You are on page 1of 8

©1995 IEEE. Personal use of this material is permitted.

However, permission to reprint/republish this material for advertising or promotional


purposes or for creating new collective works for resale or redistribution to servers or lists, or to reuse any copyrighted component of this work
in other works must be obtained from the IEEE.
IEEE TRANSACTIONS ON MAGNETICS, VOL. 40, NO. 3, MAY 2004 1617

New Electromagnetic Lift Control Method for


Magnetic Levitation Systems and Magnetic Bearings
Kent Davey

Abstract—The classical approach to gap control in active mag-


netic bearings—including those in magnetic levitation (maglev)
systems—is proportional–integral–derivative (pid) based current
correction. This paper explores a new method that simplifies
control electronics, based on repeatedly solving the governing
system equations in approximations that are valid for the next
20 to 40 ms. The method simplifies the magnetic forces by using
a Taylor approximation, one that can be evaluated rapidly by
using multivariate splines. The simplified equations of motion are
solved by the method of Frobenius. These simplified solutions are
inverted to predict the voltage necessary to achieve a desired gap
change in a specified time increment. Variations from this target
position allow for an update on inertia and mass of the levitated
object.
Index Terms—Control, feed-forward, maglev, magnetic bearing,
spline.

I. INTRODUCTION

E LECTROMAGNETIC levitation systems are generally


controlled by a proportional–integral–derivative (pid)
current correction to the lift magnet by active monitoring of Fig. 1. Comparison of current and voltage constrained movement.
the air gap [1], [2]. Newer controllers introduce fuzzy logic
control to conventional state-based control [3]. All current The voltage-controlled magnet has an immediate advantage
based controllers require a dual polarity source to rapidly drive over a current controlled system due to the reduced current that
the current up and down, as well as current interrupt devices. will be commensurate with movement. When a dc voltage
The motivation for this research was the desire to work with is placed across the coil of that electromagnet, changes in the
simpler fixed frequency voltage-controlled sources operating current are dictated by
off a three-phase bridge. Although the electronics of the
system are simpler, the control algorithm must necessarily be (1)
more robust since the current could not be driven down rapidly,
but only freewheel in a resistive decay cycle. Although the Consider two test cases, each starting with the same current and
algorithm proposed is quite appropriate for magnetic bearings, a small velocity of 0.1 mm/s to close the gap. That is, the elec-
discussion will focus on magnetic levitation (maglev) systems. tromagnet has an initial current producing a force which just
One-dimensional (1-D) control is the focus of this paper. If long equals the body’s weight, and the body is given a push to close
magnets are employed, no additional lateral control is required, the gap. The controller is disabled, fixed either with a constant
a fact adopted by high speed surface transport (HSST) [4]. current or a constant voltage. For the test, consider a 298-turn
If a fixed voltage is used to control the magnet, an advantage electromagnet with 30 A and a burden of 5600 lbs. As shown
is realized in that initial movements of the magnet toward in Fig. 1, fixing the voltage allows more reaction time since the
the track are commensurate with reduced currents within the current is compensated somewhat due to the last term in (1).
coil. Consider using an electromagnet to support a load. The The motivation behind this work is to use a simple power cir-
electromagnet can be controlled with a voltage source or a cuit, as shown in Fig. 2, to stably lift a maglev vehicle. The
current source. Of interest is the response comparison when the voltage on phases A, B, and C can be altered quickly. Master
magnet’s voltage is fixed against that when the current is fixed. drives allowing rapid voltage control like that shown are readily
available from a number of suppliers, such as Siemens Interna-
Manuscript received December 17, 2002; revised February 25, 2004. This tional. If the bridge is excited at 200 Hz, a 600-Hz ripple will
work was supported by American Maglev, Inc. appear on the magnet voltage. This fixed voltage should ripple
The author is with the Center for Electromechanics, University of Texas,
Austin, TX 78758 USA (e-mail: k.davey@mail.utexas.edu). and may suffice as a suitable signal voltage to relay the induc-
Digital Object Identifier 10.1109/TMAG.2004.827189 tance and thus the position of the magnet.
0018-9464/04$20.00 © 2004 IEEE

