment. Aiming to reduce this computing time, and considering only an
unbalance problem, a technique called Simplified Campbell is used in a way similar to that in [22], which assumes that
Ωrpm ¼ Ω. However, this technique does not change the fact that each modification in support characteristics will require the
generation of new modal parameters.

3.1.1. Simplified Campbell


In Eq. (11), the gyroscopic matrix is defined as G ¼ Ωrpm G1 , where G1 is a frequency independent matrix. In the cases of
unbalance excitation, Ωrpm ¼ Ω, and one writes G ¼ ΩG1 , and the system of Eq. (11) is then written as
Mc q€ c ðt Þ þ ΩG1c q_ c ðt Þ þKc ðΩÞqc ðt Þ ¼ f c ðt Þ: (24)
When taking the expression to the frequency domain, one obtains

 Ω2 ðMc iG1c Þ þ Kc ðΩÞ Q c ðΩÞ ¼ Fc ðΩÞ; (25)
50 E.A. Ribeiro et al. / Journal of Sound and Vibration 351 (2015) 43–56

^ ¼ Mc  iG1c , the eigenvalue problem is given by


and writing M

^ ¼ Kc ðΩÞΘ
ΛMΘ and ^ T Ψ ¼ Kc ðΩÞT Ψ :
ΛM (26)
Hence the composed system response results in

Q c ðΩÞ ¼ ΘDðΩÞ  1 Ψ T Fc ðΩÞ; (27)


where


^ þ Ψ TK
DðΩÞ ¼ iΩΨ T MΘ ^ c ðΩÞΘ : (28)

Finally, it is emphasized that, despite that simplification, any changes in the support characteristics in an optimization
environment will lead to changes in matrices M ^ and Kc ðΩÞ thus making it necessary to recalculate the modal parameters of
the composed system.

3.2. Resolution of composed system from GEP modeling

The advantage of using the GEP model is that it allows describing the dynamics of a composed system (rotorþsupport)
as a function of the generalized coordinates of the primary system only. This allows using the frequency independent modal
parameters of the primary system (Fig. 6a) as a base in n-dimensional space to determine the composed system response
(Fig. 6b). Hence, one must consider the composed system equation expressed in Eq. (29) and the primary system equation
presented in Eq. (30):
€ þ GðΩrpm ÞqðtÞ
MqðtÞ _ þðK þ Keq ðΩÞÞqðtÞ ¼ fðtÞ (29)

and
 
Mq€ ðt Þ þ G Ωrpm q_ ðt Þ þ Kqðt Þ ¼ f ðt Þ; (30)

where matrices have dimension d  d and vectors have dimension d  1d x 1. Eq. (30) represents the rotor with pre-defined
bounds (primary system). Eq. (29) represents the rotor-support composed system from the dynamically equivalent
coefficients presented in Eq. (10).
From Eq. (30) one determines the eigenvalues and eigenvectors. In an attempt to make this support model even more
efficient regarding computing time, the simplification presented in Section 3.1.1 is used. Thus, the eigenvalue problem and
its adjunct are given by
^ ¼ KΘ
ΛMΘ (31)
and

^ T Ψ ¼ KT Ψ :
ΛM (32)
Applying Eq. (18) in Eq. (29) and pre-multiplying it by Ψ , one obtains:
T



^ þΨ T KΘ þ Ψ T Keq ðΩÞΘ PðΩÞ ¼ Ψ T FðΩÞ:
 Ω2 Ψ T MΘ (33)

As the modal matrices are orthonormalized by M, ^ Eq. (33) is reduced to:


 
 Ω2 I þ Λþ Ψ T Keq ðΩÞΘ PðΩÞ ¼ Ψ T FðΩÞ: (34)

Defining
 
DðΩÞ ¼  Ω2 I þ Λ þΨ T Keq ðΩÞΘ ; (35)

it is possible to write the composed system response as

Q ðΩÞ ¼ ΘDðΩÞ  1 Ψ T FðΩÞ: (36)


The Eq. 36 can be determined with quite an adequate accuracy by using part of the eigenvectors and eigenvalues from
matrices Ψ , Θ, and Λ, namely, those lying within the frequency of interest range.

Primary system with pre-defined Compound system with GEP


boundaries

Node I Node J
Node I Node J
Node 1 Node
Node 1 Node

Fig. 6. Primary and compound system through GEP.


E.A. Ribeiro et al. / Journal of Sound and Vibration 351 (2015) 43–56 51

Besides its mathematical simplicity, this method allows introducing expression (36) in an optimization environment in
order to determine the optimal support characteristics from the point of view of the smallest vibratory response in the
rotor; furthermore, it allows calculations to be performed in a feasible amount of time.

