Proceedings of ASME Turbo Expo 2013: Turbine Technical Conference and Exposition

GT2013
June 3-7, 2013, San Antonio, Texas, USA

GT2013-95196

CONSIDERATION OF COMPLEX SUPPORT STRUCTURE DYNAMICS IN

ROTORDYNAMIC ASSESSMENTS

Thomas D. Krüger, Sauro Liberatore, Eric Knopf, Alastair Clark

ALSTOM (Switzerland) Ltd.
5401 Baden
Switzerland

ABSTRACT NOMENCLATURE
In rotordynamic analyses, support structures are commonly
represented by lumped mass systems (single-degree-of-freedom, A state matrix
SDOF). This representation is easy to implement using standard an coefficients of denominator polynom
rotordynamic tools. However, in reality the dynamic behaviour B input matrix
of the support structure (e.g. pedestals, casings, foundations) are bm coefficients of numerator polynom
in general much more complex. Only a multi-degree-of-freedom
C output matrix
(MDOF) representation provides modelling close to reality.
For many applications the dynamic behaviour of the support D direct matrix
structure significantly influences the rotordynamic characteris- D(s) denominator
tics of the shaft train and therefore needs to be included in the F force
assessment. Due to this impact, a good quality of the dynamic H frequency response function (FRF)
model used for the support structure is imperative. Regarding H matrix of FRFs
the rotor itself, the modelling is well understood and the predic- I identity matrix
tion quality is excellent, not least due to the jointless welded rotor √
j −1
design.
Numerous theoretical approaches exist for considering the K stiffness matrix
complex dynamic behaviour of the support structure, all com- M mass matrix
ing along with both drawbacks and opportunities. By discussing N(s) numerator
the characteristics of established approaches for modelling the p modal state vector
support structure, the paper particularly presents an advanced q physical displacement
theoretical approach based on a state-space representation us-
s jΩ
ing modal parameters.
A case study of a real shaft train is shown, including a u input vector
comparison of achieved results using the SDOF and the pre- x state vector
sented MDOF approach. By validating with experimental re- y output vector
sults, the excellent prediction quality of the MDOF approach is Z damping matrix
confirmed. The implementation of this approach enabled to fur-
ther improve the reliability and the efficiency, which means high
accuracy combined with low computation time, in performing
rotordynamic assessments.

Information contained in this document is indicative only. No representation or warranty is given or should be relied on that it is complete or correct or will apply to any
particular project. This will depend on the technical and commercial circumstances. It is provided without liability and is subject to change without notice.
Reproduction, use or disclosure to third parties, without express written authority, is strictly prohibited.

1 Copyright © 2013 by Alstom Technology

 Z̃ modal damping matrix
γm Hankel singular value
ζm modal damping
Φ modal matrix (mode shapes)
ϕmk , ϕml mode shape components
Ω angular frequency
Ω̃20 modal stiffness matrix
ωm eigenfrequency
INTRODUCTION
The dynamic behaviour of turbine sets is determined by the
characteristics of the rotating system itself, by the dynamic be-
haviour of the bearing (and seal) elements (fluid-structure inter-
action) and by the supporting structure (rotor-casing interaction,
rotor-foundation interaction). Looking at the application of a
heavy duty gas turbine (see Figure 1), the dynamic behaviour
of the rotor system can be significantly influenced by the gas
turbine casing and support elements, which leads to changes of
critical speeds, to differences in the forced response behaviour
and therefore in the operational behaviour of the shaft train. Par-
allel to the increasing market demands on operational flexibil-
ity, the rotordynamic assessment procedures have to react to this
by providing appropriate methods and tools to fulfill these de-
mands. The dynamic behaviour of the gas turbine casings and
support structure is usually assessed by large 2D or 3D finite ele-
ment models. The dynamic behaviour of the GT casings is quite
complex, showing numerous eigenmodes in the frequency range
of interest (multi-degree-of-freedom (MDOF) behaviour), inter-
acts with the rotor system at the bearing positions and therefore
needs to be included in the rotordynamic assessment. The paper
presents in general two approaches to include the MDOF char-
acteristics in the rotordynamic assessment:
compared to the FE model incorporating the complete cas-
ing.
• Both approaches are very well suited for sharing data be-
tween different disciplines, like structure dynamics, civil en-
gineering and rotordynamics. This ensures a smooth and
constructive interdisciplinary work.
The MDOF approach using modal parameters has been in-
troduced and established in the rotordynamic assessment with
the introduction of a new generation of GT in the Power Plant
Birr (Switzerland) configuration. In the following sections, the
individual approaches are described in detail, whereas the usu-
ally applied single-degree-of freedom (SDOF) approach serves
as a starting point and as an introduction to the topics. The use of
lumped mass systems (SDOF) for modelling the substructure dy-
namics of a rotor system is well-established since decades in ro-
tordynamic analyses. The SDOF approach is advantageous with
respect to its simplicity. Physically, its application is reasonable
as long as only one natural mode of the substructure interacts
significantly with the rotor in the frequency range of interest or
if the interaction is tolerably small.
However, depending on the complexity of the substructure
the SDOF approach may suffice only as an approximate descrip-
tion of the real dynamic behaviour. If several natural modes are

