Professional Documents
Culture Documents
a r t i c l e in fo
abstract
Article history:
Received 18 August 2009
Received in revised form
18 November 2009
Accepted 22 December 2009
Available online 7 January 2010
Keywords:
Modal analysis
Gear dynamics
Parametric vibration
Time-variant systems
1. Introduction
Modal analysis method is very frequently used with the aim of solving time-invariant linear equations of motion. The
eigenvalues and eigenvectors obtained from the solution of eigenvalue problems allow engineers to interpret and visualize
the dynamic behavior of different mechanical systems. For time-variant equations of motion the use of modal analysis in
its well-known form is not possible. In this case a stability analysis normally relies on Floquet theory. It does not deliver the
complete homogenous solution, but gives only information about the stability of the system represented by equations of
motion with time-variant coefcients [1].
In many cases as exible rotating blades, exible rotating discs, rotating shafts with non-symmetrical cross section, and
gear dynamics, the coefcients of the equations of motion vary in a periodic way. In Xu and Gasch [2] the complete
homogenous solution of periodic time-variant linear equations of motions is presented, based on Hills approximation. The
periodic time-variant matrix systems are expanded in Fourier series. Assuming that the solution also can be expanded in
Fourier series, it is possible to obtain a general homogenous solution by solving a hyper-eigenvalue problem, i.e. when the
number of Fourier coefcients is not innite. The solution of the hyper-eigenvalue problem leads also to eigenvalues and
eigenvectors, nevertheless, the eigenvectors become also periodic time-variant and the eigenvalues become dependent on
the periodicity of the parameter variation. Different contributions to the problem of periodic time-variant modal analysis
are presented in the literature, i.e. non-symmetric rotors [1,3], exible rotating discs [4,5], rotor-blade dynamics [2,6] and
ARTICLE IN PRESS
1496
later active control of rotor-blade dynamics [710]. The mathematical foundations of modal analysis for time-varying
linear systems is clearly and very nicely presented by Irretier in [11], and Bucher and Ewins in [12].
In [1] the dynamics of a simple exible shaft with non-symmetric cross section, supported by anisotropic bearings, is
theoretically investigated using Hills approach. In [3] such an investigation is carried out theoretically as well as
experimentally.
Flexible rotating discs are also an example of period time-variant structure. Their dynamics are carefully investigated
using periodic-time variant modal analysis in [4,5]. The theoretical work presented in [4], and the experimental validation
in [5], illustrates the continuation of Irretiers work [13].
Rotating exible blades are also an example of a periodic time-variant system. Their dynamics are also carefully
investigated using periodic time-variant modal analysis, as it can be seen in [2,14,6,15]. In [15] a contribution to the
experimental validation of linear and non-linear dynamic models for representing rotor-blades parametric coupled
vibrations is given. The rotor-blade dynamics is described by using three models with different levels of complexity
followed by experimental validation of such models. A deeper physical understanding of the dynamic coupling and the
behavior of the parametric vibrations are achieved. Such an understanding is of fundamental importance while developing
active control strategies. In [7] the design of time-variant modal controllers is in focus. Time-variant modal analysis and
modern control theory are integrated in an elegant way allowing the development of new control strategies. The modal
controllability and observability of bladed discs are strongly dependent on the angular velocity, a detailed analysis of such
a dependency is presented in [8]. To control rotor and blade vibration using only shaft actuation is a very challenging
problem. In [9] such a problem is investigated theoretically as well as experimentally using different control strategies. The
electromagnetic actuators are used to control a horizontal rotor-blade system (blades periodically excited by the gravity).
In [10] the same problem is theoretically as well as experimentally investigated and new strategies are developed to
control vertical rotor-blade systems.
The existing dynamic gear models can be classied based on the excitation source: transmission error (TE)-excited
models and parametrically excited models. As shown by Blankenship and Singh [16], the TE and the mesh stiffness depend
on each other, making the dynamic gear mesh modeling very complicated. To simplify the calculations, it is common to use
the TE as the only external excitation source, and/or use the varying stiffness as a parametric excitation. The validity of this
procedure is examined by Velex and Ajmi [17], who mention the problem of dening TE for helical gears as a limitation.
