You are on page 1of 6

Reconciliation of Rotordynamic

E. H. Maslen
e-mail: ehm7s@virginia.edu Models With Experimental Data
J. A. Vázquez A computationally efficient strategy is presented for adjusting analytic rotordynamic mod-
e-mail: jose@virginia.edu
els to make them consistent with experimental data. The approach permits use of conven-
tional rotordynamic models derived using finite element methods in conjunction with
Department of Mechanical and Aerospace
conventional plant identification models derived from impact or sine sweep testing in a
Engineering,
transfer function or influence coefficient format. The underlying assumption is that the
University of Virginia,
predominant uncertainties in engineered models occur at discrete points as effects like
Charlottesville, VA 22903
shrink fits, seal coefficients or foundation interactions. Further, it is assumed that these

Downloaded from http://asmedigitalcollection.asme.org/gasturbinespower/article-pdf/124/2/351/6839736/351_1.pdf by Zhejiang University user on 27 December 2022


unmodeled or poorly modeled effects are essentially linear (at least within the testing and
expected operating domains). Matching is accomplished by deriving a dynamic model for
C. K. Sortore these uncertain effects such that the resulting composite model has a transfer function
Synchrony, Inc.,
which matches that obtained experimentally. The derived augmentations are computation-
7777 Bent Mountain Road,
ally compatible with the original rotor model and valid for stability or forced response
Roanoke, VA 24018
predictions. Further, computation of this augmentation is accomplished using well devel-
e-mail: Chris.Sorte@Synchrony.com
oped and widely disseminated tools for modern control. Background theory and a com-
plete recipe for the solution are supported by a number of examples.
关DOI: 10.1115/1.1416692兴

Introduction perimental data to modify 共reconcile兲 the model at those loca-


tions where error is entering the model because of unidentified
Rotating machine analyses are based on engineering models.
effects or because the modeling techniques cannot provide enough
These models are assembled from mathematical models of various
accuracy 共关1兴兲.
components and physical effects such as the rotor, bearings, seals,
The literature in the fields of component modeling in rotating
bearing support structure, rotor couplings, aerodynamic cross-
coupling, fluid-rotor interaction, etc. The accuracy of the model of machinery and system identification is very extensive. Lund 关2兴
the machine, therefore, depends on accuracy of the models of the presented a method to calculate unbalance response and stability
individual components and of the models of the physical effects analysis of a rotor on flexible supports. This method is now
acting on the rotor. Unfortunately, not all the components and known as the transfer matrix method. Nelson and McVaugh 关3兴
physical effects acting on a rotor can be modeled accurately or developed a finite element formulation for rotor-bearing systems.
identified precisely. The accuracy of a model is determined by Vance et al. 关4兴 present the idea of using experimental modal data
comparing the response of the model and the response of the real in the refinement of models of the rotors of rotating machinery.
machine to the same excitation. The texts by Vance 关5兴, Ehrich 关6兴, and Childs 关7兴 present some of
To improve the accuracy of the machine models, work has been the most common methods to write the equations of motion of
carried out on improving modeling techniques of components like rotor-bearing systems. The text by Ljung 关8兴 includes a review of
the rotor, bearing, and seals. In the case of rotors, experimental system identification techniques. The texts by Ewins 关9兴 and Maia
methods have been used to improve the quality of the models by et al. 关10兴 include a review of the modal identification techniques
testing the rotor, extracting the modal parameters and then modi- most commonly used.
fying the model of the rotor until the modal parameters of the Direct system identification in rotating machinery has been ap-
model match the experimental data. In the case of components plied to the bearing support structure and casing. The dynamic
where the modeling is difficult or very expensive, techniques have behavior of the support structure is measured in the form of fre-
been developed to extract models from experimental data or to use quency response functions 共FRF兲 measured at the bearing loca-
the experimental data directly in the analysis. tions. This information is later used to include the flexible sup-
The process of extracting a model from experimental data is ports effects into rotordynamic calculations. Barrett et al. 关11兴 and
called system identification. Modal identification is a special case Nicholas et al. 关12兴 used FRF data to produce equivalent single
of this technique. All the current literature regarding system iden- mass support models. The bearing coefficients were modified to
tification presents a means of constructing a model whose outputs include the effects of the supports. Redmond 关13,14兴 used fre-
match the measured response for a known set of inputs. The quency response data of the support structure to produce single
‘‘identified’’ model is typically presented as a transfer function. mass support models including the effects of cross-talk between
This approach generally makes no reference to any a priori model supports. The author showed that substantial errors could arise if
of the system. the experimental data was taken with the rotor installed in the
This paper presents a method that bridges the gap between the machine and proposed a method to subtract the rotor influence
well-developed modeling techniques in rotating machinery and from the experimental data. Rouch et al. 关15兴 and Stephenson
the powerful identification techniques available. The end result is et al. 关16兴 used FRF to create equivalent finite elements models of
that the engineer can use the modeling techniques for those the support structure. Vázquez and Barrett 关17兴 showed the effects
components and effects that are well known and then use ex- of flexible supports and casings of rotating machinery can be in-
cluded in the rotordynamic analyses with polynomial transfer
Contributed by the International Gas Turbine Institute 共IGTI兲 of THE AMERICAN functions calculated from experimental data. Vázquez 关18兴 and
SOCIETY OF MECHANICAL ENGINEERS for publication in the ASME JOURNAL OF Vázquez et al. 关19,20兴 used this technique to include the support
ENGINEERING FOR GAS TURBINES AND POWER. Paper presented at the Interna-
tional Gas Turbine and Aeroengine Congress and Exhibition, Munich, Germany, May
structure of an experimental flexible rotor supported by fluid film
8 –11, 2000; Paper 00-GT-395. Manuscript received by IGTI, Nov. 1999; final revi- bearings on flexible bearing supports. The polynomial transfer
sion received by ASME Headquarters, Feb. 2000. Associate Editor: D. Wisler. functions were extracted from FRF of the support structure mea-