• description via frequency response functions (FRFs)

• description via state space representation, preferably in a
modal form

Both approaches are straightforward, whereas the state space

representation is more sophisticated and enables to perform any
typical rotordynamic calculation. The input data for both repre-
sentations can be derived from finite element models (see Figure
1) or from experimental modal analysis
Besides some differences in the capabilities (see subsequent
sections) both MDOF approaches share the following advan-
tages:

• The dynamic behaviour of the substructures is described by

a small amount of data. Therefore commercial and tailored
rotordynamic tools can be used ensuring a low computation
time. FIGURE 1: STRUCTURAL DYNAMICS FE MODEL (TOP) AND
• The accuracy of the results for the rotor system is identical FE GT CASING MODEL (TAKEN FROM [1])

2 Copyright © 2013 by Alstom Technology


FIGURE 2: DIFFERENCE BETWEEN USING SDOF AND MDOF
SYSTEMS FOR MODELLING THE SUPPORT STRUCTURE
present in the operating range featuring high interaction with the
rotor or if there is a high spacial coupling e.g. between different
bearing locations over the support structure, the approximation
might not be sufficient to reflect reality. For example, Figure 2
shows a FRF at one degree of freedom of a GT support system,
defined as dynamic compliance
q(Ω)
FRF : H(Ω) = ; [H] = m/N , (1)
F(Ω)
i.e. the ratio between dynamic displacement q(Ω) and force
F(Ω) dependent on frequency Ω. Assuming the real behaviour to
be represented by the multi-degree-of-freedom (MDOF) system,
the SDOF approach accommodates only a single mode. This can
be an acceptable approximation for a specific speed, but high de-
viations may need to be accepted for wide speed ranges. More-
over, with a SDOF model no interaction between different loca-
tions or degrees of freedom can be described. Only a MDOF
representation provides the opportunity of modelling close to re-
ality in the case of complex dynamic behaviour, characterised by
• multiple natural modes,
• cross-coupling effects, i.e. interaction between horizontal
and vertical direction, and
• cross-talking effects, i.e. interaction between several bearing
positions.
The example shown in Figure 2 emphasises the need for us-
ing a MDOF description of the support structure. Particularly
if the mentioned characteristics are present, MDOF modelling
3 Copyright © 2013 by Alstom Technology
FIGURE 3: DEFINITION OF A FRF AS LOOKUP TABLE
can be considered as an essential necessity in rotordynamic anal-
yses of high reliability. However, a MDOF description of the
support structure apparently comes along with much higher com-
plexity in the mathematical description than the SDOF approach.
Nevertheless, the mathematical description should be as simple,
concise, consolidated and systematic as possible. Preferably, it
simultaneously features the capability of performing all
• the analysis of natural behaviour,
• unbalance response calculations and
• transient calculations.
Furthermore, since complex structures consist of several compo-
nents, usually more individual disciplines are involved in provid-
ing inputs for the rotordynamic analysis. Therefore the MDOF
description should allow for the demanding need of smooth in-
terdisciplinary working.
A method of modelling the support structure has been es-
tablished, helping to meet these manifold requirements. For this
method, natural frequencies and mode shape components are the
only input needed. They are used for formulating a state space
system as constitutive mathematical description which is pro-
vided as input to the rotordynamic model.
Frequency Response Functions: Lookup Tables
FRFs represent the state-of-the-art approach of considering
complex dynamic behaviour of the support structure in rotordy-
namics. Multiple natural modes, cross-coupling as well as cross-
talking effects can be considered. Today’s rotordynamic soft-
ware is able to deal with a support structure defined via FRFs.
Usually lookup tables are used as input to define the FRFs, e.g.