Also the assumption of quasi-stationarity often used in TE-excited models is a disadvantage, as it excludes the possibility of
using the model to correctly predict dynamic behavior in a resonant region. This problem is then partly solved by
Kahraman and Singh in [1821], who use the methods of non-linear dynamics for solving the equations of motion in a
resonant region of a gear pair with a clearance-type non-linearity. The effect of the non-linearity is amplied by
introducing the varying mesh stiffness.
In the non-TE-excited models, no knowledge about the TE is needed before the calculation is performed. Peeken et al.
[2225] present models which are parametrically excited. It is shown that the equation of motion for the torsional 1-DOF
(degree-of-freedom) system reduces to the Mathieu equation, when the gear mesh stiffness is replaced by a cosine
function. More complicated mesh stiffness functions are introduced via the lowest terms of their Fourier series. In [26],
Velex and Berthe solve the equations of motion after splitting these in a stationary part (results from the mean external
load) and a dynamic part (from the varying part of the external load). The mesh stiffness is described by a Fourier series in
the time domain. Velex and Maatar [27] and Ajmi and Velex [28] extend the model by further investigations concerning
the mesh stiffness.
The goal of this work is to apply the theory of time-variant modal analysis to spur gear dynamics. The advantages and
drawbacks of using the technique for analyzing vibrations in spur gears are investigated.
2. Mathematical modeling
2.1. Gear mesh stiffness
Finding the gear mesh stiffness is an iterative process, which involves calculation of the load distribution across the
tooth and the load sharing between the contacting tooth pairs. The gear mesh model is shown schematicallyand slightly
simpliedin Fig. 1(a). In this example, three tooth pairs are taken into account, marked by 1, 2, and 3. Each tooth pair is
divided into three sections in the longitudinal direction of the tooth, as shown in Fig. 1(b). The elastic coupling between
sections within one tooth is handled by applying beam theory to simplify the plate deection problem, as shown by
Schmidt [32]. The goal for the stiffness model is to nd the stiffness matrices KHertz for the Hertzian deformation, and Ktooth
for the bending and shear in the gear tooth. For a low load P as shown in Fig. 1(a), only one section is in contact. With
increasing load, more sections will be in contact with the mating gear ank, and thus contribute to the gear mesh stiffness.
The stiffness model is based on the true involute tooth form, which is calculated from basic gear data using the model
described by Padieth [29]. The stiffness calculations are performed using the methods of Weber and Banaschek [30], Ziegler
[31], and Schmidt [32]. Following these references, the gear body deformation is included in Ktooth by regarding the gear
body as an elastic, semi-innite space. When applying a force to a tooth, the resulting stresses are transferred to the
underlying gear body, which is assumed to deform in a zone that stretches 23 tooth widths from the loaded tooth.
ARTICLE IN PRESS
R. Pedersen et al. / Mechanical Systems and Signal Processing 24 (2010) 14951508
1497
Fig. 1. (a) Gear mesh stiffness model; (b) tooth pair discretization.
The inclusion of grinding corrections is an important aspect of the contact modeling. In Fig. 1(a), the topology to the
right shows the grinding corrections. These will affect the number of contacting sections for a given load, which will then
affect the gear mesh stiffness.
The gear mesh stiffness is calculated for several relative positions of the meshing gears. This gives a load- and position
dependent stiffness. Under the high loads considered in the following dynamic analysis, it is assumed that tooth separation
will not occur. After completing the stiffness calculation, a constant rotational speed is assumed. By these assumptions, the
gear mesh stiffness becomes a periodically varying parameter in the time domain, that is, the non-linearity is removed. It
must also be noted that all teeth are considered identical, i.e. pitch errors are not included. The resulting system can be
investigated using the theory for modal analysis of linear time-variant systems.