Authorized licensed use limited to: University of Texas at Austin. Downloaded on March 30,2010 at 12:12:06 EDT from IEEE Xplore. Restrictions apply.
©1995 IEEE. Personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional
purposes or for creating new collective works for resale or redistribution to servers or lists, or to reuse any copyrighted component of this work
in other works must be obtained from the IEEE.
1618 IEEE TRANSACTIONS ON MAGNETICS, VOL. 40, NO. 3, MAY 2004

Let , , and , and use the


fact that for , (3) becomes

(5)

Fig. 2. Master drive feeding a diode H bridge excites the electromagnet. Close examination of (1) shows that and play an equal
role in determining the change of . Therefore, another approx-
imation to current other than (4) is

(6)

With this change, (5) becomes

Fig. 3. An IGBT and a passive resistor allows the current to be driven down
more quickly.
(7)
Initial work suggested that an active switching device such as This is still an approximation since represents only the ve-
an insulated gate bipolar transistor (IGBT) along with another locity at time . Most the time simulations use a time period
passive component might be required to drive the current down, of 10 ms, and for these periods, there is no practical difference
as suggested in Fig. 3. When the IGBT is fired off, all the cur- between (5) and (7).
rent will be forced through the resistor, leading to a faster The analytical solution for (5) is
decay time. A similar circuit can be constructed using a 5- F ca-
pacitor in place of the resistor. In the latter case, the energy from
the magnet charges the capacitor with a time constant .
Both options require a separate trigger circuit for the IGBT, a (8)
complication that fortunately appears to be unnecessary given
the control algorithm to be discussed.
The velocity at any time is determined by taking the derivative
II. DC EXCITATION of (8)
In its simplest form, the electromagnet will be excited with
a dc voltage that can be scaled from 0 to 400 V nearly in-
stantaneously. The equation governing the rate of decay of that
current with respect to time is (9)

(2)
Note that the constants , , and represent very accurately
Consider a single magnet with mass (including burden) at the state of the magnet with position under saturation.
time , vertical velocity , and position . The equation of
motion governing 1-D motion in the direction would be III. SOLUTION WITH OPTIMIZATION
Control is achieved by the following procedure.
1) Determine a priori a nominal update time (10 ms) and
(3) desired target gap (10 mm).
2) Determine the desired step change toward that target
Here, is the magnet force at position , current . This
(0.1 mm).
force can be computed quite accurately using boundary element
3) Establish an objective each time step in terms of both the
analysis. The forces can be fitted to a divariate spline, and the
desired gap and the velocity after the update time.
partial derivatives determined in terms of the spline coefficients
[5], [6]. Equations (2) and (3) represent a pair of coupled dif-
(10)
ferential equations which have no analytic solution; the magnet
force varies as .
4) Use sequential quadratic programming to choose the best
Approximations can be made that do allow the solution to
to minimize [7].
be worked with reasonable accuracy over short periods of time.
5) After the time step, knowing the actual distance moved,
If the inductance does not change considerably over the time
update the mass. Sequential quadratic programming is
period of interest, the well-known solution for the current is
used to minimize
(4)
(11)

Authorized licensed use limited to: University of Texas at Austin. Downloaded on March 30,2010 at 12:12:06 EDT from IEEE Xplore. Restrictions apply.
©1995 IEEE. Personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional
purposes or for creating new collective works for resale or redistribution to servers or lists, or to reuse any copyrighted component of this work
in other works must be obtained from the IEEE.
DAVEY: NEW ELECTROMAGNETIC LIFT CONTROL METHOD FOR MAGLEV SYSTEMS 1619

Fig. 4. Gap position, velocity, weight prediction, and control voltage with time.