with data columns containing frequency, amplitude and phase
(see Figure 3 as an example).
Figure 4 illustrates the use of FRFs in the rotordynamic
model of a combined cycle power plant shaft train equipped with
a gas turbine. The GT rotor is supported by a casing. The casing
structure exhibits complex dynamic behaviour and interacts sig-
nificantly with the rotor. With a 3D finite element model of the
casing a forced response analysis has therefore been performed
in order to derive FRF lookup tables. These have been attached
at the bearing positions to the rotordynamic model. Unbalance
FIGURE 4: CONSIDERING GASTURBINE CASING DYNAMICS
IN THE ROTORDYNAMIC MODEL
response predictions have been performed using this approach.
One advantage of lookup tables is the inherent flexibility of
the definition. Any input data regardless of the source – calcu-
lations or measurements – can be used. However, the analytical
capabilities are restricted to harmonic response calculations only.
Neither an eigenvalue analysis nor transient calculations can be
obtained by using FRF lookup tables. A parametric description
is needed instead.
Frequency Response Functions: Polynomial Fits
One solution of getting a parametric description of the shaft
support dynamics is to apply polynomial fits to the FRF lookup
tables. Fundamentally, a FRF can be described by the complex
rational function
M vides for good fit quality in a least squares sense only, one phys-
∑ (bm sm )
N(s) m=0
H(s) = = N
; s = jΩ (2)
D(s)
∑ (an sn )
n=0
consisting of a numerator polynomial N(s) and a denominator
polynomial D(s) with m constant coefficients bm and n constant
coefficients an respectively. Algorithms based on the least square
method exist [2, 3] in order to estimate the coefficients bm and
an from a lookup table H(Ω) representing H(s). Each individ-
ual FRF of the dynamic support description needs separate treat-
ment, including
• defining the orders of the numerator and the denominator
polynomial,
• defining the frequency range to be covered,
• a rating of the fit quality
• repeating all the steps until a satisfying quality of the fit is
reached.
4 Copyright © 2013 by Alstom Technology
FIGURE 5: POLYNOMIAL FIT OF FRF: EXAMPLE RESULT
Manual Polynomial Fits
Figure 5 shows an exemplary result of a polynomial fit with
a numerator order of 2 and a denominator order of 4 with the full
frequency range having taken into account. The fit quality can be
rated as non-satisfying, so redoing the fit is required. The man-
ual fit process is evidently based on trial and error, can be rather
time consuming and needs sound user experience. It involves po-
tential risks, e.g. partly questionable results and numerical prob-
lems, especially with high polynomial orders requested.
Another side effect of performing individual polynomial fits
manually consists in similar poles showing up for the FRFs.
Physically the structure has as many conjugate complex pole
pairs as natural modes. Hence the number of poles describing
the structure dynamics is limited. Since the fit algorithm pro-
ical pole may show up as several numerical poles close to each
other. These scattered poles (Figure 6) will occur later on in the
rotordynamic eigenvalue analysis again. The user has to decide
which eigenvalues are reasonable and which can be disregarded
– again based on experience and with additional effort.
Automatic Polynomial Fitting
The described manual process has been made automatic in
order to tackle the drawbacks coming along with it. The auto-
matic process has been implemented in a software tool providing
a stepwise increase of the orders for the numerator and denom-
inator polynomials. All physically meaningful order combina-
tions are performed, i.e. polynomial orders of equation (2) must
satisfy M ≤ N − 2 for bound mechanical structures. The quality
of each individual fit is tested by calculating the correlation be-
tween the source FRF and its estimated polynomial expression.
At the end of all fits, the solution providing the highest correla-
tion is automatically chosen.
 FIGURE 6: SCATTER OF ESTIMATED POLES
(a) FRF MAIN DIAGONAL