4. Optimization

The present work uses a hybrid optimization technique composed of genetic algorithms and nonlinear optimization
based on the Nelder–Mead method, [23]. This approach aims at combining major probabilities of obtaining the global
minimum and reducing calculation speed. Thus the optimization process begin with genetic algorithms to obtain a family of
solutions close to global minimum; later, for finalizing optimization, either the nonlinear optimization technique (TONL) by
Nelder–Mead, were used.
The optimization problem is then defined as

min f ðxÞ: R3 -R; (37)


where x is the design vector,

x ¼ x1; x2; x3 ; (38)

where x1 and x2 correspond to the support sheets thickness, and x3 corresponds to the support discrete mass added to the
composed system. The objective function is the norm of response vector given by Eq. (36) and is defined as

f ðxÞ ¼ ΘDðΩÞ  1 Ψ T FðΩÞ: (39)


Eq. (29) is subjected to inequality constraints.

5. Results

This section comprises three sequences of results: (1) comparison between the models of Sections 2.2 and 2.3: the rotor
geometry used here is a simple didactic model, and the results are presented in terms of unbalance response; (2)
comparison between the numerical results of unbalance response and transmissibility for the rotor with ball bearings, ball

2
10
DOF Addition
GEPs
1
Displ.(um)

10

0
10

1
10
0 5000 10000 15000
Frequency (rpm)

GEPs
DOF Addition
200
Phase (º)

200
0 5000 10000 15000
Frequency (rpm)

Fig. 7. Unbalance frequency response curve – comparison of support models, shaft measure position 0.4(m) (node 4).

Table 1
Shaft geometric and material characteristics.

SHAFT

Length [mm] Diameter [mm] Storage Mod. [GPa] Poisson Density [kg/m3]

1000 50 210 0.3 7850


52 E.A. Ribeiro et al. / Journal of Sound and Vibration 351 (2015) 43–56

Table 2
Disc geometric and material characteristics.

DISC

Position [mm] Ext. Diameter [mm] Thickness [mm] Density [kg/m3]

500 350 30 7850

Table 3
Dynamic load characteristics.

LOADS

Position [mm] Unbalance [gmm] Phase [1]

350 200 0
650 200 90

0.3
0.2
0.1
diameter

0
0.1
0.2
0.3

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1
length

Fig. 8. FE mesh 8 quadratic elements (dimensions in meters).

Table 4
VEM’s characteristics.

VISCOELASTIC MATERIALS

Type G0 [MPa] G1 [MPa] b1 β φ1 φ2 T0 [K]

Neoprene 5.32 148 0.0054 0.359 10.1 137 277.7


Butylic 0.18 241 0.0045 0.424 9.91 119 273

bearings composed by viscoelastic sheets, and optimal viscoelastic supports; (3) comparison between cylindrical
hydrodynamic bearings and optimal viscoelastic supports applied to a high power, two-pole electric engine.

5.1. Degree of freedom addition versus GEP

The characteristics of the rotor used to plot the curves in Fig. 7 are presented in Tables 1–3, and Fig. 8. The VEM
parameters of those used in the simulations are presented in Table 4.
Fig. 7 presents the frequency response curves for the proposed rotor using the two techniques of support insertion
presented in Sections 2.2 and 2.3. As expected, one can observe that both models are dynamically equivalent.

5.2. Support optimization in a didactic rotor

The characteristics of the rotor used in the numerical tests are presented in Table 1, Table 2, and Table 3. The VEMs used
are those presented in Table 4. In the first two simulations, the positions of ball bearings, ball bearings composed by
viscoelastic sheets, and optimal viscoelastic supports are the same, namely, 100[mm]and 900[mm].
In Fig. 8, one can observe the simulated rotor numerical model with its 8 quadratic elements along with the bearing/
support positions.
Fig. 9 presents the comparison regarding frequency response for: ball bearings, ball bearings composed by viscoelastic
sheets, and finally on optimized viscoelastic supports.
E.A. Ribeiro et al. / Journal of Sound and Vibration 351 (2015) 43–56 53

3
10
Roller bearing
Roller bearing + VE sheet
2
10 VE support

Displ. (um)
1
10

0
10

1
10
0 5000 10000 15000
Frequency (rpm)

Roller bearing
Roller bearing + VE sheet
VE support
200
Phase (º)

200
0.0 5000 10000 15000
Frequency (rpm)

Fig. 9. Unbalance frequency response for different types of bearings/support, shaft measure position 0.4(m) (node 4).

Transmissibility 1 and 2 Roller bearing


2 Transmissibility 1 and 2 Roller bearing + VE sheet
10 Transmissibility 1 and 2 VE support
Transmitted force / Applied force

1
10

0
10

1
10

0 5000 10000 15000


Frequency (rpm)

Fig. 10. Transmissibility for different types of bearings/support.

