Engineering Computations

A new approach to finite element modelling of flexible rotors

M. Geradin N. Kill
Article information:
To cite this document:
M. Geradin N. Kill, (1984),"A new approach to finite element modelling of flexible rotors", Engineering Computations, Vol. 1
Iss 1 pp. 52 - 64
Permanent link to this document:
http://dx.doi.org/10.1108/eb023560
Downloaded on: 02 November 2015, At: 01:20 (PT)
References: this document contains references to 0 other documents.
To copy this document: permissions@emeraldinsight.com
The fulltext of this document has been downloaded 186 times since 2006*
Users who downloaded this article also downloaded:
Ghulam Nabi, Rick Holden, (2008),"Graduate entrepreneurship: intentions, education and training", Education + Training,
Downloaded by UNIVERSITY OF EXETER At 01:20 02 November 2015 (PT)

Vol. 50 Iss 7 pp. 545-551 http://dx.doi.org/10.1108/00400910810909018

Rodney McAdam, Bob Mason, Josephine McCrory, (2007),"Exploring the dichotomies within the tacit knowledge literature:
towards a process of tacit knowing in organizations", Journal of Knowledge Management, Vol. 11 Iss 2 pp. 43-59 http://
dx.doi.org/10.1108/13673270710738906
Nancy Furlow, (2014),"The New Brand Spirit: How Communicating Sustainability Builds Brands, Reputations and Profits",
Journal of Product & Brand Management, Vol. 23 Iss 4/5 pp. 376-377 http://dx.doi.org/10.1108/JPBM-03-2014-0519

Access to this document was granted through an Emerald subscription provided by emerald-srm:463575 []
For Authors
If you would like to write for this, or any other Emerald publication, then please use our Emerald for Authors service
information about how to choose which publication to write for and submission guidelines are available for all. Please visit
www.emeraldinsight.com/authors for more information.
About Emerald www.emeraldinsight.com
Emerald is a global publisher linking research and practice to the benefit of society. The company manages a portfolio of
more than 290 journals and over 2,350 books and book series volumes, as well as providing an extensive range of online
products and additional customer resources and services.
Emerald is both COUNTER 4 and TRANSFER compliant. The organization is a partner of the Committee on Publication
Ethics (COPE) and also works with Portico and the LOCKSS initiative for digital archive preservation.

*Related content and download information correct at time of download.

 A new approach to finite Developing such a formalism and checking its accuracy
on an appropriate example has been the main objective of
element modelling of the paper.
As it is described in the text, two approaches are
flexible rotors possible in the analysis of rotating structures. The most
natural one, from the point of view of structural analysis,
is to derive the dynamic equilibrium equations of the
M . Geradin and N. Kill rotor in a rotating frame. In this way, the strain energy of
Universite de Liege, Laboratoire de Techniques the flexible rotor does not contain time-dependent terms.
Aéronautiques et Spatiales, rue Ernest Solvay 21, When assembling the whole machine, however, time-
4000 Liege, Belgium dependent terms appear in the equations if the fixed parts
(Received September 1983) present some anisotropy in their stiffness and damping
properties.
The second approach is to adopt a complementary
ABSTRACT point of view. If the rotor itself has perfect isotropy, it is
more rewarding to derive the dynamic equilibrium equa-
A 3-dimensionaI formalism adapted to critical speed and tions of the whole system in an inertial frame. It is then
stability analysis of rotating machinery is presented. possible to take care of arbitrary stiffness and damping
Gyroscopic effects are properly taken into account in the properties for the fixed parts of the system.
expression of the kinetic energy through a proper kine-
As both approaches have their respective advantages,
matic description which takes account of the local
both will be considered to derive the kinetic energy
changes of angular velocity induced by the deformation.
expression from which result the dynamic equations of
Two approaches are suggested according to the re- motion are written respectively in the rotating frame
Downloaded by UNIVERSITY OF EXETER At 01:20 02 November 2015 (PT)