In order to avoid similar poles, a pole check has been imple-

mented. All poles found within a user specified tolerance circle
(see Figure 6, uncertainty range) are considered as one pole. Its
value is determined by the mean value of all poles located in the
tolerance circle. The scattering poles are replaced by the mean
value.
Figures 7 and 8 show exemplary results of the automated
polynomial fitting procedure. The results in Figures 7(a) and
7(b) have been achieved by a user specified maximum polyno-
mial order of 4. Regarding the FRF on the main diagonal of the
transfer matrix (Figure 7(a)), two main peaks have been identi-
fied very well. However, due to the low maximum order, some
peaks particularly in the low frequency region are missed. A
minor fit quality has been achieved on the cross-coupling FRF
(b) CROSS-COUPLING FRF
shown in Figure 7(b). Not only many peaks are missed but also
identified peaks have been estimated poorly.
Figures 8(a) and 8(b) illustrate the result of fitting the same FIGURE 7: AUTOMATIC POLYNOMIAL FITS OF FRFS,
FRFs, but with a maximum admitted order of 72. As expected MAXIMUM ORDER 4
the final correlation between the source FRF and the fit gets
much higher (99.9 % on the main diagonal, 98.5 % on the cross-
coupling FRF). In order to achieve this result, polynomial orders
with the vector of inputs u, the vector of states x and the vector
not higher than 34 were needed.
of outputs y. The state matrix A describes the system’s natural
Another way to avoid multiple poles is to firstly identify behaviour, the input matrix B the influence of inputs onto the sys-
one denominator common to all FRFs and secondly fit the nu- tem, the output matrix C the influence of the system states onto
merators individually. This approach has been discussed by the outputs and the direct matrix D the proportionality between
R ICHARDSON and F ORMENTI [4]. inputs and outputs. Equation (3) represents a system of first order
differential equations in time.
Any differential equation of N-th order in time can be con-
State Space Representation verted to the form given by equation (3). Hence, converting the
Another approach for getting a parametric description of the second order differential equation of motion
support dynamics can be based on the use of the state space rep-
resentation. Generically a state space system is defined by
Mq̈ + Zq̇ + Kq = F (4)

ẋ = A x + B u
(3)
y = Cx + Du with mass-, damping- and stiffness terms M, Z and K respec-

5 Copyright © 2013 by Alstom Technology

 by

H(Ω) = C ( jΩ I − A)−1 B + D , (6)

while considering the equation of motion (4) the transfer be-

haviour is given by

−1
K − Ω2 M + j (Ω Z)


H(Ω) = . (7)

Equations (6) and (7) give an entirely identical result for the
FRFs H(Ω). In order to retrieve FRFs in the form proposed,
the system matrices defined in physical coordinates – or at least
(a) FRF MAIN DIAGONAL part of their entries – must be known. Particularly for the state
matrix A in equation (5) the full system matrix M is required due
to the need for its inversion. The feasibility and the handiness of
this approach is restricted, since usually 3D finite elements are
used for modelling the support structure coming along with big
system sizes. Restrictions also exist, if experimental source data
is available only.

The Benefit Of Using Modal Parameters

Equation (5) uses physical coordinates as states. By per-
forming an eigenvalue analysis on the equation of motion (4),
its eigenvalues λ and the modal matrix Φ containing the eigen-
vectors can be determined. Applying the modal transformation
q = Φp with physical coordinates q and modal coordinates p
(b) CROSS-COUPLING FRF equation (4) can be converted into its modal form