2.2. Modal analysis of time-variant system
When the only time-variant parameter in the system is the gear mesh stiffness, the equation of motion can be written
as
Mq Cq_ Ktq f
{
"
# "
0
q_
M1 Kt
q
#"
I
M
1
q
q_
0
0
0
M1
0
f
or
"
_
zAtz
p;
q
z _
q
#
3
The rst part of the analysis of Eq. (3) is to nd the solutions to the homogeneous part of the equation, that is, solve
_
zAtz
0 for the unknown z. This is done by expanding the theory of modal analysis of time-invariant systems [33] to
include periodically time-varying parameters, as shown in [6,2]. By the substitutions zt rtelt and p 0, the
homogeneous part of Eq. (3) can be written as the eigenvalue problem
r_ t lIAtrt 0
Under the assumption of constant rotational speed, the gear mesh frequency in units rad/s is called O. The stiffness matrix
Kt and therefore the state matrix At are both periodic with period T 2p=O and can be expanded into innite, complex
Fourier series.pAlso
the time-variant eigenvector rj t, belonging to the j th eigenvalue, is assumed to be periodic. In all
equations, i 1:
Kt
1
X
Kk eikOt
Aa eiaOt
k 1
At
1
X
a 1
rj t
1
X
r 1
rj;r eirOt
for j 1 . . . 2N
ARTICLE IN PRESS
1498
It can be shown that the components of the eigenvectors rj t can be found by re-writing Eq. (4):
3
3 2
2
7 6 0 7
6r
7
6 j;2 7 6
7
7 6
6
6 rj;1 7 6 0 7
7
7 6
6
7
7 6
^ 6
^ A
lj I
6 rj;0 7 6 0 7
7
7 6
6
6 rj;1 7 6 0 7
7
7 6
6
7
7 6
6
4 rj;2 5 4 0 5
where
2
&
6
6
6
6
6
6
A^ 6
6
6
6
6
4
2iOI A0
A1
A1
iOI A0
A2
A1
A3
A2
A4
A3
A2
A1
A0
A1
A2
A3
A2
A1
iOI A0
A1
7
7
7
7
7
7
7
7
7
7
7
5
&
A4
A3
A2
A1
2iOI A0
In an exact solution to Eq. (8), A^ is innitely large. However, the magnitude of the Fourier components of At depend directly
on the Fourier components of Kt. Therefore, if the gear mesh stiffness function included in Kt is smooth, only a few of the
Fourier components of At will have a magnitude that will signicantly inuence the resulting displacements and velocities
in the vector zt. In a later section, it will be shown how to calculate the necessary number of Fourier components to be
^ in order to obtain accurate results. If the number of included Fourier components is n, in the sense that
included in A,
Kt
n
X
Kk eikOt
10
k n
and the number of degrees of freedom in the model is N, the size of matrix A^ will be 2N2n 1 2N2n 1.
There is a great amount of redundant information in the solution of Eq. (8). Only the basis eigenvalues and basis
eigenvectors, which are identied as described in a later section, are needed in the further analysis. The 2N basis
eigenvalues are stored in a diagonal matrix K, and the Fourier components of the basis eigenvectors rj;0 are stored in the
three-dimensional array R. This array can be visualized as 2n 1 layers of matrices of size 2N 2N. The k th layer of R
will have the structure Rk r1;k r2;k r2N1;k r2N;k for k n; . . . ; n
To nd the solution q of Eq. (1), also the left eigenvectors Lt, which are the solution to the equation RtLt I are
needed. These can be found by solving the matrix equation:
RtLt I
{
1
X
Rr eirOt
r 1
{
2
6
6
6
6
6
6
6
6
6
6
6
4
11
1
X
Ll eilOt I
12
l 1
R0
R1
R2
R3
R4
R1
R0
R1
R2
R3
R2
R1
R0
R1
R2
R3
R2
R1
R0
R1
R4
R3
R2
R1
R0
3
3 2
6
7
7
6
76 L2 7 6 0 7
7
7
7 6
76
6 L1 7 6 0 7
7
7
7 6
76
7
7 6
76
76 L 0 7 6 I 7
7
7 6
76
7
7 6
6
7
76 L 1 7 6 0 7
7
7 6
76
54 L 2 5 4 0 5
32
13
In most practical cases, the magnitudes of the submatrices Rn will decrease by increasing n. Therefore, a solution based on
the central rows and columns of the innitely large system of Eqs. (13) will be sufciently accurate. For n 2, Eq. (13) can
be approximated by
32
3 2 3
2
L2
R0 R1 R2
0
0
0
7
7 607
6R
6
R
R
R
0
L
0
1
2
76 1 7 6 7
6 1
76
7 6 7
6
7 6 7
6 R 2 R1
6