spective stiffness and geometric properties of the rotating (section 1) and in the inertial frame (section 2). Section 3
and fixed parts: the rotating frame approach and the deals with the finite element discretization of the 3-
inertial frame approach. In both cases, an axisymmetric dimensional continuum. As the rotating parts of a
finite element modelling of the rotor is proposed which machine have generally revolution symmetry, an axisym-
takes into account the 3-dimensional nature of the system metric modelling of the displacement field with Fourier
while keeping the number of degrees of freedom to a expansion in the azimuthal direction is adopted, and it is
reasonable level. further assumed that only bending deformations (cor-
In order to perform the stability analysis, a preliminary responding to cos θ and sin θ harmonics) are significant.
reduction of the system is achieved using the component Because of the occurrence of gyroscopic effects, the
mode method. Critical speeds are calculated next either successive analyses cannot be separated as in a linear
by the classical sweeping procedure or by a direct method analysis of the structure at rest. The coupling introduced
when the restrictive conditions of its applicability are met. by the gyroscopic matrix makes it necessary to keep both
The concepts proposed are then applied to an example harmonics in the same analysis and thus, doubles the size
in order to demonstrate their adequacy. of the system to be solved.
As shown in section 4, the general form of the eigen­
INTRODUCTION value problem arising in the analysis of rotating systems
is:
An abundant literature exists on the subject of critical
speed and stability analysis of rotating ma-
chinery 1 - 5 . The models that are generally used are
limited to the following assumptions: (l) the rotating and where Ω is the rotation speed, M the mass matrix of the
fixed parts of the structure have linear material and structure, K(Ω) a possibly non-symmetric stiffness matrix
geometric behaviour; (2) the rotating shaft is represented and B(Ω) a matrix collecting damping and gyroscopic
by beam finite elements (with shear deformation and terms. The application of a standard eigenvalue extrac­
rotary inertia effects of right sections possibly included); tion method such as subspace bi-iteration to perform a
(3) the disks attached to the shaft are infinitely rigid; (4) the stability analysis of the system necessitates to recast it in
stiffness and damping properties of bearings and seals are first order form. When an axisymmetric model is adopted,
linear functions of displacements and velocities. it is thus not computationally efficient to solve the above
Beam models are very useful in the preliminary analysis equation directly. It is rather advisable to reduce its size
of machines such as aircraft engines, turbopumps, electri- first using a reduction method such as component mode
cal turbomachinery, etc. but are not always adequate for a synthesis.
detailed modelling of such structures. On the other hand, Even after reduction, the stability analysis of the system
it is well known that a linear 3-dimensional elastic model is a lengthy procedure since it becomes necessary to solve
of rotating structures is not appropriate when the inertia a different eigenvalue problem for each value of the
forces due to large moments of inertia become important; rotation speed. Fortunately, there are cases where the
the gyroscopic moment which arises from the rotation of a calculation of the critical speeds of the system does not
disk about its symmetry axis is not properly taken into require this sweeping procedure. In the absence of damp­
account in a fully linear formalism, since it arises from ing and provided that no cross-coupling terms appear
non-linear terms associated with the deflection of the between transverse directions in the stiffness matrix, a
rotation axis. direct method exists which allows the determination of
There is thus a need for a 3-dimensional formalism the critical speeds by solving only one eigenvalue pro­
which includes these gyroscopic effects as does the blem. Its principle is given in section 5.
classical theory of rotating beams with attached disks. The concepts developed in the paper have been applied
0264-4401/84/010052-13$2.00
© 1984 Pineridge Press Ltd
52 Eng. Comput., 1984, Vol. 1 , March
 Finite element modelling of flexible rotors: M. Geradin and N. Kill

frames, and parallel to the convected frame that results

from local deformation (Figure 1.2).
The matrix transformation from the inertial frame to
the dynamic frame consists of three elementary rotations:
a rotation f about Ozi which reduces to f = Ωt when
torsional deformations are assumed zero; a rotation θ
about the new axis Oyr; a rotation ψ about the body axis
Ox,,. The matrix representation Pd of the position vector of
a given point P in the dynamic frame is related to its
expression pi in the inertial frame by:

where R is the orthogonal rotation matrix

in section 6 to a rotating structure with properties such
that a comparison between a beam model and a 3-
dimensional model is of interest. Among others, it shows
the importance that gyroscopic effects may take in the
dynamic behaviour of a flexible rotor, and illustrates the
very good agreement between the beam and axisymmetric
models when the rotor has moderately massive
properties.
Downloaded by UNIVERSITY OF EXETER At 01:20 02 November 2015 (PT)

As (1.2) is to be used later to calculate a quadratic

1. ROTATING FRAME APPROACH TO ROTOR
approximation to the kinetic energy, eqn (1.2) can be
KINEMATICS AND DYNAMICS
limited to second order terms in the rotations (θ, ψ):
A convenient way to derive the equations of motion of a
flexible rotor is to express them in a frame linked to the
prescribed rotation of the hypothetical rigid shaft. This
approach presents a certain number of advantages,
namely: (a) it allows a possible anisotropy in the stiffness
properties of the shaft to be taken into account; (b) under
some restrictive conditions, it leads to an easy calculation
of the critical speeds of the system. However, the rotating
frame approach also presents limitations. Periodic terms
appear in the equations of motion as soon as the bearings
exhibit anisotropic stiffness properties, or when an ap­
propriate modelling of fixed parts (housings, carter,...) is
required. And also, the method fails in the case of
machines with several rotors.
Let us now denote by (ur, vr, wr) the components of the
1.1 Kinematics elastic displacement field in the rotating frame. Due to
Let us consider a flexible rotating body with prescribed combined rotation and deformation, a given point P
uniform angular velocity Ω about its reference axis Oz. located at rest in the plane Ox r y r transforms into a point
The inertial frame Ox i y i z i is defined such that Ozi P' with coordinates:
coincides with the rotation axis.
The rotating frame Ox r y r z r is defined next such that its
axis Oz coincides also with the rotation axis, and has a
relative angular velocity Ω with respect to Ox i y i z i (Figure
1.1). As a result of rotation, the rotor undergoes defor­
mation and the actual angular velocity changes from
point to point. In order to express the combined effects of
deformation and prescribed rotation speed on the kinetic
energy expression, let us adopt the following hypotheses:
(1) material strains and rotations remain infinitesimal, so
that the kinetic energy expression can be limited to a
quadratic form in the generalized displacements and
velocities associated with the elastic deformation; (2)for
the same reason, the eulerian and lagrangian description
of elastic deformation may be confused. At every point,
the strain energy and the kinetic energy are computed in a
frame attached to local deformation.
The expression of the local angular velocity requires the
further definition of a dynamic reference frame Oxdydzd
having a common origin with the inertial and rotating

Eng. Comput., 1984, Vol. 1 , March 53

 Finite element modelling of flexible rotors: M. Geradin and N. Kill

Equation (1.11) will be used to calculate the axial velocity

of point P and to get a quadratic expression of the kinetic
energy compatible with that of the strain energy.
In order to calculate the velocity of point P in the
dynamic frame one notes that:
in the rotating frame. Its coordinates may also be
expressed in the dynamic frame:

which yields to:

in which case pr and pd are related by the transformation

matrix D from the dynamic frame to the rotating frame:

where [to] is the antisymmetric matrix of angular

or explicitly: velocities:

The explicit form of the result (1.13) is:

Downloaded by UNIVERSITY OF EXETER At 01:20 02 November 2015 (PT)

Let us next imagine to express pd in terms of quantities

defined in the rotating frame:

in which case u and v combine the displacements (ud, vd)

associated to deformation with an additional component
due to the rotation of the shaft section in which P is
located. Substituting (1.8) into (1.7) and deleting
higher order terms yields to the second order
approximation:

1.2 Kinetic energy of theflexiblerotor

Integrating the kinetic energy on the underformed
configuration yields to:

Since there is no difference to thefirstorder between (u, v)

and (ur, vr), the approximations ur@u and vr@v may be
adopted. This is equivalent to assuming that eulerian and
lagrangian descriptions of in-plane displacements are
confused. Moreover, since the angular displacements 0 Substituting (1.14) into (1.15) provides a decomposition of
and ψ are infinitesimal, they can be related to the elastic T into three terms:
displacements at the point of interest:

where (l)the relative kinetic energy:

On the other hand,wtand wd differ by a first order term, so

that the following relation is introduced:

54 Eng. Comput., 1984, Vol. 1, March

 Finite element modelling of flexible rotors: M. Geradin and N. Kill

coincides with the total kinetic energy of the same and the three contributions to the kinetic energy of the
structure at rest. Its discretization provides the mass beam are then rewritten in the form:
matrix of the structure; (i) the relative kinetic energy:
(2) the coupling kinetic energy contains five terms

which splits then into terms representing translation and

rotation of cross-sections, respectively;
(ii) the coupling kinetic energy is unchanged:

(iii) the centrifugal kinetic energy contains a negative

contribution due to cross-section rotation:

but one verifies in a straightforward manner that the only

one to contribute to inertia forces is:
Downloaded by UNIVERSITY OF EXETER At 01:20 02 November 2015 (PT)

They correspond both to the classical expressions for

rotating beams with gyroscopic effects included6. One
observes also that only the centrifugal kinetic energy is
Its discretization will yield to the gyroscopic matrix of the affected by the additional gyroscopic effects induced by
structure; the deformation; (1.22) and (1.23) would not differ in
(3) the last term is centrifugal kinetic energy: a purely linear description of the kinematics.

Axisymmetric structures. The second type of structural

model corresponds to axisymmetric structures, in which
case cylindrical coordinates (p, θ, z) are adopted to
describe the deformation in the rotating frame. The
transformation from cartesian coordinates to cylindrical
coordinates is:

As (1.20) is a quadratic form of the displacements, it

will combine with the strain energy of the structure to give
a modification of its apparent stiffness. The first term gives
rise to translational centrifugal forces, and the remaining
terms produce rotary inertia forces. Their introduction is
classical in the dynamics of rigid disks, but not in the
analysis of flexible rotors.
The expressions (1.17), (1.19) and (1.20) of the com­ and the components of the displacement field transform
ponents of the kinetic energy (1.16) can be specialized to accordingly:
the structural models which are of common use in rotor
dynamics.

Beam-like behaviour. The first case corresponds to

beam-like behaviour of the rotor, in which case the
Bernoulli hypothesis that cross-sections are conserved
and remain orthogonal to the neutral axis after defor­ The angular displacements become:
mation has to be introduced. Equation (1.11) provides
then the condition:

Eng. Comput., 1984, Vol. 1 , March 55

 Finite element modelling of flexible rotors: M. Geradin and N. Kill

with the definitions: bearing with anisotropic properties, which are frequently
encountered in practice, are easily modelled. Similarly, the
analysis of machines with several rotors and/orfixedparts
(carter) does not raise additional difficulties when a fixed
reference is used.
The corresponding expressions of the components of On the other hand, the inertial frame approach be­
kinetic energy are: comes cumbersome when the rotor presents anisotropic
geometric stiffness properties, since periodic inertia and
stiffness coefficients appear in these cases. The analysis of
unsymmetric rotors is thus discarded in the present
analysis.

2.1 Kinematics
The relation (1.14) which gives the velocityvdof a point
expressed in the dynamic frame has been calculated in
terms of quantities in the rotating frame. It can be
rewritten in the form:

The expressions (1.29) to (1.31) are the basis for a

Downloaded by UNIVERSITY OF EXETER At 01:20 02 November 2015 (PT)

discretization of the kinetic energy in axisymmetric

structures. In the azimuthal direction the displacement
field is decomposed into Fourier harmonics, while in the
meridian plane each Fourier coefficient is discretized in
terms of 2-D shape functions.

1.3 Strain energy of the flexible rotor In order to obtain the same expression in terms of
quantities defined in the inertial frame, let us build the
In order to express the strain energy of the flexible rotor transformation R from inertial to dynamic frame in terms
the infinitesimal strain tensor: of three new rotations defined in the following order: an
elastic rotation θi about the inertial axis Oyi; a second
elastic rotation ψi about the new axis Ox; a final rotation
ψ=Ωt about Ozd. The resulting rotation matrix is:
is related to the displacement field by:

and the associated angular velocity vector, calculated

from [ω] = — RRT and reduced to second order terms, is:
The total energy of the flexible rotor is then expressed in
the classical form:

where H is the Hooke matrix of elastic coefficients of the To obtain the final expression to (2.1), material coor­
continuum. It is straightforward to specialize (1.34) either dinates, displacements and rotations are also transformed
to beam behaviour (with or without shear deformation from rotating to inertial axes:
included) or to axisymmetric structures.