FIGURE 8: AUTOMATIC POLYNOMIAL FITS OF FRFS, (ΦT MΦ) p̈ + (ΦT ZΦ) ṗ + (ΦT KΦ) p = ΦT F (8)
| {z } | {z } | {z }
MAXIMUM ORDER 72
1 Z̃ Ω̃20

tively into the state space representation yields assuming the modal matrix Φ to be mass normalised. Ω20 is a
diagonal matrix containing the squares of the eigenfrequencies.
ẋ A x B u If the damping of the system can be described proportional to
z}|{ z }| { z}|{ z }| { z}|{ M and K, Z̃ is diagonal as well, containing the real parts of the
  
q̇ 0 I
  
q 0
 eigenvalues λ . In equation (8) the system is described modally
= + F decoupled. Once the eigenfrequencies ωm and the modal matrix
q̈ −M−1 K −M−1 Z q̇ M−1
(5) Φ are known, the FRF from input l to output k can be calculated
  
 
 q   by the modal superposition
q = I 0 + 0 F
q̇
|{z} | {z } |{z} | {z } |{z} N
y x u ϕmk ϕml
C D Hkl (Ω) = ∑ , (9)
m=1 (ω 2 − Ω2 ) +
m j (2 ζm ωm Ω)

by determining forces F as inputs and displacements q as out-

puts. Considering the state space representation (3), the transfer making use of the N mode shape components ϕml at input l and
behaviour between its inputs u and outputs y in general is defined ϕmk at output k. ζm represents the modal damping of mode m.

6 Copyright © 2013 by Alstom Technology

The state space representation in modal form of system (8) be- time consuming and error prone. This should be avoided, auto-
comes matic procedures are recommendable instead. Automatic proce-
dures for reducing the number of considered modes are presented
  
ṗ 0 I
   
p 0 in the later section about System Reduction.
= + T F
p̈ −Ω̃20 −Z̃ ṗ Φ
(10)
  Case study
 p
The following section shows a case study of a new gen-
    
q = Φ 0 + 0 F .
ṗ eration GT where the FRFs have been introduced and applied
for considering the GT casing dynamics in the rotordynamic
The only difference between the state space systems (5) and assessment of the Power Plant Birr in Switzerland (see Fig-
(10) consists in the definition of their state vectors, using physi- ure 4). For this shaft train configuration, significant rotor-
cal q and modal p respectively. The described dynamic properties structure-interaction was expected.
stay the same, the forces F (inputs) and the displacements q (out- As mentioned, the casing exhibits rather complex dynamic
puts) as well. Both inputs and outputs stay in physical represen- behaviour, including 64 modes within the frequency range of in-
tation even in the modal form (10). Therefore, also the transfer terest, cross-coupling as well as significant cross-talking between
behaviour determined by equation (6) remains unchanged in the the bearings. A 3D finite element model of the casing has been
case of using state space system (10). Furthermore, not the full used to properly estimate its eigenfrequencies and mode shapes.
modal matrix Φ is needed to build the state space system, but The casing interacts with the shaft train via two radial journal
only the mode shape components at the in- and outputs (here: bearings. For each bearing, the casing dynamics horizontally
forces/displacements respectively). and vertically has been taken into account. These 4 degrees of
Besides the representation shown here, other definitions of freedom require 10 individual FRFs1 (see also Figure 12) for de-
state space systems exist. S CH ÖNHOFF ET AL . [5] present a sam-
ple of these definitions and discuss their individual properties
in further detail. For a more generic discussion about capabili- 1 due to the symmetry of the transfer matrix
ties and fields of application for state space formulations see e.g.
G AWRONSKI [6].
A couple of benefits come along with the use of modal para-
meters. The source can be either a calculation or an experimental
modal analysis. Using finite element models, an eigenvalue cal-
culation requires significantly less computational time compared
to harmonic response calculations. A state space representation
can be formulated. If needed, FRFs as lookup tables are deter-
minable via either equation (6) or equation (9). In contrast to
the polynomial fit method, the use of modal parameters ensures
physically entirely meaningful results in all rotordynamic analy-
ses. Additionally the comparatively high efforts for performing
polynomial fits are avoided. System reductions can be easily per-
(a) 1st MODE SHAPE
formed by using the modal state space representation (10). Fur-
thermore, modal parameters represent a simple and unique data
format ensuring an efficient interdisciplinary exchange. Together
with the use of the state space representation the approach is very
well formalised and therefore easy to handle. Used in the rotor-
dynamic model, it is applicable to eigenvalue analyses as well as
to unbalance and transient response calculations.
The number of modes for typical support structures like cas-
ings or foundations can be rather high. However, by using the
modal representation even a high number of modes can be han-
dled with low numerical effort. The result is then identical com-
pared to the source model, e.g. FE model of the GT casing. The
(b) 2nd MODE SHAPE
effort for the user is minimised, particularly a manual selection
between significant and unimportant modes is not necessary. The
experience shows that especially the manual selection can be FIGURE 9: SDOF SUPPORT, CALCULATED MODE SHAPES AT
CRITICAL SPEEDS

