You are on page 1of 7

274 IEEE TRANSACTIONS ON MAGNETICS, VOL. 41, NO.

1, JANUARY 2005

Rotordynamic Modeling Using Bond Graphs:


Modeling the Jeffcott Rotor
Joaquín Campos, Mark Crawford, and Raul Longoria

Abstract—Visualizing the complex and counterintuitive phe-


nomenology associated with rotordynamics has been problematic
since Rankine first scientifically studied it in the early 1900s. To
address this, a simple model capable of allowing visualization of
the parameters and problematic aspects associated with an imbal-
anced rotor (Jeffcott rotor) will be proposed using bond graphs,
which provide a structured and unified method for modeling a
large class of nonlinear, multienergetic systems. The purpose of
this paper is to provide a high-fidelity, experimentally proven bond
graph model of the Jeffcott rotor to aid in the design and analysis
of high-speed rotational machinery often associated with pulsesd
energy systems. The intent of the model is the ability to use it as
a modular and foundational piece in more complex rotordynamic Fig. 1. Schematic of Jeffcott rotor with flexible shaft.
models.
Index Terms—Bond graph, Jeffcot rotor, pulsed power, rotordy- lematic since Rankine first scientifically studied it in the early
namic models, rotordynamics.
1900s [1]. In order to address this, a simple model capable
of allowing visualization of the parameters and problematic
I. INTRODUCTION aspects associated with a simple imbalanced rotor system
(Jeffcott rotor) will be proposed using a formulation technique
T HE complex interaction between rotordynamics and
electromagnetics has been of considerable interest as
high-speed rotating electrical machines have become the log-
known as bond graphs. This technique inherently allows the
modeler to “see” causality in physical systems, which for
ical choice for mobile pulsed power applications. Due to the the current application greatly aids in understanding system
inherent size and operating speeds of the proposed machines, phenomena. Specifically, the method described in this paper
great care must be put into the design and analysis of both the shows how a Lagrangian formulation can be readily integrated
mechanical and electrical aspects of the machine. In this paper, with the bond graph approach. The analytical formulation
a small portion of the overall analysis problem—the basic followed in a Lagrangian approach is especially useful in this
rotordynamics modeling—will be addressed. The intent is to application where the modular Jeffcott (Lagrangian) model
produce an analytical model that is simple, robust, and can be can be integrated with a bond graph model representing an
used as a modular and foundational piece to the more complex electromechanical drive train.
overall system models.
B. Physical System and Kinematics
II. PROBLEM STATEMENT The system used as a case study is the classical Jeffcott Rotor
A. Dilemma and Approach developed by Jeffcott et al. to study the whirl phenomena of
imbalanced rotors [1]–[3]. The Jeffcott rotor consists of a rigidly
Visualizing the complex and counterintuitive phe-
supported isotropic shaft (free to rotate along its bearing axis)
nomenology associated with rotordynamics has been prob-
with finite elasticity, on which a geometrically symmetrical but
inertially asymmetrical rotor is rigidly fastened (see Fig. 1).
Manuscript received December 19, 2003. The research reported in this work Using a Cartesian coordinate system with an Eulerian angle
was performed in connection with Contract DAAD17-01-D-0001 with the U.S.
Army Research Laboratory. The views and conclusions contained in this doc- description, the following diagram is constructed (Fig. 2),
ument are those of the authors and should not be interpreted as presenting the where the position vector and spin angle initially define the
official policies or position, either expressed or implied, of the U.S. Army Re- system’s center of mass with respect to the inertial reference
search Laboratory or the U.S. Government unless so designated by other au-
thorized documents. Citation of manufacturers’ or trade names does not consti- frame.
tute an official endorsement or approval of the use thereof. The U.S. Govern- Breaking the position vector down into its components yields
ment is authorized to reproduce and distribute reprints for Government purposes
notwithstanding any copyright notation hereon.
J. Campos and M. Crawford are with the Institute for Advanced Technology, (1)
The University of Texas at Austin, Austin, TX 78759-5316 USA (e-mail:
joaquin_campos@iat.utexas.edu; mark_crawford@iat.utexas.edu). Differentiating the position vector with respect to time yields
R. Longoria is with the Department of Mechanical Engineering, The
University of Texas at Austin, Austin, TX 78712-0292 USA (e-mail:
the velocity vector
r.longoria@mail.utexas.edu).
Digital Object Identifier 10.1109/TMAG.2004.838924 (2)
0018-9464/$20.00 © 2005 IEEE
CAMPOS et al.: ROTORDYNAMIC MODELING USING BOND GRAPHS 275

Fig. 2. End view of Jeffcot rotor with flexible shaft.