2 INERTIAL FRAME APPROACH

The dynamic equations of motion of flexible rotors may
also be expressed in an inertial frame. In which case

56 Eng. Comput., 1984, Vol. 1, March

 Finite element modelling of flexible rotors: M. Geradin and N. Kill

Owing to the hypothesis of isotropic cross-sections there

is not periodic dependence with respect to time of these
expressions, since the boundary of the integration volume
Vi is then insensitive to rotation.
After substitution of (2.3) and (2.4) in (2.1), the resulting
velocity vector vd has for components:
Beam-like behaviour. Let us particularize the kinetic
energy expression just as we did in the rotating frame
approach to beam-like behaviour. The conservation of
right cross-sections implies the condition:

in which case the kinetic energy transforms into:

T = T1 + T2 where T1 combines the translational and
rotational kinetic energies

Just as in the rotating frame approach, the elastic angular and T2 is the kinetic energy term of gyroscopic origin
displacements are related to lateral displacement com-
ponents by:
Downloaded by UNIVERSITY OF EXETER At 01:20 02 November 2015 (PT)

Axisymmetric structures. Similarly, for axisymmetric

2.2 Kinetic energy of the flexible body solids, the kinetic energy is expressible in terms of inertial
Let us introduce (2.5) into (1.15) to obtain the three quantities defined in axisymmetric coordinates:
components of the kinetic energy. After simplification the
final result is:

with (i) the translational kinetic energy

The transformation from cartesian to axisymmetric com-

ponents of displacements is:

(ii)the kinetic energy associated to rotation:

while the angular displacements become:

which does not contribute to inertia forces; (iii) the kinetic

energy due to gyroscopic effects:

with

The only terms in eqn (2.10) to contribute to the inertia The substitution of (2.14), (2.15) and (2.16) into (2.8) and
forces are: (2.10) yields the kinetic energy expression:

Eng. Comput., 1984, Vol. 1, March 57

 Finite element modelling of flexible rotors: M. Geradin and N. Kill

2.3 Strain energy When dealing with linear structures at rest each term of
If, in eqn (1.34), the strains ε are expressed in terms of eqn (3.3) is calculated from a separate analysis, and the
inertial displacements, the resulting expression of the final displacements and stresses result from a recom­
strain energy is formally identical to that obtained in the bination of the successive harmonics.
rotating frame. Let us assume next that the rotating flexible rotor is
Again, if the structure had geometric anisotropy, the subject to bending deformation only. This means that one
integration volume would be time dependent and would is only interested in the analyses corresponding to a
introduce a periodic dependence of the strain energy. On circumferential wave n = l , with m=0 or 1. If both
the other hand, in a structure with anisotropic stiffness analyses were made separately, however, gyroscopic
properties, the time dependence would come from the effects would be neglected since they result from a
matrix of elastic coefficients. Both cases are discarded coupling between bending modes in two orthogonal
since they would introduce periodic terms in the stiffness directions. Both analyses describing bending in ortho­
matrix. gonal directions have thus to be performed
simultaneously.
Let us introduce the decomposition of displacement
components:
3. FINITE ELEMENT DISCRETIZATION

Beam-like structures
Let us consider two types of finite element discreti­
zation. The first one is to assimilate the flexible rotor to a
prismatic beam. The discretization consists then of repre­
senting the bending deflections of the beam with cubic
shape functions defined in terms of end displacementsu,v,
Downloaded by UNIVERSITY OF EXETER At 01:20 02 November 2015 (PT)

0 andψ.If the unicolumn matrix q collects all the degrees

of freedom of the discretized structure, the discretized
forms of the kinetic and strain energies are:
(i) in the rotating frame approach:
to calculate the kinetic energy.
In the rotating frame approach, the relative kinetic
energy (1.29) becomes:

(ii)in the inertial frame approach:

where M is the mass matrix, K the stiffness matrix, N is the whereμis the volumic mass and S is the generating surface
centrifugal stiffness matrix and G'r,G'iare the gyroscopic of the revolution body. The coupling energy term (1.30)
matrices expressed respectively in the rotating and inertial becomes:
frames. Since isotropy of the structure has been assumed,
the mass and stiffness matrices are identical in both
frames.

Axisymmetric structures
The second type of discretization corresponds to a 3-
dimensional modelling of the axisymmetric rotor. and one observes that it is responsible for the coupling
Cylindrical coordinates are used, and the finite element between both analyses. The last term is the centrifugal
discretization involves a Fourier expansion of the dis­ kinetic energy:
placement field in the circumferential direction. In a
meridian plane, each Fourier coefficient is represented in
terms of 2-dimensional shape functions.
In the general case, the Fourier expansion is:

58 Eng. Comput., 1984, Vol. 1 , March

 Finite element modelling of flexible rotors: M. Geradin and N. Kill

In the inertial frame approach, the translational kinetic dynamic equilibrium equations of the free-free substruc-
energy (2.8) becomes ture at zero rotation speed are partitioned accordingly;

The constrained modes are obtained by imposing a unit

deflection at each boundary node successively. They are
thus given by the matrix resulting from static
and the kinetic energy due to gyroscopic effects trans- condensation:
forms into:

The internal vibration modes are obtained by solving the

free vibration problem of the substructure clamped on its
boundary:
Let us note that the displacement field has to be de-
termined in a unique manner, a condition which is not
automatically achieved on the revolution axis. In p=0,
the components of the displacement field are no longer
functions of 0. This gives rise to the equalities: They are collected in the matrix of normal modes:
Downloaded by UNIVERSITY OF EXETER At 01:20 02 November 2015 (PT)