R0 R1 R2 7
14
76 L 0 7 6 I 7
6
7
7 6 7
6 0
6
R
R
R
R
L
0
4 5
4
2
1
0
1 54 1 5
0
0
R2
R1
R0
L2
0
ARTICLE IN PRESS
R. Pedersen et al. / Mechanical Systems and Signal Processing 24 (2010) 14951508
1499
With Ll found for l n n, the solutions to the homogeneous and the inhomogeneous equation of motion, Eq. (1), can
now be found [6,2]. After introducing the modal coordinates yt as
_
_ Rtyt
_
zt Rtyt ) zt
Rtyt
15
and inserting into Eq. (3), the terms are multiplied from the left with Lt and rearranged. The result is the decoupled modal
equation of motion
2
3
&
0
6
7
lk
_
16
yt
4
5yt Lt
1 ft
M
&
P
By inserting Lt nl n Ll eilOt and the oscillating force ft p eiO t into the right hand side of (16), the equation
becomes
2
3
&
n
X
0
6
7
iO lOt
lk
_
Ll
17
yt
4
5yt
1 p e
M
l n
&
Now a modal solution yt yeiO lOt is assumed and inserted into (17). The transformation back to the zt coordinates is
Pn
performed using Eq. (15), and Rt is substituted by its equivalent sum of
r n Rr . After rearranging the terms, the
resulting equation shows the forced response:
3
2
&
n
n
7 0
6
X X
1
7
6
iO Or lt
Rr 6
18
zinhom t
7Ll
1 p e
O
l
O
l
i
5
4
M
k
r n l n
&
When using a nite n and the initial condition zt0 z0 , the solution to the homogeneous equation of motion becomes
2
3
&
6
7
elk tt0
19
zhom t Rt4
5Lt0 z0
&
lk;expected lk1
dl
l l
DO lk1 k1 k2 Ok Ok1
dO
Ok1 Ok2
20
The eigenvalues lk;expected predicted by Eq. (20) are then compared to those actually obtained by solving the hypereigenvalue problem, lk . The absolute difference between lk;expected and lk is calculated, and the eigenvalues in lk are then
identied with the eigenvalue in lk;expected , which shows the minimum difference. A few requirements must be fullled in
order to use the Taylor expansion sorting method:
1. The hyper-eigenvalue problem must be solved for a monotonically increasing value of O.
2. At least two solutions, for k2 and for k1 must be computed without the Taylor expansion sorting.
3. The values of O should not increase too much in each step in order to get a good estimation of lk from Eq. (20), i.e., a
certain smoothness of the function lO is required.
ARTICLE IN PRESS
1500
f2
Im ()
Im ()
f2
f1
f1
Fig. 2. Solutions to hyper-eigenvalue problem: (a) low n or low O; (b) high n or high O.
72 eigenvalues
8 basis eigenvalues
Im ()
Im ()
f1
0
0
f1
Re ()
Fig. 3. Complex conjugate eigenvalues: (a) eigenvalue distribution, (b) high n or high O.
When the eigenvalues have been sorted within each family, the families are sorted relative to each other based on the
mean value of imaginary part of the eigenvalues in the family. The basis eigenvalues are now dened as the central
member of the family, based on the imaginary part.
3. Results
3.1. Presentation of model
The gear pair, the attached shafts and the inertias J1 J4 are shown in Fig. 4. The degrees of freedom (DOF) of the model
are the rotational angles of the four inertias and will be referred to as q q1 q2 q3 q4 T . The model has intentionally been
kept very simple in order to keep the results easy to interpret. The basic data of the gears used in this work are presented in
Table 1. The data are loosely based on the intermediate stage of a 1 MW wind turbine gear box, except a zero degree helix
angle and zero prole shift on both gear and pinion is used. The calculated gear mesh stiffness function is shown in
Fig. 5(a). For a constant rotational speed O, the periodic stiffness can be described using its complex Fourier expansion.
When t is the time, the Fourier series is dened as
f t
1
X
j 1
cj eijt
21
ARTICLE IN PRESS
R. Pedersen et al. / Mechanical Systems and Signal Processing 24 (2010) 14951508
J2
1501
J1
J4
J3
Table 1
Basic gear data.