Journal of Engineering for Gas Turbines and Power APRIL 2002, Vol. 124 Õ 351
Copyright © 2002 by ASME
sured at the bearing locations using an identification technique equations of motion is assumed by the modern control tools that
introduced by Sanathanan and Koerner 关21兴 and later applied to will be used in the reconciliation process. The transformation of
magnetic bearings by Gähler and Herzog 关22兴. the equations of motion of a rotor-bearing system from the stan-
The work on the bearing support structures has shown that the dard M, C, K representation into the state-space representation is
bearing support structure can be represented as transfer functions straightforward and it is presented by Maslen and Bielk 关25兴.
acting at the bearings of rotating machinery. This is an important The state-space equations for the nominal rotor model with
conclusion. In the field of modern control, these transfer functions the addition of some uncharacterized disturbance forces can be
are represented as local feedback mechanisms acting on the plant expressed as
共rotor-bearing model兲. The importance of this representation is
that there are formal mathematical tools that permit the synthesis ż⫽Az⫹Bc u c ⫹Bk f k (1)
of transfer functions of local feedback mechanisms from experi- y m ⫽Cm z (2)
mental data. Therefore, we can use the synthesis tools used in
modern controls to identify effects acting in rotor models without y k ⫽Ck z. (3)
having to measure these effects independently.
Parallel work in the area of modern controls includes Wolodkin The vector f k represents the suspected uncharacterized distur-

Downloaded from http://asmedigitalcollection.asme.org/gasturbinespower/article-pdf/124/2/351/6839736/351_1.pdf by Zhejiang University user on 27 December 2022


et al. 关23兴 and Fisher et al. 关24兴. Wolodkin et al. showed that local bance forces 共i.e., uncertainties兲 and B k is a model of how they
feedback can be used successfully for the identification of the enter the model 共B k is a matrix with the location of the uncertain-
parameters 共elements兲 of a constant matrix. Fisher et al. 关24兴 used ties兲. The vector y k is the response at the disturbance locations. In
local feedback for the refinement of the model obtained during the the present method, the response at these locations and the distur-
identification of high-order lightly damped systems. The authors bances are related through 共presently unknown兲 feedback as
obtained a transfer function from frequency response data and f k ⫽K共 s 兲 y k . (4)
then used local feedback to refine the model with time-domain
data as it became available. We use a similar idea on this work but The model reconciliation strategy is to compute a dynamic sys-
we start with a mathematical model developed using finite ele- tem, K(s), which minimizes the error between the actual 共identi-
ments. Also, the corrections to the mathematical model are evalu- fied兲 response and the corresponding response computed by the
ated at discrete points. In this manner, it is possible to keep the nominal model at the same locations. Figure 2 illustrates the con-
mathematical model in physical coordinates and only modify the struction of the identification problem.
parts of the model that are expected to contribute error to the The state-space model for this system is