7 Copyright © 2013 by Alstom Technology

 (a) 1st MODE SHAPE
(b) 2nd MODE SHAPE
FIGURE 10: MDOF SUPPORT, CALCULATED MODE SHAPES
AT CRITICAL SPEEDS
scribing all cross-coupling and cross-talking terms. Instead of
generating FRF lookup tables and fitting polynomials, the modal
state space representation (10) has been used for the dynamic
analysis. A speed dependent eigenvalue analysis was performed
resulting in a Campbell diagram and the prediction of critical
speeds. The comparison is shown exemplarily for the first and
second critical speed of the GT shaft.
The mode shapes of these critical speeds are shown in Fig-
ures 9 and 10, whereas Figure 9 represents the SDOF casing
model and Figure 10 shows the mode shapes for the MDOF
approach. It becomes evident that the differences in the mode
shapes between the two support models are comparatively small,
since these modes are shaft dominated.
The higher quality of the MDOF support model becomes
obvious when comparing the calculated critical speeds with the
ones identified in the Power Plant. The relative deviations are
shown in Figure 11. Using a SDOF model for the support struc-
ture, deviations up to 17 % need to be accepted, whilst the MDOF
model provides maximum deviations from the identified values
around 2 % which is remarkable for such a complex system.
If the complex support structure dynamics are considered
properly, the noticeable effect of dynamic interaction between
the rotor and its support is taken into account and results get
8 Copyright © 2013 by Alstom Technology
FIGURE 11: DEVIATIONS IN CRITICAL SPEEDS
closer to reality. Especially, the sound representation of the
phase characteristics (see Figure 2) and the cross-talking (see
Figure 12) plays a significant role and contributes to the im-
proved prediction.
System Reduction
In the case study the high number of modes to model the
support structure has been mentioned. Many of the modes can
be less significant for the overall rotordynamic behaviour e.g.
because of their local character. But each mode considered to
describe the support structure dynamics will appear as mode in
an eigenvalue analysis of the coupled system as well, regardless
of it’s relevance for assessing the coupled system behaviour. This
fact emphasizes the benefit of reducing the system.
Deriving proper FRFs from equation (9) by involving all
modes is straightforward. However, getting equivalent FRFs
based on a reduced set of modes requires additional consider-
ation. A reduced system should be as small as possible and –
to some degree contradictory – still reflect the system dynam-
ics realistically. In modal representation a system reduction can
be achieved by truncating modes. In order to do this, limiting
the frequency range of interest is the first step. However, if for
instance runup behaviour of the shaft train is of interest, limit-
ing the frequency range may remain pointless. Rather adjuvant
instead is to identify relevant modes and neglect the irrelevant
ones. One approach in this manner is offered e.g. by evaluating
Hankel singular values, representing a measure for the energy
each mode adds to the system dynamics. Determining Hankel
singular values in an exact manner is time-consuming [6], espe-
 FIGURE 12: EXAMPLE OF SYSTEM REDUCTION USING HANKEL SINGULAR VALUES
cially for big systems, but
||Bm || · ||Cm ||
γm = (11)
4ζm ωm
provides a sufficient approximation. Having defined system dy-
namics by a modal state space representation, equation (11) eas-
ily can be evaluated. Bm and Cm are the parts of the input and
output matrix referring to the m-th mode, || · || denote their ma-
trix norms. For each individual mode one Hankel singular value
γm is determinable. Low values indicate insignificance of the ac-
cording mode shape.
The capability of applying Hankel singular values for sys-
tem reduction is illustrated in Figure 12 using the example of the
GT casing (cf. Figure 4). The plot is the result of an automatic
procedure driven by achieving a target correlation between the
full and the reduced system. For the given combination of inputs
and outputs, an overall correlation of 99.927 % has been reached
with including only half of the poles of the full system (28 of 62).
At this point, another advantage of the presented method re-
veals. Only those modes are automatically taken into account
which are observable and/or excitable at the interface positions,
9 Copyright © 2013 by Alstom Technology
usually the bearing locations. The use of Hankel singular val-
ues offers a promising way of reducing even very big systems
efficiently without noticeable loss of quality.
Conclusion
The paper summarises different approaches to include the
dynamic characteristics of structures like gas turbine casings and
support elements in rotordynamic analysis. The lumped mass
(SDOF) approach has still its value for early stage analysis or for
systems not showing significant rotor-structure interaction (e.g.
pedestal supported generator rotors). It is clear that the more so-
phisticated MDOF approaches based on FRFs or the state space
deliver a more realistic result over the interesting speed ranges.
With the availability of accurate finite element models, the re-
quired input data for the computation of the FRFs as well as for
the modal state space representation are at hand. By the repre-
sentation of the support structure in the state space using modal
parameters, an efficient, reliable, user-independent way of con-
sidering complex support dynamics has been implemented. The
approach inherently ensures physically meaningful results only
and can be used consistently for all calculations needed for a ro-
tordynamic assessment: eigenvalue analysis, unbalance as well
as transient response. This approach combines the advantages
of high accuracy with comparably low computing time, which is
a big advantage for rotordynamic analysis, where a lot of load
cases need to be assessed. With the use of the modal parame-
ters for the representation of support structures, the interfaces
to other disciplines like civil engineering (e.g. for foundation)
and structure dynamics (GT casing) can be easily standardised
and the data exchange is independent from the calculation tools
used. This is an additional benefit as every discipline usually
uses special tools for their part of the assessment. With the intro-
duction of a new GT generation, it has been experienced that the
presented methods are very well suited for close to reality pre-
dictions of the rotordynamic behaviour, while ensuring a stan-
dardised and simplified data to manage the interactions between
the different engineering disciplines.