Driving gear
z
95
an
Unit
Description
22
deg
deg
mm
mm
mm
Nm
Number of teeth
Normal pressure angle
Helix angle
Center distance
Normal module
Prole shift
Tooth width
Torque
20
0
486
8
0
215
74875.84
0
225
17339.67
30
25
2.5
a
mn
x
b
Mt
Driven gear
20
15
10
Original
Using 2 Fourier coeffs.
Using 5 Fourier coeffs.
Using 20 Fourier coeffs.
5
0
0
2
1.5
1
0.5
0
10
15
20
25
30
Fig. 5. (a) Gear mesh stiffness, (b) Fourier expansion of gear mesh stiffness.
In Fig. 5(b), the rst 30 Fourier components of the signals are shown, expressed as Aj 2 abscj . Because of the symmetry
in the complex Fourier components, where cj conjcj , only cj for j 1; . . . ; 30 are shown. The mean value of the gear
mesh stiffness c0 26:66 N mm mm1 is not shown in order to illustrate the other components more clearly.
ARTICLE IN PRESS
1502
30
Gear mesh stiffness [N (mm m)1]
30
25
20
15
c
c
10
cth KissSoft
cth, KissSoft
0
25
20
15
10
c
c
c KissSoft
c KissSoft
0
0
Fig. 6. Validation of stiffness c and mean stiffness cg by comparing to program KissSoft: (a) no load-induced increase in contact ratio, (b) including loadinduced increase in contact ratio.
Table 2
Fundamental eigenfrequencies for n 0.
Mode no.
Frequency (Hz)
1
2
3
4
5
6
7
8
0
0
399
399
585
585
3216
3216
ARTICLE IN PRESS
R. Pedersen et al. / Mechanical Systems and Signal Processing 24 (2010) 14951508
Mode 1, 2
1503
Mode 3, 4
1.5
1
q1
q2
0.5
0.5
q3
q4
0.5
0.5
0
/2
3/2
Mode 5, 6
/2
3/2
/2
3/2
Gear mesh position [rad]
Mode 7, 8
1.5
1.5
0.5
0.5
0.5
0.5
0
/2
3/2
Gear mesh position [rad]
Fig. 7. Mode shapes as a function of gear mesh position y Ot, calculated for n 10.
n
X
Aa eiaOt1
22
a n
At1 is then inserted into Eq. (4), and the eigenvalue problem is solved. The resulting (positive) eigenfrequencies are
shown in Fig. 8 for two different values of n. From the gure, it can be seen that the highest eigenfrequency at roughly
3200 Hz is much more sensitive to the variations in the gear mesh stiffness than the other eigenfrequencies. Also, n has an
inuence on the extent of the frequency interval, in which the highest eigenfrequency is located. As a conclusion to this
quasi-static analysis, n can be expected to play an important role when determining the free and the forced response of
the time-variant system.
3.5. Necessary number of Fourier components
The accuracy of the time-variant modal analysis depends on the number of Fourier components, n, included in the
analysis. When the applied force f in Eq. (1) is constant or zero, only the changing stiffness in Kt can excite the system.
From Eq. (10), it can be expected, that only frequencies up to approximately f nO will be excited. From the time-invariant
modal analysis of the system it is known that the highest eigenfrequency is around 3215 Hz. When the gear mesh
frequency is O 1357:59 rad=s 216 Hz, this eigenfrequency can be expected to be excited only when n Z3215=216 15.