IV. OPTIMIZATION CONTROL RESULTS 2) Increasing the number of terms employed to represent the
magnetic force; specifically
Consider a 3.7-kg body supported by one magnet. Place it in
the field with a 12-mm gap and an initial closure velocity of
0.2 mm/s. Change the weight randomly by 5% every 100 ms.
Shown in Fig. 4 is the response of a single magnet test. The
value is set to 3 10 . This parameter represents a com-
promise between control and reaching the target gap. A large
is commensurate with a controller that takes awhile to reach its
objective. The mass is graphed normalized.
Every system perturbation is accompanied by a period of
about ten time periods (100 ms) for the controller to overcome
the commensurate transient. Shown in Fig. 5 is the performance
where
simulation when the mass is held constant.

V. DC VOLTAGE CONTROL WITH INCREASED SPEED


It is necessary to decrease the computation time and eliminate
the sequential quadratic programming step. Equation (5) can be
solved by the method of Frobenius, assuming that the gap
can be represented by

(12)

Inserting this series into (5) and equating terms of like power in
yields the solution
(14)

Inserting (12) into (5) yields the result


(13)

A bit more accuracy can be achieved by the following.


1) Increasing the number of terms in (12). (15)

Authorized licensed use limited to: University of Texas at Austin. Downloaded on March 30,2010 at 12:12:06 EDT from IEEE Xplore. Restrictions apply.
©1995 IEEE. Personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional
purposes or for creating new collective works for resale or redistribution to servers or lists, or to reuse any copyrighted component of this work
in other works must be obtained from the IEEE.
1620 IEEE TRANSACTIONS ON MAGNETICS, VOL. 40, NO. 3, MAY 2004

Fig. 5. Transient when the mass is kept constant.

If the mass is known, (15) can be solved for the voltage neces- At the end of each control cycle, the following events occur.
sary to move the vehicle to the next desired gap in seconds a) Measure the gap.
b) Compute the gap velocity.
c) Solve (17) for the mass.
d) Compute the next desired gap position.
e) Solve (16) for the necessary voltage to move the vehicle
to that position.
One additional perturbation is necessary to control with
(16)
voltage, and that is to force the velocity to be small. Equation
(15) can be differentiated to find the velocity
The mass of the real vehicle is never known a priori since the
number of people vary between trips. In addition, their distribu-
tion can change during the trip. Certainly the uplift on the nose
and tail will be dependent on speed. If a solid reaction rail is (18)
employed in the secondary, the ferromagnetic lift force of the The voltage necessary to drive this velocity to zero at the next
magnet to the guideway will be mitigated by speed-dependent time step is
eddy currents induced in that guideway. One way to deal with
these dynamics is to update the mass every cycle. If the starting
conditions are known ( , , , and ), the mass can be com-
puted in terms of the measured gap displacement

(19)

The voltage from (16) and (19) should be combined with a heavy
weighting on , e.g., 10/90 weight as

(20)

Using this method, the loop computation time on a 1.4-GHz


(17) Athalon computer is 7 ms.

Authorized licensed use limited to: University of Texas at Austin. Downloaded on March 30,2010 at 12:12:06 EDT from IEEE Xplore. Restrictions apply.
©1995 IEEE. Personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional
purposes or for creating new collective works for resale or redistribution to servers or lists, or to reuse any copyrighted component of this work
in other works must be obtained from the IEEE.
DAVEY: NEW ELECTROMAGNETIC LIFT CONTROL METHOD FOR MAGLEV SYSTEMS 1621

is the gap three time steps back. These first- and second-order
approximations to the gap are also displayed with the simulated
velocity.

VII. SIMULATION WITH THREE-PHASE EXCITATION


When the three-phase master drive inverter feeds the full
wave bridge, the resulting voltage is no longer a simple dc
signal. Set the voltage from the master drive to be

(22)

Suppose the frequency is 200 Hz. The signal out of the bridge
will have a 600-Hz ripple, as shown in Fig. 8. It can be approx-
imated by a Fourier series as
Fig. 6. Comparison of approximations for the equation of motion.