Fig. 4. Orthogonal spring and dashpot physical model of Jeffcot rotor.

the rotor. These vibrations usually occur at multiples of the nat-


ural frequency of the shaft-rotor system and are typically at su-
percritical speed (speeds that are usually twice the natural fre-
Fig. 3. Free body diagram of the shaft-rotor system. quency) [6], [7]. This phenomenon and its subphenomena are
investigated analytically, and in a limited form, experimentally.
but since the magnitude of the eccentricity does not change Due to the high speeds and the size of the rotor-shaft system
with time the equation becomes that was constructed, experimental testing is kept under super-
critical speed. Thus, the test setup reported in this paper will not
(3) experience any of the phenomena associated with asynchronous
Differentiating the position vector once more yields the ac- whirl. The analytical results will also be compared to those in
celeration vector literature to measure the level of fidelity of the proposed model.

IV. PHYSICAL MODEL


(4) The first, most common, and simplest method for the phys-
ical modeling of this system is to construct it by supporting the
for the center of mass of the Jeffcott rotor. Now that the accel-
imbalanced flywheel at its geometric center by two orthogonal
eration vector has been attained, the free body diagram of the
springs [2], [4] with the same stiffness , each in
system can be constructed (Fig. 3).
parallel with a dashpot, as in Fig. 4 [4].
In this figure, is the deflection of the rotor shaft, is the
There are two coordinate sets for this method, one that is
shaft input torque, is the stiffness of the rotor shaft, is the
fixed to the flywheel (rotating with it), and the second that is
rotational damping, and is the internal damping of the shaft.
the inertial frame. If an initial point of view is taken from the in-
ertial frame, the flywheel, when rotating, imparts a sinusoidal
III. PHENOMENOLOGY forced vibration on the two orthogonal springs. This method
The principal phenomena to address with the model are those best describes the case where the supports are much more com-
associated with whirl. There are two types of “whirl,” the first pliant than the shaft on which the flywheel rotates, which is
being the most common form known as “synchronous whirl,” not the case in this study. The second and less common method
which produces a forcing function frequency that is the same of modeling this system uses an inertial cylindrical coordinate
as the drive or “spin” frequency. The second form of whirl is system as shown in Fig. 5, in which the compliance is a single
“asynchronous whirl.” This phenomenon is not as common, but spring–dashpot system that rotates with the body [2], [8].
is the most dangerous, since vibrations due to it are exceed- The loading on the structural member modeled in this system
ingly violent [4] and are usually caused by system instabilities is constant, as opposed to the above case where the loading
at very high speeds (a term that is relative to the properties of is sinusoidal. This is an important difference that greatly af-
the system, i.e., shaft natural frequency). fects the outcome of the models where system damping is con-
Synchronous whirl is a very well characterized phenomenon cerned. This method is less popular because of the mathemat-
and has been the subject of numerous papers and journal arti- ical complexity introduced by the cylindrical coordinates, but it
cles. As such, there is a great deal of information on the subject has its conceptual advantage as being more realistic to an actual
matter [5]. The primary focus of study for synchronous whirl system, where the supports are much more rigid than the shaft,
is the critical speed, which is the speed at which the amplitude as in the current application. In creating the model for this study,
of vibration increases dramatically, a phenomenon often incor- elements from both simplified models will be used to construct
rectly associated with the natural frequency of the rotor-shaft a more appropriate one. In the current case, it is known that the
system. supporting structure is much more rigid than the shaft, so the
Asynchronous whirl, as its name implies, is a vibrational phe- second model is a more viable candidate for application. In order
nomenon that is not synchronous with the spin frequency of to simplify the analytical portion of this study, the spring force
276 IEEE TRANSACTIONS ON MAGNETICS, VOL. 41, NO. 1, JANUARY 2005

that uses Lagrange’s equation to construct the state equations.