再 冎 冋 册再 冎 冋 册 冋 册
model.
ẑ˙ Â 0 ẑ B̂ 0
This work presents a method that will allow the automation of ⫽ ⫹ c u c⫹ f (5)
model reconciliation by making use of existing mathematical ż 0 A z Bc Bk k

再冎
models. The system is proposed as a nominal model with uncer-
tainties at known locations. The idea is to identify the transfer ẑ
e ym ⫽ 关 Ĉm ⫺Cm 兴 ⫹␧D12 f k (6)
functions of the mechanisms acting at the uncertainty locations z

再冎
such that the response of the mathematical model matches the
experimental data. ẑ
y k ⫽ 关 0 Ck 兴 ⫹␧D21u c (7)
z
Model Reconciliation f k ⫽K共 s 兲 y k . (8)
The reconciliation method starts with a nominal engineering
model of a rotating machine and experimental data measured at The constant matrices Â, B̂c , and Ĉm represent the state-space
some known locations. The purpose of the method is to modify realization of the identified rotor. This state-space realization is
the engineering model such that the output of the model matches calculated from the experimental FRF 共Gähler and Herzog 关22兴兲.
the experimental data. The assumption is that the error in the D12 and D21 are full-rank matrices and ␧ is a very small number.
engineering model occurs at discrete points and/or components. These values do not affect the results but are needed to avoid
These components could be aerodynamic effects, seal effects, or numerical problems.
interference fits between disks and journal—these are features or Any suitable method for designing the feedback controller may
conditions which may not have an accurate accounting within the be used: H⬁ , H2 , etc. 共关26兴兲. The identification process is a mat-
model and are usually difficult to identify precisely. These effects ter of identifying parameters 共sources of model error兲, construct-
are represented by a local feedback mechanism as shown in Fig. ing the error supermodel 共Fig. 2兲 and then turning the ‘‘crank’’ of
1. The ‘‘controller’’ K(s) is the correction included in the nominal a chosen controller design algorithm. If the controller is designed
engineering model. It can be interpreted as complex frequency- properly, the reconciled model will be matched to the measured
dependent stiffness or transfer functions acting at discrete points
in the model 共these points are selected by the engineer兲.

State-Space Model
The equations of motion of the engineering model will be writ-
ten using a state-space representation. This representation of the

Fig. 1 Reconciled model representation Fig. 2 Model identification block diagram

352 Õ Vol. 124, APRIL 2002 Transactions of the ASME


data in the sense that, when the original test signals are applied to tation is imposed to simplify the examples. The method, however,
this modified model, its output will match the original data. can be used directly for coupling between the vertical and hori-
If the uncertainty locations are chosen appropriately, then it is zontal direction.
expected that the internal states of the system will also be matched
to the measured plant, not just the output measurements. This Example 1. It is posited that the true rotor contains some
method does not ignore 共as other currently available identification dynamics at the thrust bearing end of the shaft, say due to some
methods do兲 but takes advantage of existing well-developed mod- interaction with a coupling at the end of the shaft. In this example
els of the rotor, bearings, seals, etc. If the particular choice of a stiffness of 900,000 N/m is found in the ‘‘identified’’ model at
uncertainty locations was chosen poorly 共e.g., assuming the wrong this location. In the nominal model, only the location of the un-
location of a disturbance effect兲, the designed controller, K(s), known spring is assumed to be known. This is the only assump-
may still force a match between the response of the modeled and tion made in the model reconciliation process. The value of ␧ in
measured plants. In this case, however, it is less likely that the Eqs. 共6兲 and 共7兲 is selected to be small enough such that the error
subsequent use of this reconciled model to construct transfer func- in the transfer function of the dynamic system does not change
tions to and from other points 共points other than that which were with the value of ␧ and big enough to allow good convergence of
used to perform the measurement兲 will be accurate. the H ⬁ crank. The actual value of ␧ depends on the dynamics of