The basic assumption of the component mode method

The half-meridian section of the rotors is modelled in consists to assume that the behaviour of a given substruc-
terms of triangular or quadrangular isoparametric finite ture in the global system can be accurately described in
elements with linear, quadratic or cubic shape functions. terms of the constrained modes augmented by a reduced
The discretized displacement field is thus written in the set of normal modes. It can indeed be expected that an
form: accurate description of the dynamic deflection of the
substructure will be preserved if only some lower fre-
quency modes are kept in the model.
One combines then eqns (4.2) and (4.4) to build up the
where W(p, z) is the matrix of shape functions, and q matrix which expresses the transformation from the basis
collects the generalized degrees of freedom defined in of local degrees of freedom q to boundary displacements
axisymmetric coordinates. After discretization, the kinetic qB and intensities of normal modes p:
and strain energies are written in either one of the forms
(3.1) or (3.2) according as the rotating frame or the inertial
frame approach is adopted. At the solution level, there is
thus no difference between the beam-like and the axisym-
metric structure discretizations. It can be written in the compact form:

4. REDUCTION OF SYSTEM SIZE BY THE Let us introduce the reduction relationship (4.6) into the
COMPONENT MODE METHOD kinetic and strain energy expressions (3.1) obtained in a
The analysis of flexible rotors leads to systems with rotating frame. The result is:
relatively large number of degrees of freedom, at least
when an axisymmetric model of the 3-dimensional struc-
ture is adopted. It is therefore necessary to reduce the size
of the system by a reduction method such as the now
classical component mode method7.
For this purpose let us split the global system in a with the reduced matrices:
certain number of substructures, the boundary of each
substructure being defined as its boundary with adjacent
substructures. The behaviour of each substructure is then
described in terms of two kinds of component modes: the
constrained modes which are static response modes to In a similar manner, one would obtain in an inertial frame
boundary deflections, and the normal modes which are the reduced matrices and
free vibration modes in clamped boundary configuration. From the point of view of the finite element discreti-
To express these modes, let us partition the degrees of zation, each substructure is now defined as a superelement
freedom of the substructure into boundary displacements characterized by its mass, stiffness and gyroscopic mat-
qB and internal displacements q1, in which case the rices (4.8).

Eng. Comput., 1984, Vol. 1 , March 59

 Finite element modelling of flexible rotors: M. Geradin and N. Kill

5. EQUATIONS OF MOTION AND STABILITY Equation (5.6) can be rewritten in the form of a first order
ANALYSIS system of equations:
The dynamic equilibrium equations of the system result
directly from the application of Hamilton's principle.
The kinetic energy and the strain energy of rotating
parts are given by either (3.1) or (3.2) depending
or, if K - 1 exists:
on whether the reference frame is of rotating or inertial
type. The dynamic behaviour of bearings and seals is
assumed linear in the displacements and velocities and
thus modelled by stiffness and damping matrices of
arbitrary form. These matrices are decomposed in their
symmetric and antisymmetric parts: with λ= 1/p.
This is a classical eigenvalue problem:

with A non-symmetric and x T =[y T λy T ]. The eigenvalues

and eigenvectors of (5.11) are calculated using a
in which case Ks generates conservative elastic forces and standard eigenvalue extraction algorithm such as Francis'
Kas gives rise to non-conservative circulatory forces. In a QR method 9 .
similar manner, B as is associated with conservative gyro­ When the structure contains supporting parts de­
scopic forces while B s corresponds to dissipative damping scribed by arbitrary stiffness and damping matrices, its
forces8. analysis involves the solution of a series of eigenvalue
The stiffness of the bearings and supporting parts may
Downloaded by UNIVERSITY OF EXETER At 01:20 02 November 2015 (PT)

problems for increasing values of the rotating speed Ω.

be characterized by their strain energy: Instability occurs when the real part of an eigenvalue
becomes positive.
Besides the determination of the instability threshold, a
sweep with increasing rotation speed leads to the de­
and the damping and gyroscopic forces that they generate termination of critical speeds. In a broad sense, a critical
are introduced through a virtual work expression: speed is defined as a rotation speed Ω at which the
frequency of the vibration motion is a multiple of Ω. A
potential source of excitation of the vibration is then
provided by the rotation motion. It is well known that two
From the application of Hamilton's principle in terms of kinds of critical speeds may occur. Aforward critical speed
inertial quantities: occurs when the corresponding eigenmode corresponds
to a forward precession motion (as viewed by a fixed
observer) characterized by a precession velocity equal to
Ω. If, on the other hand, the motion as seen by a fixed
observer is a backward precession at velocity Ω, the
corresponding velocity is a backward critical speed. For­
ward critical speeds may be interpreted as a fundamental
one deduces the homogeneous equations of dynamic instability corresponding to stationary deflection of the
equilibrium expressed in the inertial frame: system when observed by an observer attached to the
rotating frame.Their cause of excitation is unbalance, and
the damping in the system has no reduction effect on the
amplitude of the associated motion. Backward critical
which can be rewritten in the more compact form: speeds, on the other hand, correspond to a vibration
resonance in the rotating frame. They are unlikely to
appear with isotropic supports, and their effect is reduced
by internal damping. In general, forward critical speeds
where M is symmetric and positive definite, while B and K may thus be expected to appear in a much more violent
are arbitrary matrices. manner. It is fully confirmed by experiment.
Similarly, if one assumes that the supporting parts have
isotropic stiffness properties, the application of Let us consider the restrictive case of an undamped
Hamilton's principle leads to the dynamic equations of system with no stiffness coupling between degrees of
motion in the rotating frame: freedom corresponding to x and y deflections. It is then
possible to avoid the classical sweeping procedure and
determine the critical rotation speeds directly1,10. To this
purpose the dynamic equation of motion in the rotating
In a general manner, the structure is thus described in frame may be split in terms of x and y motions:
both representation frames in the form of (5.6). Let us
suppose that the displacements take the exponential
form:

which complex amplitude y and time constant p =α+ iω.