Recall the Lagrangian

(6)

where is the Lagrangian, is the kinetic coenergy, and


is the potential energy. The Hamiltonian formulation starts with
the identification of the kinetic and potential energies and forms
the equations of state from the derivative of these two energy
functions [9] in the form of Lagrange’s equation. Since the Jeff-
Fig. 5. Single spring and dashpot physical model of Jeffcot rotor.
cott rotor only has one inertia—the rotor—the formulation will
be mainly the description of its rate of change of motion. The
general form of the Hamiltonian is
is decomposed into its and components to avoid the need
for cylindrical coordinates. This in effect produces two springs (7)
of equivalent stiffness as in the first model above. It is im-
portant at this point to emphasize that, even though the system
looks like the first model, it is still conceptually equivalent to Using the procedure as shown by Vance [10], the following state
the second model. The reason for this is to emphasize that the equations are derived:
internal damping in the system is present in the shaft only
if the shaft is being cyclically loaded and/or has some .
These effects will be seen primarily during acceleration and de-
celeration of the rotor and during whirl, where instabilities are
caused by internal damping. The internal damping affects the
critical speed and the maximum amplitude reached at critical
speed. The rotational or viscous damping is a form of braking
of the flywheel, decreasing the angular velocity and in turn de-
creasing the amplitude of apparent vibration.

V. FORMULATION
Two formulation techniques will be used to construct the
system’s state equations. This will be done to compare and
contrast a commonly used technique known as Hamilton’s
method to the proposed technique of bond graphing.
From the free-body diagram (Fig. 2), the following equations
of motion are derived: (8)

direction: This formulation is particularly tedious for the problem studied.


Albeit a common method for rotordynamics, it would not be
this author’s preferred method for obtaining the system’s state
direction: equations.

B. Bond Graph Formulation


direction:
The bond graph has an advantage. Besides being a formu-
(5)
lation technique, it acts as a conceptual aid enabling the user
to view the flow of power and track its effects—a tool that is
This set of equations of motion will be the starting point from
very useful for the subject of rotordynamics, where there are so
which the comparative methods of modeling will be con-
many counterintuitive phenomena [4]. There does not appear to
structed. Each equation of motion will be verified using each of
be a great deal of work in open literature on the subject of mod-
the formulations methods proposed. These equations of motion
eling rotordynamics using bond graphs other than the notes by
should be reproduced by the integration of the state equations,
Beaman and Paynter [11], text by Karnopp et al. [12], and one
to be derived by the methods described above. So they will be
reference for explicit bond graph modeling of rotordynamics,
used as checks on the formulation techniques.
from Hubbard [13].
The bond graph technique that will be used in this paper will
A. Hamiltonian Formulation be the Lagrangian bond graph technique as laid out by Beaman
The most common analytical technique used to develop and Paynter [14]. The basic structure of the Lagrangian bond
the system’s state equations in rotordynamics is known as graph is as a multiport capacitive element; although it represents
Hamilton’s method. This technique is an energy formulation storage of both kinetic and potential energies, all the energy is
CAMPOS et al.: ROTORDYNAMIC MODELING USING BOND GRAPHS 277

Fig. 6. Basic Langrangian bond graph of Jeffcot rotor.

Fig. 8. Final bond graph model of Jeffcot rotor.