Downloaded from http://asmedigitalcollection.asme.org/gasturbinespower/article-pdf/124/2/351/6839736/351_1.pdf by Zhejiang University user on 27 December 2022


the rotor being analyzed. For the rotor in these examples, values
of ␧ between 10⫺9 and 10⫺10 are appropriate.
Steps in the Reconciliation Process Figures 4 – 6 show the results of this process. The maximum
singular value of each system is used for comparison. The maxi-
1 Measure FRFs of the actual rotor-bearing system.
mum singular value is a composite measure of the component
2 Identify a dynamic system from FRF 共state-space model for
transfer functions in a matrix and provides useful and direct com-
the transfer functions兲.
parison of the overall model of the system 共关27兴兲. Figure 4 shows
3 Construct a nominal engineering model 共rotor, bearing, seals,
etc.兲.
4 Identify the points of likely modeling uncertainty 共interfer-
ence fits, seal locations, aerodynamic cross-coupling, etc.兲.
5 Assemble the error supermodel 共consists of engineering
model coupled to the experimentally identified model, Fig.
2兲.
6 Turn H ⬁ crank⇒produces K(s).
K(s) is a representation of the physical effects operating at the
points of uncertainty. It will be an n⫻n matrix where n is the
number of uncertainties. The elements of this matrix are transfer
functions of order equal to the sums of the orders of the identified
plant and the mathematical model.

Examples
Consider a long, flexible three-bearing rotor supported by
magnetic bearings. The geometry of this rotor is illustrated in
Fig. 3.
The bearing layout of this rotor is intended to be representative
of a large boiler feed pump. The rotor is approximately 2.46 m
long and weights 1036 N. The expected continuous running
speed of the pump is 5200 rpm. The stiffness and damping Fig. 4 Maximum singular value comparison between nominal
coefficients for all three bearings are 8.75⫻106 N/m and 17,500 model and the identified rotor transfer functions from measur-
N s/m, respectively. able inputs „bearings… to measurable outputs „sensors…,
The initial measurement of FRF and subsequent model identi- Example 1
fication is not actually carried out here because we are using a
numerical model. Instead, for the purposes of this example, the
‘‘identified model’’ is the same as the ‘‘nominal model’’ but with
the actual values of the unknown effects included.
The model of the rotor is assembled from undamped mode
shapes. Two rigid-body mode shapes and four flexible mode
shapes are used. For the purposes of this work the model is con-
sidered planar and no gyroscopic effects are included. This limi-

Fig. 5 Comparison of the error of the nominal model and the


Fig. 3 Three-bearing rotor reconciled model, Example 1

Journal of Engineering for Gas Turbines and Power APRIL 2002, Vol. 124 Õ 353
Downloaded from http://asmedigitalcollection.asme.org/gasturbinespower/article-pdf/124/2/351/6839736/351_1.pdf by Zhejiang University user on 27 December 2022
Fig. 6 Matching controller transfer function gain, Example 1 Fig. 8 Comparison of the error of the nominal model and the
reconciled model, Example 2

a noticeable difference between the identified transfer function


and nominal model transfer function at frequencies in the vicinity
of the operating speed.
Figure 5 shows the error between the identified rotor and
the nominal and reconciled rotor models. At the running speed,
the difference between the identified rotor model and the nomi-
nal model is 12.35 percent. At the same speed, the difference
between the identified rotor and the reconciled model is 0.00083
percent. The reconciliation method decreases the error by five
orders of magnitude. As shown in Fig. 6, the H ⬁ controller syn-
thesis exactly identifies the unknown spring with a stiffness of
900,000 N/m.
Example 2. As a second example, the identified rotor now
includes the effects of the coupling described in Example 1 and
the effects of a seal located at midspan. The seal has a stiffness
coefficient of 350,000 N/m and damping coefficient of 4,300 N
s/m. For the reconciliation process, the location of the coupling
effect is identified but the seal is ignored. Figure 7 shows the
comparison between maximum singular value of the identified
rotor system and the nominal model. There is a noticeable differ-
ence in the vicinity of the operating speed of the rotor. This is
expected because the model does not include the effects of the Fig. 9 Matching controller transfer function gain, Example 2
seal and the stiffness at the coupling end. Figure 8 shows the error

Fig. 7 Maximum singular value comparison between nominal


model and the identified rotor transfer functions from measur-
able inputs „bearings… to measurable outputs „sensors…, Fig. 10 Comparison of the error of the nominal model and the
Example 2 reconciled model, Example 3

354 Õ Vol. 124, APRIL 2002 Transactions of the ASME


Table 1 Comparison between the eigenvalues of the nominal
model, identified rotor-bearing system, and the reconciled
model