Table 5
Geometric and material characteristics of shaft.

SHAFT  SECTIONS

Length [mm] Diameter [mm] Storage Mod. [GPa] Poisson Density [kg/m3]

170 80 210 0.3 7850


310 100
115 160
1305 190
214 200
139 160
365 100
60 85

The comparison between types of simulated bearings/support regarding transmissibility of the system is presented in
Fig. 10. Transmissibility 1 and 2 represent the transmissibilities for the first and second bearings; and first and second
supports.
54 E.A. Ribeiro et al. / Journal of Sound and Vibration 351 (2015) 43–56

Table 6
Geometric and material characteristics of discs.

DISCS

Position [mm] Ext. diameter [mm] Thickness [mm] Density [kg/m3]

982.5 482.6 245 7850


1227.5
1472.5
1717.5

Table 7
Load characteristics.

LOADS

Position [mm] Unbalance [gmm] Phase [1]

860 313 0
1840 182 0

0.4

0.2
diameter

0.2

0.4

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6
length

Fig. 11. FE mesh with 24 quadratic elements (dimensions in meters).

5.3. Comercial rotor

The characteristics of the rotor used in this test are presented in Tables 5, 6 and 7. VEMs are those presented in Table 4.
This simulated rotor is used in electrical engines with more than 500 ½cv of power and 3600 ½rpm nominal rotation.
Fig. 11 presents a finite element numerical model composed of 24 elements.
In this simulation, the supports are placed on 360½mm and 2380½mm. The types of support to be compared are:
hydrodynamic bearings and optimized viscoelastic supports. The hydrodynamic bearing presents some limitations
regarding rotation frequencies above 10; 000½rpm, from which the pressure field shape of the oil film becomes unstable.
Thus, the frequency ranges used to plot the frequency response curves were limited to 10; 000½rpm. Fig. 12a and b shows
the unbalance frequency response curves for hydrodynamic bearings and optimized viscoelastic supports.

6. Conclusions

The present work presents a methodology for controlling vibrations in rotating machines using viscoelastic supports.
This methodology based on the four-parameter fractional derivative model and GEP allows considering any support made
with VEM, independently of their geometric complexity or degrees of freedom of the support (translational or rotational
ones). In this work the support is composed with tunable intermediate mass.
Two ways of inserting supports into the numerical model of the primary system were presented: a traditional one,
inserting degrees of freedom; and the one proposed in the present work using GEP. The latter allows writing the motion
equation of the compound system as a function of the primary system coordinates. Both are completely equivalent
regarding their dynamic behavior. However, when using the GEP technique, the model renders more agile and robust,
regarding computing time, when inserted in an optimization environment for an optimal design of the support. For
calculating the objective function using GEP, the time spent is approximately four seconds whereas, when using the usual
degree of freedom insertion technique, the amount of time spent is about 20 min (the system used to overcome this setback
E.A. Ribeiro et al. / Journal of Sound and Vibration 351 (2015) 43–56 55

2
10
VE support VE support
1
Journal bearing 10 Journal Bearing
1
Displ. (um)

10

Displ. (um)
0
0
10
10

1 1
10 10

0 5000 10000 0 5000 10000


Frequency (rpm) Frequency (rpm)

VE support VE support
Journal bearing 200 Journal Bearing
200
Phase (º)

Phase (º)
0 0

200 200
0 5000 10000 0 5000 10000
Frequency (rpm) Frequency (rpm)

Fig. 12. Unbalance frequency response curve – comparison between a hydrodynamic bearing and an optimized VE support, shaft measure position 0.86(m)
(node 10), coordinate z (a) and x (b).

is presented in Section 6.3). This reduction is possible only because the eigenvalue problem is solved only once when
using GEP.
In order to generalize the control solution, the vibration control proposal via optimal supports was tested using two
examples: a simple didactic rotor and a real one used in electric engines. As can be observed in the present work, the
viscoelastic supports presented a 97.5% reduction in vibration amplitude when compared to ball bearings, and up to an
86.4% reduction when compared to hydrodynamic bearings. Besides, the proposal of viscoelastic supports attained even a
second objective regarding the transmissibility of the system, which was improved in a consistent way. A reduction in the
transmissibility of the rotor system together with the unbalance response reduction means a longer life cycle for mechanical
components in rotating engines. The numerical simulations show promising results in passive control using supports
composed with viscoelastic materials and tuning mass.
The stability of the system is not studied in the present work, but due to the symmetrical system support and the high
damping introduced through the viscoelastic support, one can hypothesize that the rotor has a large rotational speed band
in stable conditions.