TABLE I
MATERIAL PROPERTIES FOR THE SHAFT AND THE ROTOR

Fig. 7. Initial condition input bond graph. VI. EXPERIMENTAL SETUP


The experimental setup for this study was constructed as an
representatively stored in the capacitive element ( ). The for- instrumented Jeffcott-like rotor. Although great care was put
mulation is similar to Lagrange’s method, only the equations of into the design and the construction of the test device, it is dif-
motion are expressed in the form of first-order ordinary differen- ficult to produce an ideal Jeffcott rotor, and any test setup will
tial equations (ODEs). The bond graph structure for the system have inconsistencies and imperfections in the shaft-rotor system
under study will have the general structure of the bond graph in as well as the supporting structure. The test setup is a simple
Fig. 6. shaft-rotor system suspended on rigid bearings and whose rotor
The transformation matrix, , in this case, is an identity ma- accommodates eccentric placement of mass in order to precisely
trix, and the inputs to the Lagrangian are the generalized mo- control the position of the center of mass of the rotor, so that
ments , , and . The generalized conservative efforts, , an imbalance can be induced. Table I shows the material prop-
, and are the causal outputs for which the Lagrangian re- erties for the shaft and the rotor. The rotor is constructed of
turns the flows , , and . The nonconservative inputs to the 6061-T6 aluminum, using 19-mm plate stock, and has a diam-
system are , , and . eter of 300 mm. Controlled eccentricity of the center of mass
In order to account for the losses in the system (nonconser- is accomplished by means of three rows of 1/4-20 tapped holes
vative forces) and to account for the transformation of the input spaced 20 mm apart and symmetrically located 180 on either
torque (effort) into three different flows in the three directions— side of the center of the rotor. The tapped holes provide a consis-
, , and spin—the structure shown in Fig. 7 is needed. tent mechanism for introducing an eccentric mass by means of
Putting the entire bond graph together gives Fig. 8. either 1-in-long 1/4-20 steel hex screws having an average mass
Using the same notation as Beaman and Paynter [14], the of 6.03 g each, or 0.6-in-long 1/4-20 steel hex screws, having an
general form of the Lagrange equation is average mass of 4.96 g each. The rotor shaft is 303 stainless steel
with a 0.50-in diameter. The shaft length is 15.96 in, of which
(9) only 12.25 in is used as the supporting length of the rotor.
The rotor was driven by a Kollmorgen servo motor that was
where is the generalized flow variable, is the generalized limited to 5000 rpm. Measurement of the rotor lateral position
velocity, is the Lagrangian, is the kinetic was done via an orthogonal pair of position-sensitive laser diode
coenergy function, and is the potential energy function. The pairs from LMI/Selcom that were mounted on a rigid structure
procedure used for the Lagrangian bond graph generally fol- surrounding the rotor (see Fig. 9). A total of 88 tests were con-
lowed that presented by Beaman and Paynter [14], which gives ducted with two variants of the above setup.
the following state equations:
VII. ANALYSIS
Using the data gathered from experiments, a comparison will
be made between actual and simulated data. Three representa-
tive transient (spin-up) cases will be presented for comparison
with the bond graph simulation data.

A. Formulation Comparison
To gain confidence in the formulation procedures, the two
modeling techniques are compared using an identical set of ini-
tial conditions for a simulated experiment of the experimental
(10) rotor spinning up through its critical speed. As can be noted, the
278 IEEE TRANSACTIONS ON MAGNETICS, VOL. 41, NO. 1, JANUARY 2005

Fig. 9. Photograph of final test setup.


Fig. 11. Plot of radial deflection of shaft versus time of both Hamiltonian and
bond graph model simulations.
TABLE II
SIMULATION INITIAL CONDITIONS FOR MODEL COMPARISON
TABLE III
3600-RPM TEST

Fig. 12. Plot of experimental versus simulation data of x-center position for
3600 rpm test.

physics. This provides verification for both techniques, giving a


check on the respective model formulations.

B. Comparison With Experimental Data


Fig. 10. Plot of center position trace versus time of both Hamiltonian and bond The first case presented, spin-up to 3600 rpm, is the only
graph simulations. spin-up test presented that is conducted at subcritical speed.
Table III shows the conditions for this test, and Fig. 12 plots
state equations resulting from the Hamiltonian derivation and the results.
the Lagrangian bond graph derivation are slightly different from The lighter color is the bond graph simulation data and the
each other. The following simulation comparison should serve darker is the experimental. There are three main differences be-
to verify these derivations. The conditions in Table II were used tween the test and simulation results. The main and most ob-
in the simulation comparison. vious difference is the amplitude of vibration. There seems to be
The following results show the Hamiltonian simulation re- a resonance at about 3500 rpm that does not match the bump-test
sults plotted with the Lagrangian bond graph model results. The resonance for the direction of the experimental rotor, but is
first plot (Fig. 10) is of the trace of the rotor center. The second close to the resonant frequency of the direction, which is
plot (Fig. 11) is of the radial displacement as seen rotating with 3540 rpm. It is possible that the resonant amplitude of vibra-
the rotor. tion is coupling into the direction from the direction, and the
The simulation results are almost identical. Computed values fact that the resonant peaks are occurring roughly at the orthog-
are approximately 0.2% of each other. So, even if the formula- onal direction resonance is evidence. This test is a good example
tion produced state equations that may have seemed different, of how the ideal case and the actual case differ because of the as-
they are in fact similar and produce similar results for the same sumption of a symmetric system. The third difference between
CAMPOS et al.: ROTORDYNAMIC MODELING USING BOND GRAPHS 279