Nominal Model Identified System Reconciled Model


Real Exp Frequency Real Exp Frequency Real Exp Frequency
共1/s兲 共rad/s兲 共1/s兲 共rad/s兲 共1/s兲 共rad/s兲
⫺1255.6 142.73
⫺122.24 184.13
⫺1870.8 351.84
⫺105.40 407.23 ⫺131.12 424.51 ⫺131.12 424.51
⫺284.03 479.69 ⫺302.85 497.53 ⫺302.85 497.53
⫺80.116 549.56 ⫺474.39 575.84 ⫺474.39 575.84
⫺473.07 590.60 ⫺117.07 670.79 ⫺117.08 670.79
⫺311.28 1073.1 ⫺301.82 1050.3 ⫺301.82 1050.3
⫺161.96 1808.7 ⫺173.79 1842.9 ⫺173.80 1842.9

Downloaded from http://asmedigitalcollection.asme.org/gasturbinespower/article-pdf/124/2/351/6839736/351_1.pdf by Zhejiang University user on 27 December 2022


⫺55804. 2827.6
⫺1384.8 23698.
⫺220.77 102950
⫺326.41
Fig. 11 Matching controller transfer function gain, Example 3 ⫺348.87
⫺421.84
⫺1140.0
⫺1822.1
⫺2929.3
⫺40650.
⫺60953.

standard H ⬁ or H 2 cranks to compute K(s). The cross-coupling is


an inherent result of the synthesis method and cannot be avoided
at present. However, forcing the synthesis of uncoupled transfer
functions is an active area of research in the controls community.
Once such methods become available, they can be used directly in
this reconciliation technique.
Figure 12 shows the percentage error between the transfer
functions of the coupling stiffness and seal effects and the trans-
fer functions calculated by the H ⬁ synthesis. The error of the
coupling stiffness is below 0.004 percent up to 130 Hz. The
maximum error in the range of interest is 0.28 percent at 811 Hz.
The error of the seal transfer function is below 0.55 percent up to
470 Hz.
Fig. 12 Percentage error of matching controller transfer func- Table 1 compares the eigenvalues of the nominal model, the
tion gain, Example 2 identified rotor-bearing system and the reconciled model. The rec-
onciliation process improves the eigenvalues of the system. The
reconciled model includes all the eigenvalues of the true rotor-
of the nominal model and the error of the reconciled model. The bearing system plus some eigenvalues that are artifacts of the
reconciliation process 共which is unaware of the seal兲 does not reconciliation process. The additional eigenvalues are all well
usefully improve the overall error of the model. Figure 9 shows damped and do not significantly affect the stability analyses.
the transfer function of the matched dynamic system. In this case,
the synthesis method is trying to match the effect of the coupling
and the seal with a transfer function at the coupling location. This Conclusions
example shows both a limitation and strength of this method. The
limitation is that the sources of uncertainty must be known in A method is presented to reconcile engineering models of ro-
order for the method to improve the model. The strength is that, if tating machinery with experimental data. The underlying assump-
the sources of uncertainty are poorly chosen, then the resulting tion is that the difference between the engineering models and the
poor match suggests that other sources should be included. experimental data is caused by uncharacterized disturbances 共un-
certainties兲 acting at discrete points in the model. The forces
Example 3. As a last example, the true model includes the caused by the uncertainties are represented as local feedback
effects of the coupling and the seal as introduced in Example 2. mechanisms. The method makes use of the existing engineering
For the reconciliation process, however, the locations of the un- model of the system and likely location of uncertainties where
certainties 共i.e., the location of the coupling and seal兲 are both error enters the model. Transfer functions acting at the location of
known. Figure 10 shows the error in the maximum singular value uncertainties are calculated so that the response of the reconciled
of the nominal model and the reconciled model. The error in this engineering model matches the experimental data.
case is also reduced for the reconciled model, showing a large This method has several advantages over other parameter iden-
improvement. The error is now four orders of magnitude smaller tification schemes. One is that it effectively shrinks the number of
than the error of the nominal model. Figure 11 shows the transfer identification variables 共uncertainties兲 to a reasonable size and al-
function of the matching dynamic system. The transfer functions lows the engineer direct control over where the model adjustments
of the coupling effects and the seal are clearly identified. are made. Another advantage is that the uncertainties have a direct
The synthesis method also produces cross-coupling between the analogue to real, physical effects—interference fits, aerodynamic
two uncertainty locations, although their magnitudes are much behavior, etc. The identification of the location of the uncertainties
smaller than the direct, physically significant transfer functions. is a critical step. Given the current state of the art in rotordynam-
Generation of undesired cross-couplings is a drawback of using ics modeling, it is expected that the experienced rotordynamicist