The accuracy of the method can be determined by comparing the solution in the time domain to the corresponding
solution obtained by numerical integration. For this purpose, q3 has been chosen, as this DOF shows the largest difference
between the two calculation methods, as seen from Fig. 9(a). It might be interesting to represent the time variations of the
dynamic mesh force (as opposed to q3 in Fig. 9) since it also gives an indication on the presence of non-linearity (contact
losses when the mesh force is negative). Since this loss of contact is not considered in the present gear mesh model, the
dynamic mesh force is calculated as the difference between the displacements of the two gears multiplied by the mesh
stiffness, Ft ctq3 tr3 q2 tr2 , where r2 and r3 are the radii of the two gears. Comparison based on a single DOF is the
simplest way, since Ft offers no new information on the dynamic behavior of the system. It is clear from Fig. 9(a) how the
ARTICLE IN PRESS
R. Pedersen et al. / Mechanical Systems and Signal Processing 24 (2010) 14951508
3500
3500
3000
3000
Eigenfrequencies [Hz]
Eigenfrequencies [Hz]
1504
2500
2000
1500
1000
500
2500
2000
1500
1000
500
0
/2
3/2
/2
3/2
x 104
Numerical integration
Timevariant modal analysis
104
q2
Difference [rad]
1
q3 (t) [rad]
q1
0
1
q3
105
q4
106
2
0.1
0.11
5
10
15
20
Number of Fourier coefficients, n
Fig. 9. Comparison of solutions: (a) maximum difference between numerical integration and time-variant modal analysis, (b) comparison of timedomain solutions (zoom). Only q3 t is shown.
accuracy of the modal solution increases signicantly when n reaches 15. An increase in n beyond 18 does not increase
accuracy much. A zoom of the two solutions is shown in Fig. 9(b).
3.6. Change in fundamental frequencies
For each number of n, the eigenvalue problem, Eq. (8), changes. It can therefore be expected that both the fundamental
eigenfrequencies (for k 0) and the higher order parametric eigenfrequencies (for k 2 n; n; ka0) will change as a
function of n. In the following analysis, the change in the fundamental eigenfrequencies, evaluated at the nominal speed
O 1357:59 rad=s, is investigated. The results are shown in Fig. 10, where the change (in per cent) of the eigenfrequencies
belonging to the three elastic modes, relative to a time-invariant modal analysis (n 0) are plotted. It can be seen that
there are no signicant changes in the fundamental eigenfrequencies. At no point, the frequencies change more than 0.16
per cent.
3.7. Normalization factors
For each of the 2N eigenvectors, the importance of the k th harmonic parametric vibration mode can be calculated. This
is done by nding the relative magnitude of the harmonic components of the eigenvector. For the k th harmonic
ARTICLE IN PRESS
R. Pedersen et al. / Mechanical Systems and Signal Processing 24 (2010) 14951508
1505
0
Modes 7 and 8
Modes 5 and 6
Modes 3 and 4
0.05
0.1
0.15
0.2
0
10
15
20
1
NFj;k q
T
rj;k rj;k
rj;n
6 7
7
6
7
6
6 rj;1 7
7
6
7
6
with rj 6 rj;0 7 for j 1; . . . ; 2N
7
6
6 rj;1 7
7
6
6 7
5
4
rj;n
23
In Figs. 11 and 12, normalization factors for all eight modes are plotted versus the gear mesh frequency O (in rad/s), for
n 3 and 10, respectively. The NF-axes in the plots are logarithmic and show the interval NF 2 102 ; 101 . NF for the
fundamental harmonic component, k 0, is set to NFj;0 1 in all cases. With these denitions, the normalization factors
show how the fundamental part of the mode shape vector r0 is related to the time-varying parts rk for k n; . . . ; n
depending on the operational speed O. The k th harmonic component showing the smallest NFj;k will be predominant in
the j th mode shape at a particular gear mesh frequency O. A number of observations can be made from the normalization
factor plots, Figs. 11 and 12:
1. For the rigid-body modes 1 and 2, the fundamental harmonic k 0 is predominant for all gear mesh frequencies O 4 0.
2. For all elastic modes, there is a symmetry between the positive (left column) and the negative frequencies (right column
in the plots): For the j th positive eigenfrequency, the normalization factor NFj;k is equal to NFj;k for the corresponding
negative eigenfrequency.
3. For modes 36, there exist certain mesh frequency ranges, for which the time-varying part of the mode shape
(parametric vibrations) will be more important than the stationary. These frequency ranges are independent of the
number of Fourier components included in the analysis. For instance, for mode 3 in Fig. 11, NF for the 2 harmonic
component shows a local minimum around O 600 rad=s. This potential 2 harmonic resonance is found for all
n Z 2.
4. Modes 7 and 8 at roughly f 7 3216 Hz behave fundamentally different from the other elastic modes (modes 36). No
n-independent frequencies with low NF are observed.