(23)
VI. SOLUTION WITH TIME SERIES REPRESENTATION
It should be clear from (12) and (14) that both the order of the
Note that the signal repeats with a period equal to one-third of
time series and the order of the force approximation are vari-
that coming from the master drive. Analysis reveals that
ables affecting the accuracy of these solutions. Increasing the
order of the representation of the magnetic force, i.e., the partial
derivatives in current and space, always enhances solution accu-
racy as long as the current and gap are not significantly different
from their starting values. If the time step is large, increasing the (24)
number of terms in the time series of (12) may not increase the
accuracy. Shown in Fig. 6 is a comparison of various approxi-
mations. As expected, increasing the representation order of the Here, is the voltage peak set by the master drive. Fig. 8 shows
force is helpful. But as the order of the force representation in- how well the fit results with 19 and 1000 terms.
creases, so must the terms in the power series so that those addi- The effect of this ripple turns out to be rather small for the
tional terms can be correctly allocated. The lesson to be learned control algorithm. Suppose the weight is changed randomly
from Fig. 6 is that the all representations are equally valid for every 300 ms by 5%. At each step, the voltage is computed
time steps ms. Larger time steps will incur error. to place the bogie at the next desired position. A step size of
Shown in Fig. 7 is a simulation of the lift using these rela- 0.1 mm is chosen for every 10 ms, attempting to move the
tions with a 3777-kg vehicle, changing the mass randomly every vehicle back to the target gap of 10 mm. Shown in Fig. 9 is
200 ms. The time step is chosen at 10 ms. The target gap is the gap position, velocity, weight, and voltage as a function of
10 mm, and a gap change of 0.1 mm is requested every 10 ms time.
toward the target. The algorithm is able to adequately adjust When the weight is not changed every 600 ms, the algorithm
the correct mass every time step. In addition to required con- is able to reach the target goal in about 1 s, as shown in Fig. 10.
trol voltage, mass, and resulting gap, the gap velocity, current,
and change in gap per time step are displayed. The gap velocity VIII. SELF SENSING
is computed using the temporal sequence of gap
The circuit in Fig. 2 is to be excited with a fixed frequency
( 200 Hz). The three-phase bridge dictates that there will be
a ripple signal at three times the excitation frequency (600 Hz)
on the dc signal going to the magnet. Presently, about 1.7 ms
or is required to collect data for a complete cycle of this 600-Hz
(21) ripple. The circuit in Fig. 11 is employed to sample data using
both a voltage and a current lem module. The capacitive–resistor
arrangement places a zero on the dc signal. If ),
Here, is the present gap measurement, is the gap one the ac signal which passes through relatively unabated. If these
time step back, is the gap two time steps back, and signals are sampled every 83 s, we gather 20 samples in a

Authorized licensed use limited to: University of Texas at Austin. Downloaded on March 30,2010 at 12:12:06 EDT from IEEE Xplore. Restrictions apply.
©1995 IEEE. Personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional
purposes or for creating new collective works for resale or redistribution to servers or lists, or to reuse any copyrighted component of this work
in other works must be obtained from the IEEE.
1622 IEEE TRANSACTIONS ON MAGNETICS, VOL. 40, NO. 3, MAY 2004

Fig. 7. Stable levitation using (16) and (17) to solve for control voltage and vehicle mass.

where . Let and be the subscripts for the voltage


and current signals, respectively. The inductance independent of
the relatively small resistance is

(26)

The values and are determined very easily from a two-by-


two matrix. Suppose samples are collected by the A–D card,
designated as . The matrix equation determining and is

.. .. .. (27)
. . .

Rearranging yields

Fig. 8. Voltage delivered out of the diode bridge.

period. The 600-Hz component for either the voltage or the cur-
rent can be represented as (28)
Signal (25)

Authorized licensed use limited to: University of Texas at Austin. Downloaded on March 30,2010 at 12:12:06 EDT from IEEE Xplore. Restrictions apply.
©1995 IEEE. Personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional
purposes or for creating new collective works for resale or redistribution to servers or lists, or to reuse any copyrighted component of this work
in other works must be obtained from the IEEE.
DAVEY: NEW ELECTROMAGNETIC LIFT CONTROL METHOD FOR MAGLEV SYSTEMS 1623

Fig. 9. Performance of the voltage control algorithm when the weight is changed randomly.