60 Eng. Comput., 1984, Vol. 1, March

 Finite element modelling of flexible rotors: M. Geradin and N. Kill

where qx and qy collect the degrees of freedom correspond­ value analysis is performed either by the Lanczos
ing to displacements in the transverse planes Oxz and Oyz algorithm 11 or using a subspace iteration scheme.
respectively. Because of the assumptions of zero damping When the inertia frame approach is adopted, the system
and no stiffness coupling between qx and qy, the eigen­ (5.14) can also be solved to determine the critical speeds
values are purely imaginary. One may thus assume a directly. If fixed parts and bearings are introduced,
motion of the form: however, it is necessary to perform the full stability
analysis through a sweeping procedure on the rotation
speed. In the latter case the eigenproblem (5.11) has
complex eigensolutions and the eigenvalue extraction
method is a bi-iteration combined with the QR algorithm
for solving the interaction eigenvalue problem12. The finite
in which case the eigenvalue problem associated with eqn element model can be defined in terms of either beam or
(5.12) is rewritten in the form: axisymmetric solid elements. The 3-dimensional model
gives rise, of course, to a much larger number of degrees of
freedom, and a component mode representation of the
system is always adopted in that case to reduce the size of
where X is the proportionality factor between vibration the initial system.
frequency and rotation speed, yi is the vector of displace­ An application of interest has been taken from the work
ment amplitudes y T =[u T v T ] in inertial coordinates and of Nelson and MacVaugh 2 . The rotor configuration is
G*i is the symmetric matrix associated to Gi.. The form of described by Figure 6.1 and consists of a flexible shaft
eqn (5.14) is similar to that of the free vibration problem: supported on two bearings. A disk of relatively large
inertia is attached to its overhanging part. The con­
figuration data are given in Table 1. Various models with
Downloaded by UNIVERSITY OF EXETER At 01:20 02 November 2015 (PT)

different characteristics have been considered as sum­

provided that M is replaced by the modified inertia matrix marized in Table 2.
λ 2 M-λG*i. The rotor is a fairly massive construction, and is thus
One is easily convinced that X= + 1 corresponds to appropriate for a comparison between beam and 3-
forward critical speeds while X= — 1 allows to calculate dimensional models. The disk part is also such that the
the critical speeds corresponding to backward precession. assumption of infinite transverse stiffness might be
In a similar manner the rotating frame approach leads unappropriate.
to the eigenproblem In the beam model, the shaft is discretized into 17 beam
elements with shear deformation and rotary inertia effects
of right sections included. The bending shape functions
are cubic, and the disk is assumed infinitely rigid. The
where yr is the vector of displacement amplitudes in resulting model is represented by Figure 6.2.
rotating coordinates, and G*r is the symmetric matrix In the 3-dimensional model, the shaft and the disk are
associated to G r For an observer attached to the rotating discretized in terms of axisymmetric solid elements with
frame, the forward critical speeds are obtained when λ=0 isoparametric shape functions, and developed in Fourier
while the backward critical speeds correspond to X = — 2. series as described in section 4. Only the terms of the series
This direct procedure for computation of critical speeds corresponding to bending have been retained, and they
leads to a significant saving in computing effort by are coupled through the gyroscopic terms. The whole
comparison with the usual sweeping method. rotor model numbers 30 elements and is represented in
Figure 6.3. The axial and radial displacement fields are
either quadratic or cubic. The model with cubic elements
6. APPLICATION numbers 1216 degrees of freedom, and its stability
Software for dynamic analysis of rotating structures has analysis cannot be performed without using the com­
been written and integrated into the finite element system ponent mode method to reduce the size of the system. It
SAMCEF developed by the Aerospace Laboratory of the has thus been decomposed in 4 substructures as indicated
University of Liege, Belgium. The program has the by Figure 6.4. The boundary degrees of freedom are
capability of analysing rotating structures defined either located on adjacent edges and on the rotation axis at
in a rotating or in an inertial frame. bearing locations.
In the rotating frame approach, (5.16) is solved to Let us consider first the analysis in the rotating frame, in
determine critical speeds for structures discretized using which case the calculation of forward critical speeds has
either beam or axisymmetric solid elements. The eigen- been limited to the direct method of (5.16). Forward

Eng. Comput., 1984, Vol. 1 , March 61

 Finite element modelling of flexible rotors: M. Geradin and N. Kill

Table 7 Rotor configuration data critical speeds correspond to λ=0, and the eigenvalue
Discretization of Outer diameter Inner diameter
problem is thus:
Oz axis (m) (m) (m)

0.0 0.0102
0.0127 — 0.0204 By comparison with the free vibration eigenproblem, the
0.0508 — 0.0152 mass matrix is replaced by the centrifugal inertia matrix
0.0762 — 0.0406 deduced from (1.20).
0.1016 — 0.0660
Table 3 compares the three forward critical speeds
0.1067 0.0304 0.0660
0.1143 0.0356 0.0508 obtained from the beam and axisymmetric models (de­
0.1270 — 0.0508 gree 2) to the corresponding vibration frequencies of the
0.1346 — 0.0254 structure at rest. The comparison shows that the axisym­
0.1905 — 0.0304 metric model takes into account correctly the gyroscopic
0.2667 — 0.0254
effects, and that it introduces only a slight additional
0.3048 — 0.0762
0.3150 — 0.0406 flexibility in the structural model since the third critical
0.3454 0.0304 0.0406 speed is decreased by only 3%.
The remaining part of the analysis has been performed
E = 2.078 x 1011 N / m 2 μ = 7806 k g / m 3
Disk location on Oz axis: 0.0889 m
in the inertial frame. In a first step, the same bearing
Disk properties: m = 1.401 kg configuration (case a) has been considered. Again, since
Jp = 2 x 1 0 - 3 k g m 2 the bearing has isotropic properties and no stiffness
Jd = 1.36 x 1 0 - 3 k g m 2 coupling terms, it is possible to calculate the critical
speeds directly from (5.14).
Table 2 Bearing configuration cases The beam model gives the same forward critical speeds
Downloaded by UNIVERSITY OF EXETER At 01:20 02 November 2015 (PT)