5. For modes 7 and 8, frequency intervals exist in which NF for one or more harmonic components of the mode shape are
smaller than 1. In these frequency intervals, which depend on the number of Fourier components included but are
generally located in O 2 0; 600 rad=s, the content of the time-varying part of the mode in the overall mode shape is
larger than the fundamental component. These modes are strongly dependent on the relative angular movement
between the two gears, and are strongly affected by the time-variant tooth stiffness.
ARTICLE IN PRESS
1506
Mode 1
Mode 2
100
NF
NF
100
102
1
2
102
0
500
1000
1500
Mode 3
500
1000
1500
Mode 4
NF
100
NF
100
102
102
0
500
1000
1500
Mode 5
500
1000
1500
Mode 6
NF
100
NF
100
102
102
0
500
1000
1500
Mode 7
500
1000
1500
Mode 8
NF
100
NF
100
102
102
0
500
1000
[rad/sec]
1500
500
1000
[rad/sec]
1500
4. Conclusion
The theory of modal analysis of time-variant systems has been applied to a simple spur gear pair with a periodically
time-varying gear mesh stiffness. It has been made clear that a large number of terms in the Fourier expansion of the
system matrices is necessary in order to yield results of sufcient accuracy. This is a direct result of the jumps in the gear
mesh stiffness function. In the cases of exible rotors with non-symmetrical cross section, exible rotating discs, and
exible rotating blades, a very reduced number of Fourier components is needed, normally n 2, as the time-variant
coefcients are normally sine and cosine functions.
It has been shown that there exist regions, in which the higher order parametric contributions to the overall mode
shapes will be signicant. For the case studied here, these effects were mainly observed at low gear mesh frequencies.
The elastic mode with the highest frequency behaved differently from the other elastic modes. While the two lower
modes showed parametric resonance frequencies that were largely independent of the number of Fourier components
included in the analysis n, the parametric resonance areas of the high frequency mode strongly depended on n. Overall, the
vibrations related to the higher frequency mode seemed to be more sensitive to the time-variant nature of the gear mesh
stiffness.
The system studied in this work consisted of a single gear stage. A typical modern wind turbine gearbox consists of one
or two planetary stages followed by one or two parallel gear stages, with a total of 815 gear meshes. When applying the
theory of modal analysis of time-variant systems to such a complex system, great care must be taken when interpreting
the results. The larger the number of Fourier components needed to expand the periodic time-variant coefcients, the
larger the hyper-eigenvalue problem becomes. It means also, it becomes more complicated and complex to physically
interpret the basic and parametric mode shapes. Nevertheless, by exploring the denition of NF, the importance of the
time-varying part of the mode shapes (parametric modes) can be investigated and quantied as a function of the gear
mesh frequency. Compared to time-step integration schemes, the main advantage of the time-variant modal analysis is
ARTICLE IN PRESS
R. Pedersen et al. / Mechanical Systems and Signal Processing 24 (2010) 14951508
Mode 1
Mode 2
10
100
NF
NF
100
102
8
7
102
0
500
1000
Mode 3
1500
500
1000
Mode 4
1500
1
0
102
500
1000
Mode 5
1500
500
1000
Mode 6
1500
1
2
3
4
100
NF
NF
100
NF
102
6
4
3
100
NF
100
1507
6
2
10
10
500
1000
Mode 7
1500
500
1000
Mode 8
1500
7
8
9
10
NF
100
NF
100
102
102
0
500
1000
[rad/sec]
1500
500
1000
[rad/sec]
1500
that it offers an analytical solution to the vibration problem. Therefore a modal truncation is possible, which removes the
need for the very short time step used for a numerical integration of a system with high eigenfrequencies. Also the method
allows to expand the analysis to the concepts of observability and controllability [8], which offer a quantication of the
parametric vibrations.
References
[1] M. Ertz, A. Reister, R. Nordmann, Zur Berechnung der Eigenschwingungen von Strukturen mit periodisch zeitvarianten Bewegungsgleichungen, in:
H. Irretier, R. Nordmann (Eds.), Schwingungen in rotierenden Maschinen, vol. III, Springer, Vieweg, Braunschweig, Germany, 1995, pp. 288296.