Fig. 10. Voltage control using the H bridge when the weight is altered every 600 ms.

Authorized licensed use limited to: University of Texas at Austin. Downloaded on March 30,2010 at 12:12:06 EDT from IEEE Xplore. Restrictions apply.
©1995 IEEE. Personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional
purposes or for creating new collective works for resale or redistribution to servers or lists, or to reuse any copyrighted component of this work
in other works must be obtained from the IEEE.
1624 IEEE TRANSACTIONS ON MAGNETICS, VOL. 40, NO. 3, MAY 2004

termination of and requires only 45 multiples. Ten sample


points may suffice.

IX. CONCLUSION
To satisfactorily control a magnet in lift using only dc voltage
which can never go negative, careful attention must be given to
the velocity of the magnet. Both this velocity and the gap po-
sition must enter into the criteria for selecting the voltage. The
voltage is chosen a priori to place the system at the next tar-
geted position in the next time interval. Time intervals greater
than 10 ms are not recommended. The ripple frequency at three
times the excitation frequency can be used to achieve a sensor
Fig. 11. Sensing circuit used to determine the inductance and thus the gap free lift by solving a simple equation that bypasses a fast Fourier
position.
transform sweep, the more conventional waveform analyzer ap-
proach. Although 1-D control has been the focus of this paper,
Solving for and yields the solution analysis shows that the method can be used to simultaneously
solve for lateral and vertical position.

REFERENCES
[1] A. D’Arrigo and R. Alfred, “Integrated electromagnetic levitation and
guidance system for the Swiss-Metro Project,” in Proc. Maglev 2000
Conf., Rio de Janeiro, Brazil, June 7–10, 2000, pp. 263–268.
[2] M. Sichinori, E. Masada, J. Baba, and T. Shimogata, “A hybrid control
scheme for electromagnetic suspension system of HSST,” in Proc. Ma-
glev 2000 Conf., Rio de Janeiro, Brazil, June 7–10, 2000, pp. 185–190.
[3] J. Van Goethem, F. Weber, and G. Henneberger, “Fuzzy control of a mag-
netic levitation system for a linear drive and comparison with a state con-
(29) trol,” in Proc. Magnetically Levitated Systems and Linear Drives Conf.,
Lausanne, Switzerland, Sept. 2002, PP08103.
[4] M. Jiasu, M. Tanaka, and S. Ishimoto, “Total running test operation of
HSST 100 and the project of East Hill side in Nagoya,” in Proc. Maglev
2000 Conf., Rio de Janeiro, Brazil, June 7–10, 2000, pp. 35–43.
This expression may appear foreboding, but the reader should [5] K. R. Davey, “Examination of various techniques for the acceleration
of multivariable multiminima optimization problems,” in 10th Biennial
note for a fixed number of sample points, and a fixed master IEEE Conf. Electromagnetic Field Computation, Perugia, Italy, June
drive excitation frequency, all the terms in the first large term in 16–19, 2002, p. 302.
(29) are known a priori. That is, the execution of (29) involves a [6] , “Use of tensor product splines in magnet optimization,” IEEE
Trans. Magn., vol. 35, pp. 1714–1717, May 1999.
few multiplies and is very fast. All of the sine and cosine terms [7] D. Goldfarb, “A family of variable metric updates derived by variational
are evaluated a priori. If 20 sample points are involved, the de- means,” Math. Comput., vol. 24, pp. 23–26, 1970.

Authorized licensed use limited to: University of Texas at Austin. Downloaded on March 30,2010 at 12:12:06 EDT from IEEE Xplore. Restrictions apply.

You might also like