2 identical bearings
as in the previous analysis, but the axisymmetric model
location on Oz a x i s : 0.1651 m and 0.287 m has now degree 3. It is thus necessary to apply the
component method to reduce its size. In each substruc­
Case A isotropic bearings, no damping ture, the boundary degrees of freedom have been kept
Kxx = Kyy = 4.378 x 1 0 7 N/m and Kxy = Kyx = 0 together with 0,12 and 14 normal modes successively. The
Case B orthotropic bearings, no damping
calculated critical speeds are given in Table 4. One
Kxx = Kyy = .0503 x 107 N/m and Kxy = Kyx = observes that the degree 3 model is somewhat more
8.756 x 106 N/m flexible, and that a slightly larger discrepancy between
axisymmetric and beam models occurs for backward
Case C isotropic bearings, w i t h damping critical speeds. It is also of interest to note that 12 normal
Kxx = Kyy = 4.378 x 1 0 7 N/m and Kxy = Kyx = 0
Bxx = Byy = 2627 Ns/m and Bxy = Byx = 0
modes in the component mode representation are quite

62 Eng. Comput., 1984, Vol. 1 , March

 Finite element modelling of flexible rotors: M. Geradin and N. Kill

sufficient to give accurate results. The model with zero between x and y directions. A plot of eigenfrequencies as
normal modes has variable accuracy, the discrepancy functions of rotation speed Ω is represented in Figure 6.6.
being the largest for the second backward and forward The critical speeds are obtained using the same structural
critical speeds. This is due to the fact that the reduced models as previously, and are given in Table 5. When
model has an insufficient number of d of in the region of comparing the critical speeds to the vibration frequencies
substructure IV where most of the kinetic energy is of the structure at rest one observes that the gyroscopic
concentrated for this mode. This is illustrated by Figure effect is far from being negligible. The backward critical
6.5 where a perspective view of the corresponding modal speeds decrease with rotation speed while forward critical
shape is given for the forward precession motion. speeds increase. It is also worthwhile noticing that the
In order to calculate the critical speeds of the rotor with difference between eigenfrequencies and critical speeds on
the bearing characteristics of case (b), it is necessary to one hand, and between beam-like and axisymmetric
perform a sweep on rotation speed since coupling occurs models on the other hand increase with frequency.
Finally, Table 6 gives the forward and backward
Table3 Vibrationfrequenciesandforwardcriticalspeeds-case(a)
critical speeds for case (c). They are fairly close to the
results of case (a). The damping terms have only a slight
Eigen- Forward critical speeds (rpm)
frequencies
(Ω = 0) Axisymmetric model Beam MacVaugh
(c/min) (degree 2) model model

16 119 17 035 17 148 17159

47 761 49 378 5 0 123 49 983
75 155 93 481 95 882 96 457
Downloaded by UNIVERSITY OF EXETER At 01:20 02 November 2015 (PT)

Table 4 Critical speeds - c a s e (a) (rpm)

Axisymmetric model

40 normal 12 normal 0 normal

modes modes mode Beam model

B 14 987.3 14 987.3 14 984 15 486.32

F 17 163.5 17 163.5 17 173.9 17 148.04
B 45 570.1 45 570.1 51 248.6 46 744.95
F 49 596.6 49 596.7 53 232.3 50 122.69
B 62 450.3 62 452.4 62 500.7 64 857.56
F 92 219.1 92 219.8 94 094.4 95 881.52

Eng. Comput., 1984, Vol. 1, March 63

 Finite element modelling of flexible rotors: M. Geradin and N. Kill

Table 5 Critical speeds - case (b) (rpm)

Structure at rest Axisymmetric model

axisymmetric model
(c/min) 40 normal modes 12 normal modes 0 normal mode Beam model

B 13 990.4 13 724.8 13 724.8 13 727.6 13 875

F 16 029.3 16 340.0 16 340.0 16 342.4 16 545.1
B 39 620.1 39 383.4 39 383.4 41 938.7 39 993.6
F 47 605.6 47 647.7 47 647.8 52 046.8 48 375.9
B 65 661.4 57 816.5 57 721.2 58 180.1 59 716.7
F 74 661.9 88 765.5 88 766.3 91 309.7 92 585.8

Table 6 Critical speeds - case (c) (rpm)

Structure at rest Axisymmetric model (3d degree)

axisymmetric model
(c/min) 4 0 normal modes 12 normal modes 0 normal mode Beam model

16 022.05 15 211.4 15 211.4 15 214.3 15 501.7

16 022.05 16 955.2 16 955.1 16 9 58.8 17 168.8
47 574.39 45 659.2 45 659.3 51 162.4 46 950.8
47 574.39 49 681.5 49 681.6 52 916.7 50 254.4
74 621.2 61 762.1 61 762.2 61 675.7 64 039.3
74 621.2 92 806.8 92 807.7 95 010.6 96 365.9
Downloaded by UNIVERSITY OF EXETER At 01:20 02 November 2015 (PT)

ties, one is led to systems of equations with periodic

coefficients. They require adequate methods of solution
which are still to be developed; (3)a refinement in the
rotor model should be accompanied by a similar refine-
ment in the modelling of supporting parts (bearings and
seals in particular).