TABLE IV
4200-RPM TEST INITIAL CONDITIONS

Fig. 14. Plot of experimental versus simulation data of x-center position for
4400 rpm test.

TABLE V
4400-RPM TEST INITIAL CONDITIONS

Fig. 13. Plot of experimental versus simulation data of y -center position for
4200 rpm test.

the simulation and the experiment is that the start-up vibration ference between the average measured and adjusted values for
amplitudes are very different. The simulation shows hardly any the damping coefficient. It should be noted that this adjusted
start-up displacement amplitude, whereas the experiment shows value of the damping coefficient was still within the range of
comparatively large displacement amplitude. It is hypothesized measured values—between 0.01 and 0.001.
that this is due to the supporting structure giving during rapid The final case to be studied as representative of the series was
acceleration of the offset mass. The simulation assumes an in- conducted at 4400 rpm (see Fig. 14), the highest speed tested.
finitely stiff supporting structure, so start-up displacement am- Testing conditions were as follows in Table V.
plitudes due to high accelerations cannot occur. Since the differ- There was a 1.563-s delay ( ) between when the data input
ences in amplitude for this test were so different from the sim- was initiated and when motion began in the flywheel for the
ulation data, it is only presented as a qualitative comparison. experiment. Again, the data had to be shifted in time by in
The next studied test took place at speed of 4200 rpm, which order to compare with the simulated data. As in the 4200-rpm
was just above the critical speed of the shaft-rotor system in the test, there is good agreement between the model and the exper-
direction ( 64 Hz or 3840 rpm). This test gave an op- iment. Again, we do not have the pre-critical-speed envelope
portunity to observe transition through critical speed and to ob- centered at the same rpm as in the 4200-rpm case, which was
serve the qualitative shape of the amplitude envelope. Table IV 54 Hz 2 Hz. The transition between envelopes also occurs at
contains conditions for the 4200-rpm test, and Fig. 13 plots the about the same rpm as in the previous case. The overall per-
results. formance for the given adjusted values gave a 6% difference
There was a 1.768-s delay ( ) between when the data input between measured and adjusted values of stiffness and again
was initiated and when motion began in the flywheel for the ex- a 38% difference between the average measured and adjusted
periment, so the data had to be shifted in time by in order values for the damping coefficient, keeping in mind the previous
to compare with the of the simulated data. The results were case’s discussion on the value of damping coefficient.
typical of those that crossed the critical speed with a character-
istic envelope shape. The shape and amplitude was found to be VIII. CONCLUSION
highly sensitive to the damping ratio and the stiffness. This was
observed while varying the values of the stiffness and damping The primary goal of this study was to create an experimen-
coefficient in the model. The experimental data shows an enve- tally validated bond graph model of the Jeffcott rotor. This was
lope that precedes the predicted critical speed of 64 Hz. This done successfully. Three main points on the performance of the
envelope appears to be centered at about 54 Hz. This phenom- Lagrangian Bond Graph model are as follows.
enon is due to one of two things; either it is a coupling in the • There is excellent agreement ( 99.8 ) between the
shaft displacement amplitude from the direction, as discussed Hamiltonian model and the Lagrangian bond graph
previously, due to the asymmetric nature of the stiffnesses, or model.
something in the supporting structure has a resonance at this par- • The transient case for subcritical speeds shows poor agree-
ticular speed. Excluding this precritical-speed amplitude peak, ment with the experimental data, but is due to modeling as-
the simulation generally follows the shape and form of the crit- sumption of ideal Jeffcott rotor conditions, i.e., structural
ical-speed amplitude envelope. There is some slight difference resonances not accounted for in the supporting structure.
in the drop-off and in the transition between envelopes in the • The transient case for the spin-up through critical speed
experiment which does not exist in the simulation. The overall shows very good agreement ( 97 ) between the exper-
performance, given adjusted values, gave a 7% difference be- iment and simulation in the amplitude and shape of the
tween measured and adjusted values of stiffness and a 38% dif- resonant amplitude envelope if the resonant envelope that
280 IEEE TRANSACTIONS ON MAGNETICS, VOL. 41, NO. 1, JANUARY 2005