Journal of Engineering for Gas Turbines and Power APRIL 2002, Vol. 124 Õ 355
should be capable of sensibly identifying the likely sources of Modal Analysis, Research Studies Press Ltd., Tauton, Somerset, UK.
关11兴 Barrett, L. E., Nicholas, J. C., and Dhar, D., 1986, ‘‘The Dynamic Analysis of
uncharacterized disturbances acting upon the system.
Rotor-Bearing Systems Using Experimental Bearing Support Compliance
An interesting benefit to this method is that it can actually pro- Data,’’ Proceedings of the Fourth International Modal Analysis Conference,
duce a scaled measure of the disturbance effect. Numerical ex- Union College, Schenectady, NY, pp. 1531–1535.
amples are presented where coupling stiffness and seal effects are 关12兴 Nicholas, J. C., Whalen, J. K., and Franklin, S. D., 1986, ‘‘Improving Critical
identified from frequency response data of the rotor-bearing sys- Speed Calculations Using Flexible Bearing Support FRF Compliance Data,’’
Proceedings of the 15th Turbomachinery Symposium, Texas A&M University,
tem. The only assumption used in the process is the location of the College Station, TX.
uncertainties in the model. The reconciliation process reduces the 关13兴 Redmond, I., 1995, ‘‘Practical Rotordynamics Modeling Using Combined
error between the engineering model and the experimental data by Measured and Theoretical Data,’’ Proceedings of the 13th International Modal
four orders of magnitude. The eigenvalues 共stability analysis兲 of Analysis Conference, Nashville, TN, Society for Experimental Mechanics,
the reconciled model match the eigenvalues of the actual rotor- Bethel, CT.
关14兴 Redmond, I., 1996, ‘‘Rotordynamic Modelling Utilizing Dynamic Support
bearing system. The reconciliation process adds some eigenvalues Data Obtained From Field Impact Tests,’’ Proceedings of Sixth International
as the product of the added transfer function. However, these Conference on Vibrations in Rotating Machinery, Paper C500/055/96.
added eigenvalues are all well damped. 关15兴 Rouch, K. E., McMains, T. H., and Stephenson, R. W., 1989, ‘‘Modeling of
If the locations of all uncertainties acting in the system are

Downloaded from http://asmedigitalcollection.asme.org/gasturbinespower/article-pdf/124/2/351/6839736/351_1.pdf by Zhejiang University user on 27 December 2022