REFERENCES
[1] U GEL , D.; C LARK , A.; K NOPF, E.: The Validation of FE
Simulated Structural Dynamic Behaviour of Heavy Duty
Gas Turbines. – ASME Turbo Expo 2013, Conf.Proc.,
GT2013-94873.
[2] L EVI , E.C.: Complex-Curve Fitting. – IRE Trans. on Au-
tomatic Control, Vol.AC-4 (1959), pp.37-44.
[3] D ENNIS , J.E., J R .; S CHNABEL , R.B.: Numerical Meth-
ods for Unconstrained Optimization and Nonlinear Equa-
tions. – Englewood Cliffs, NJ: Prentice-Hall, 1983.
[4] R ICHARDSON , M.H.; F ORMENTI , D.L.: Parameter Es-
timation From Frequency Response Measurements Using
Rational Fractional Polynomials. – 1st IMAC Conference,
Orlando, FL, November 1982.
[5] S CH ÖNHOFF , U.; E ISENTR ÄGER , P.; N ORDMANN , R.:
Reduction of finite element models of flexible structures for
controller design and integrated modelling. – Invited Pa-
per to the International Conference on Noise and Vibration
Engineering, ISMA25. Leuven, Belgium: 2000.
[6] G AWRONSKI , W.: Dynamics and Control of Structures – A
Modal Approach. – Springer-Verlag. New-York: 1998.
[7] C LARK , A.S.; J URJEVIC , Z.: Fast Simulation of Dy-
namic Behaviour of Heavy Duty Gas Turbines for Qual-
ity Improvement and Reduced Design Cycle Time. – ASME
Turbo Expo 2007, Conf.Proc., pp. 367-377.

10 Copyright © 2013 by Alstom Technology