Acknowledgments

E.A. Ribeiro acknowledges the financial support from the Interdisciplinary Program in Petroleum Engineering and
Natural Gas of the Federal University of Parana, PRH24.
C.A. Bavastri acknowledges the financial support from CNPq.

References

[1] A. Muszynska, Stability of whirl and whip in rotor/bearing systems, Journal of Sound and Vibration 127 (1988) 49–64.
[2] C.H. Cloud, E.H. Maslen, L.E. Barret, Rotor stability estimation with competing tilting pad bearing models, Mechanical Systems and Signal Processing 29
(2012) 90–106.
[3] A.S. Das, M.C. Nighil, J.K. Dutt, H. Irretier, Vibration control and stability analysis of rotor-shaft system with electromagnetic exciters, Mechanism and
Machine Theory 43 (2008) 1295–1316.
[4] Y. Lee, T. Kim, C. KIM, N. LEE, D. CHO, Dynamic characteristics of a flexible rotor system supported by a viscoelastic foil bearing (VEFB), Tribology
International 37 (2004) 679–687.
[5] J.K. Dutt, T. Toi, Rotor vibration reduction with polymeric sectors, Journal of Sound and Vibration 262 (2003) 769–793.
[6] C.A. Bavastri, E.M.S. Ferreira, J.J. Espíndola, E.M.O. Lopes, Modeling of dynamic rotors with flexible bearings due to the use of viscoelastic materials,
Journal of the Brazilian Society of Mechanical Sciences and Engineering 30 (2008) 22–29.
[7] R.G. Kirk, E.J. Gunter, The effect of support flexibility and damping on the synchronous response of a single-mass flexible rotor, ASME Journal of
Engineering for Industry 94 (1972) 221–232.
[8] K.C. Panda, J.K. Dutt, Optimum support characteristics for rotor–shaft system with preloaded rolling element bearings, Journal of Sound and Vibration
260 (2003) 731–755.
[9] C.H. Kang, W.C. Hsu, E.K. Lee, T.N. Shiau, Dynamic analysis of gear-rotor system with viscoelastic supports under residual shaft bow effect, Mechanism
and Machine Theory 46 (2011) 264–275.
[10] G Genta, N. Amatti, Hysteretic damping in rotordynamics: An equivalent formulation, Journal of Sound and Vibration 329 (2010) 4772–4784.
56 E.A. Ribeiro et al. / Journal of Sound and Vibration 351 (2015) 43–56

[11] J. Padovan, Y. Guo, General response of viscoelastic systems modelled by fractional operators, Journal of Franklin Institute 325 (1988) 247–275.
[12] F. Mainardi, Fractional Calculus and Waves in Linear Viscoelasticity, Imperial College Press, London, 2010.
[13] L. Sun, L. Chen, Free vibrations of a taul cable with a general viscoelastic damper modeled by fractional derivatives, Journal of Sound and Vibration 335
(2015) 19–33.
[14] P. Varney, I. Green, Rotordynamic analysis using complex transfer matrix: an application to elastomer supports using viscoelastic correspondence
principle, Journal of Sound and Vibration 333 (2014) 6258–6272.
[15] T. Pritz, Analysis of four-parameter fractional derivative model of real solid materials, Journal of Sound and Vibration 8 (1996) 103–115.
[16] R.L. Bagley, P.J. Torvik, A theoretical basis for the application of fractional calculus to viscoelasticity, Journal of Rheology 27 (1983) 201–210.
[17] Y.M. Haddad, Viscoelasticity of Engineering Materials, Chapman & Hall, London, 1995.
[18] J.D. Ferry, Viscoelastic Properties of Polymers, 3rd edition, John Wiley and Sons, New York, 1980.
[19] J.J. Espíndola, H.P. Silva, Modal reduction of vibration by dynamic neutralizers: a generalized approach, Proceedings of the 10th International Modal
Analysis Conference, Vol. 2, 1992, pp. 1367  1373.
[20] J.J. de Espíndola, C.A. Bavastri, E.M.O. Lopes, On the passive control of vibrations with viscoelastic dynamics absorbers of ordinary and pendulum types,
Journal of Franklin Institute 347 (2010) 102–115.
[21] D.J. Ewins, Modal Testing: Theory, Practice, Application, 2nd edition, Research Studies Press, Philadelphia, 2000.
[22] J.J. Espíndola, C.A. Bavastri, An efficient concept of transmissibility for a general equipment isolation system, in: DETC'97/VIB-4120,– Design
Engineering Technical Conferences, ASME, 1997.
[23] H. Rahami, A. Kaveh, M. Aslani, R. Asl Najian, A hybrid modified genetic-Nelder mead simplex algorithm for large scale truss optimization,
International Journal of Optimization in Civil Engineering 1 (2011) 29–46.

You might also like