Rotor-Foundation Systems Using Frequency-Response Functions in a Finite
not included in the reconciliation process, the method will try to Element Approach,’’ Proceedings of the 1989 ASME Design Technical Con-
ference 12th Biennial Conference on Mechanical Vibration and Noise, Mont-
match the response of the nominal model to the experimental real, Canada, ASME, New York, pp. 157–166.
data using only the identified locations of uncertainties. In the 关16兴 Stephenson, R. W., and Rouch, K. E., 1992, ‘‘Generating Matrices of the
example presented, the reconciliation process does not usefully Foundation Structure of a Rotor System From Test Data,’’ J. Sound Vib., 154,
improve the error between the reconciled model and the experi- No. 3, pp. 467– 484.
mental data. This indicates that other sources of uncertainty need 关17兴 Vázquez, J. A., and Barrett, L. E., 1999, ‘‘Transfer Function Representation of
Flexible Supports and Casings of Rotating Machinery,’’ Proceedings of the
to be identified. 17th International Modal Analysis Conference, Orlando, FL, Feb., Society for
The reconciliation process produces a measure of the distur- Experimental Mechanics, Bethel, CT.
bance effects acting on the engineering model plus some undes- 关18兴 Vázquez, J. A., 1999, ‘‘Using Transfer Functions to Model Flexible Supports
ired cross-coupling effects. The cross-coupling is an inherent re- and Casings of Rotating Machinery,’’ Ph.D. dissertation, University of Vir-
ginia, Charlottesville, VA, Jan.
sult of the synthesis method and cannot be avoided at present. 关19兴 Vázquez, J. A., Barrett, L. E., and Flack, R. D., 1999, ‘‘A Flexible Rotor on
However, forcing the synthesis of uncoupled transfer functions is Flexible Bearing Supports. Part I: Stability,’’ Proceedings of the 1999 Vibra-
an active area of research in the controls community. Once such tion Conference, Paper DETC99/VIB-8285. Las Vegas, NV, Sept. 12–15,
methods become available, they can be used directly in this rec- ASME, New York.
onciliation technique. 关20兴 Vázquez, J. A., Barrett, L. E., and Flack, R. D., 1999, ‘‘A Flexible Rotor on
Flexible Bearing Supports. Part II: Unbalance Response,’’ Proceedings of the
1999 Vibration Conference, Paper DETC99/VIB-8286. Las Vegas, NV, Sept.
References 12–15, ASME, New York.
关21兴 Sanathanan, C. K., and Koerner, J., 1963, ‘‘Transfer Function Synthesis as a
关1兴 Sortore, C. K., 1999, ‘‘Observer Based Critical Response Estimation in Rotat-
Ratio of Two Complex Polynomials,’’ IEEE Trans. Autom. Control, 8, pp.
ing Machinery,’’ Ph.D. dissertation, University of Virginia, Jan.
56 –58.
关2兴 Lund, J. W., 1974, ‘‘Stability and Damped Critical Speeds of a Flexible Rotor
in Fluid-Film Bearings,’’ J. Eng. Ind., 96, pp. 509–517. 关22兴 Gähler, C., and Herzog, R., 1994, ‘‘Identification of Magnetic Bearing Sys-
关3兴 Nelson, H. D., and McVaugh, J. M., 1976, ‘‘The Dynamics of Rotor-Bearing tems,’’ Proceedings of the Fourth International Symposium on Magnetic Bear-
Systems Using Finite Elements,’’ J. Eng. Ind., 98, pp. 593– 600. ings, ETH Zurich, Aug., University Press, Swiss Federal Technical University,
关4兴 Vance, J. M., Murphy, B. T., and Tripp, H. A., 1987, ‘‘Critical Speeds of Zurich, Switzerland, pp. 293–298.
Turbomachinery: Computer Predictions vs. Experimental Measurements—Part 关23兴 Wolodkin, G., Rangan, S., and Poolla, K., 1997, ‘‘LFT Approach to Parameter
I: The Rotor Mass–Elastic Model,’’ ASME J. Vib., Acoust., Stress, Reliab. Estimation,’’ Proceedings of the American Control Conference, Vol. 3, June
Des., 109, pp. 1–7. 4 – 6, IEEE, Piscataway, NJ, pp. 2088 –2092.
关5兴 Vance, J. M., 1988, Rotordynamics of Turbomachinery, John Wiley and Sons, 关24兴 Fisher, D., Jue, D., Packard, A., and Poolla, K., 1999, ‘‘On the Identification of
New York. High-Order Lightly-Damped Multivariable Systems,’’ Proceedings of the 1999
关6兴 Ehrich, F. F., 1992, Handbook of Rotordynamics, McGraw-Hill, New York. American Control Conference, San Diego, CA, June 2– 4, IEEE, Piscataway,
关7兴 Childs, D., 1993, Turbomachinery Rotordynamics, John Wiley and Sons, New NJ, pp. 848 – 853.
York. 关25兴 Maslen, E. H., and Bielk, J. R., 1992, ‘‘A Stability Model for Flexible Rotors
关8兴 Ljung, L., 1999, System Identification, Theory for the user, 2nd Ed., Prentice- With Magnetic Bearings,’’ ASME J. Dyn. Syst., Meas., Control, 114, pp. 172–
Hall, Englewood Cliffs, NJ. 175.
关9兴 Ewins, D. J., 1984, Modal Testing: Theory and Practice, Research Studies 关26兴 Zhou, K., Doyle, J. C., and Glover, K., 1996, Robust and Optimal Control,
Press Ltd., Tauton, Somerset, UK. Prentice Hall, Englewood Cliffs, NJ.
关10兴 Maia, N. M. M., Silva, J. M. M., He, J., Lieven, N. A. J., Lin, R. M., Skingle, 关27兴 Maslen, E. H., 1991, ‘‘Magnetic Bearing Synthesis for Rotating Machinery,’’
G. W., To, W. M., and Urgueria, A. P. V., 1997, Theoretical and Experimental Ph.D. dissertation, University of Virginia, Jan.

356 Õ Vol. 124, APRIL 2002 Transactions of the ASME

You might also like