does not exist in the simulated data is excluded. Recalling [2] R. G. Loewy and V. J. Piarulli, [SVM-4] Dynamics of Rotating
again that the assumption of an ideal Jeffcott rotor is used Shafts. Washington, D.C.: The Shock and Vibration Information
Center, U.S. Department of Defense, 1969, pp. 1–4.
for the model, which is being compared to a “nonideal” [3] D. Childs, Turbomachinery Rotordynamics—Phenomena, Modeling,
simple shaft-rotor system. and Analysis. New York: Wiley, 1993, pp. 2–8.
[4] R. G. Loewy and V. J. Piarulli, [SVM-4] Dynamics of Rotating
The major modeling issue to come out of this study is that Shafts. Washington, D.C.: The Shock and Vibration Information
using the Hamiltonian method to analytically model this system Center, U.S. Department of Defense, 1969, pp. 6–10.
proved extremely tedious, which indicates that there could have [5] , [SVM-4] Dynamics of Rotating Shafts. Washington, D.C.: The
Shock and Vibration Information Center, U.S. Department of Defense,
been a better approach chosen as the baseline to the bond graph 1969, p. 2.
model. It should also be noted that most rotor/shaft systems of [6] J. M. Vance, Rotordynamics of Turbomachinery. New York: Wiley,
greater than two degrees of freedom will act like the two-de- 1988, p. 17.
[7] R. G. Loewy and V. J. Piarulli, [SVM-4] Dynamics of Rotating
grees-of-freedom case (i.e., the Jeffcott rotor) as observed by Shafts. Washington, D.C.: The Shock and Vibration Information
Dimentberg [15] in his canonical text on flexural vibrations of Center, U.S. Department of Defense, 1969, pp. 31–34.
rotating shafts. So, it should be concluded that the bond graph [8] J. M. Vance, Rotordynamics of Turbomachinery. New York: Wiley,
1988, p. 21.
model constructed should readily be able to describe rotor sys- [9] R. Neptune, “Class Notes on Hamilton’s Method,” Dept. Mech. Eng.,
tems of greater degrees of freedom. Because of the bond graph’s The University of Texas at Austin, 2003.
innate ability to connect systems and multiple energy domains, [10] J. M. Vance, Rotordynamics of Turbomachinery. New York: Wiley,
1988, p. 295.
the current bond graph model is well suited for use in more com- [11] J. J. Beaman and H. M. Paynter, Modeling of Physical Systems, 1993,
plex systems, thus making it a robust and adaptable element for unpublished, pp. 3.69–3.71.
modeling larger systems where multiple energy domains and [12] D. C. Karnopp, D. L. Margolis, and R. C. Rosenberg, System Dynamics:
A Unified Approach, 2nd ed. New York: Wiley, 1990, pp. 316–317.
multiple systems are interacting. [13] M. Hubbard, “Whirl dynamics of pendulous flywheels using bond
graphs,” J. Franklin Inst., vol. 308, no. 4, pp. 505–421, Oct. 1979.
[14] J. J. Beaman and H. M. Paynter, Modeling of Physical Systems, 1993,
REFERENCES
unpublished, pp. 6.22–6.35.
[1] J. M. Vance, Rotordynamics of Turbomachinery. New York: Wiley, [15] F. M. Dimentburg, Flexural Vibrations of Rotating Shafts. London,
1988, pp. 3–6. U.K.: Buttersworth, 1961, p. 43.

You might also like