REFERENCES
1 Childs, D. W. and Graviss, K. A note on critical-speed solutions
for finite-element-based rotor models, J. Mech. Design (ASME),
104,412-416(1982)
2 Nelson, H. D. and MacVaugh, J. M. The dynamics of rotor-
bearing systems using finite elements, J. Eng. Ind. (ASME)
593-600 (1976)
3 Jacker, M. Vibration analysis of large rotor-bearing-foundation-
influence on eigenfrequencies. The modal damping coef- systems using a modal condensation for the reduction of
unknowns, Vibration in Rotating Machinery, I. Mech. E., 195-202
ficients are plotted as a function of rotation speed in (1980)
Figure 6.7. The first three modes of the system remains 4 Tondl, A. Some Problems of Rotor Dynamics, Chapman & Hall,
stable in the speed domain analysed since all the coef- London (1965)
ficients remain negative. 5 Lalanne, M., Ferraris, G.,Tran, D. M., Queau, J. P. and Berthier,
P. Comportement dynamique des rotors de turbomachines,
Rapport final DRET, (1982)
CONCLUSIONS 6 Dubigeon, S. and Michon, J. C. Utilisation de la méthode des
elements finis pour les calculs des pulsations propres d'arbres en
A formalism has been proposed which allows a 3- tenant compte de l'effet gyroscopique, Revue Fr. Méc. (53), 25-30
dimensional modelling of rotating machinery for its (1975)
7 Craig, R. R. and Bampton, M. C. Coupling of substructures for
critical speed and stability analysis. Its implementation dynamic analyses, AIAA J. 6, (7), 1313-1319 (1968)
has been achieved in the case of rotors with revolution 8 Adams, M. L. and Padovan, J. Insights into linearised rotor
symmetry using axisymmetric elements, and its adequacy dynamics, J. Sound Vibration, 76(1), 129-142 (1981)
has been demonstrated on a numerical example. 9 Wilkinson, J. H. The Algebraic Eigenvalue Problem, Clarendon
Further developments of this approach to rotor dy- Press, Oxford, (1965)
10 Geradin, M. Sur les vitesses critiques de lignes d'arbres, Revue-M,
namics should also include the following aspects: (l)for SBM, 25, (2) (1979)
rotors which do not have the revolution symmetry but 11 Carnoy, E G. and Geradin, M. On the practical use of the
keep some form of cyclic symmetry, the principles of the Lanczos algorithm in finite element applications to vibration and
present analysis should be combined with the methods of bifurcation problems, Proc. Conf. "Matrix Pencil" Pitea, Sweden,
(1982)
analysis applicable to periodic structures; (2)in cases 12 Bauer, F. L. Das Verfahren der Treppeniteration und verwandte
where the rotor and thefixedparts present simultaneously Verfahren zur Lösung algebraischer Eigenwertprobleme, ZAMP
some anisotropy in this geometric and material proper- VIII 214-233 (1957)

64 Eng. Comput., 1984, Vol. 1 , March

 This article has been cited by:

1. Zhong-xiu Fei, Shui-guang Tong, Chao Wei. 2013. Investigation of the dynamic characteristics of a dual rotor system and its
start-up simulation based on finite element method. Journal of Zhejiang University SCIENCE A 14, 268-280. [CrossRef]
2. C. Carmignani, P. Forte, G. Melani. 2012. A Duhamel approach for forced vibrations of rotating axisymmetric bodies.
International Journal for Numerical Methods in Engineering 90:10.1002/nme.v90.7, 928-938. [CrossRef]
3. A.S. Das, J.K. Dutt. 2012. A reduced rotor model using modified SEREP approach for vibration control of rotors. Mechanical
Systems and Signal Processing 26, 167-180. [CrossRef]
4. D. Combescure, A. Lazarus. 2008. Refined finite element modelling for the vibration analysis of large rotating machines:
Application to the gas turbine modular helium reactor power conversion unit. Journal of Sound and Vibration 318, 1262-1280.
[CrossRef]
5. A.S. Das, J.K. Dutt. 2008. Reduced model of a rotor-shaft system using modified SEREP. Mechanics Research Communications
35, 398-407. [CrossRef]
6. Eric Chatelet, Flavio D’Ambrosio, Georges Jacquet-Richardet. 2005. Toward global modelling approaches for dynamic
analyses of rotating assemblies of turbomachines. Journal of Sound and Vibration 282, 163-178. [CrossRef]
7. G. Genta, C. Delprete, D. Bassani. 1996. DYNROT: a finite element code for rotordynamic analysis based on complex co‐
ordinates. Engineering Computations 13:6, 86-109. [Abstract] [PDF]
8. G. Genta. 1992. A fast modal technique for the computation of the Campbell diagram of multi-degree-of-freedom rotors.
Downloaded by UNIVERSITY OF EXETER At 01:20 02 November 2015 (PT)

Journal of Sound and Vibration 155, 385-402. [CrossRef]

9. G. Genta, F. De Bona. 1990. Unbalance response of rotors: A modal approach with some extensions to damped natural
systems. Journal of Sound and Vibration 140, 129-153. [CrossRef]
10. R.W. Stephenson, K.E. Rouch, R. Arora. 1989. Modelling of rotors with axisymmetric solid harmonic elements. Journal of
Sound and Vibration 131, 431-443. [CrossRef]
11. G. Genta. 1988. Whirling of unsymmetrical rotors: A finite element approach based on complex co-ordinates. Journal of
Sound and Vibration 124, 27-53. [CrossRef]