[2] J. Xu, R. Gasch, Modale Behandlung linearer periodisch zeitvarianter Bewegungsgleichungen, Archive of Applied MechanicsIngenieur Archiv 65 (3)
(1995) 178.
[3] F.E. Boru, H. Irretier, Numerical and experimental dynamic analysis of a rotor with non-circular shaft mounted in anisotropic bearings, in: SIRM
20098th International Conference on Vibrations in Rotating Machines, Vienna, Austria, 2009, pp. 110.
ARTICLE IN PRESS
1508
[13] H. Irretier, F. Reuter, Experimentelle Modalanalyse an einer rotierenden Scheibe, in: H. Irretier, R. Nordmann (Eds.), Schwingungen in rotierenden
Maschinen, vol. I, Springer, Vieweg, Braunschweig, Germany, 1991, pp. 6677.
[14] J. Bienert, Anwendung der Strukturmodikation zur Vorhersage der Kreiselwirkung von symmetrischen und unsymmetrischen Rotoren, in:
H. Irretier, R. Nordmann (Eds.), Schwingungen in rotierenden Maschinen, vol. IV, Springer, Vieweg, Braunschweig, Germany, 1997, pp. 97104.
[15] I. Santos, C. Saracho, J. Smith, J. Eiland, Contribution to experimental validation of linear and non-linear dynamic models for representing rotor-blade
parametric coupled vibrations, Journal of Sound and Vibration 271 (35) (2004) 883904.
[16] G.W. Blankenship, R. Singh, Comparative study of selected gear mesh interface dynamic models, Advancing Power Transmission into the 21st
Century and American Society of Mechanical Engineers, Design Engineering Division (Publication) DE 43 pt 1, 1992, pp. 137146.
[17] P. Velex, M. Ajmi, On the modelling of excitations in geared systems by transmission errors, Journal of Sound and Vibration 290 (3) (2006) 882909.
[18] A. Kahraman, R. Singh, Non-linear dynamics of a spur gear pair, Journal of Sound and Vibration 142 (1) (1990) 4975.
[19] A. Kahraman, R. Singh, Non-linear dynamics of a geared rotor-bearing system with multiple clearances, Journal of Sound and Vibration 144 (3)
(1991) 469506.
[20] A. Kahraman, R. Singh, Interactions between time-varying mesh stiffness and clearance non-linearities in a geared system, Journal of Sound and
Vibration 146 (1) (1991) 135156.
[21] A. Kahraman, R. Singh, Dynamics of an oscillator with both clearance and continuous non-linearities, Journal of Sound and Vibration 153 (1) (1992)
180185.
[22] H. Peeken, C. Troeder, G. Diekhans, Parametererregte Getriebeschwingungen, Teil 1 VDI-Z 122 (20) (1980) 869877.
[23] H. Peeken, C. Troeder, G. Diekhans, Parametererregte Getriebeschwingungen, Teil 2 VDI-Z 122 (21) (1980) 967977.
[24] H. Peeken, C. Troeder, G. Diekhans, Parametererregte Getriebeschwingungen, Teil 3 VDI-Z 122 (22) (1980) 10291043.
[25] H. Peeken, C. Troeder, G. Diekhans, Parametererregte Getriebeschwingungen, Teil 4 VDI-Z 122 (23/24) (1980) 11011113.
[26] P. Velex, D. Berthe, Dynamic tooth loads on geared train, in: Proceedings of the 1989 International Power Transmission and Gearing Conference:
New Technologies for Power Transmissions of the 1990s, 1989, pp. 447454.
[27] P. Velex, M. Maatar, A mathematical model for analyzing the inuence of shape deviations and mounting errors on gear dynamic behaviour, Journal
of Sound and Vibration 191 (5) (1996) 629660.
[28] M. Ajmi, P. Velex, A model for simulating the quasi-static and dynamic behaviour of solid wide-faced spur and helical gears, Mechanism and
Machine Theory 40 (2) (2005) 173190.
[29] R. Padieth, Exakte Ermittlung der Zahnform, Antriebtechnik 17 (10) (1978) 434436.
Universitat
Germany, 1973.
[33] L. Meirovitch, Fundamentals of Vibrations, McGraw-Hill, 2001.