Variable Bias Current in Magnetic Bearings for Energy Optimization

Variable Bias Current in Magnetic Bearings
for Energy Optimization
M. Necip Sahinkaya1 and Ahu E. Hartavi2
Department of Mechanical Engineering, University of Bath, Bath, U.K.
Faculty of Mechanical Engineering Automotive Controls and Mechatronics Research Center, Istanbul Technical University,
Istanbul, Turkey
We introduce an adaptive variable bias current control scheme to minimize the energy consumption of magnetic bearings without
altering the dynamic performance. We developed analytical expressions for the optimum bias current settings for both differential and
unidirectional modes of operation as a function of the orbit size and the desired bearing stiffness. The orbit size is measured in situ by a
recursive Fourier coefficient calculation. The analytical results are presented in normalized variables, and hence can be applied to any
size and type of magnetic bearing. Both zero and nonzero static load cases are considered. We show analytically that, under variable
bias current operation, the differential mode provides better energy efficiency and stiffness capacity than the unidirectional mode. We
present the energy consumption data obtained experimentally on a flexible rotor magnetic bearing rig. The data show that, with the
proposed variable bias current approach, a significant energy saving can be obtained without deterioration of dynamic performance.
Index Terms—Adaptive control, bias current, energy optimization, magnetic bearings.
PPLICATION areas of magnetic bearings are steadily expanding because of the significant advantages they offer
compared with conventional bearings [1]. They provide contact-free and oil-free support of rotating machinery without the
losses encountered due to friction in oil film bearings. They
can operate in hostile environments, at very high temperatures,
and at high speeds. They are also an ideal choice in applications where contamination is an issue such as turbo-molecular
pumps and artificial hearts. The power loss in magnetic bearings
is lower than for fluid film bearings, but there is potential for further significant reduction. Low power consumption is important
in several application areas, such as high-speed energy storage
flywheel systems, transport applications including future military vehicles where available power is concentrated to critical
operational functions, remote and unmanned air sea and land vehicles where the energy is provided by a battery, and micro/nano
Ohmic loss is one of the major energy losses in magnetic
bearings, and this is proportional to the square of the current
flowing in the coils. Other losses such as the eddy current and
hysteresis in the rotor and stator, and circuit loss in the electrical
equipment (cable, electronic cabinet) are also functions of coil
current. Therefore, minimization of the current has a significant
effect on the overall energy efficiency of magnetic bearings. Furthermore, the heat generated by the magnetic bearings will be
reduced with a decrease in coil current.
The design of low-loss magnetic bearings can be achieved by
modifying hardware and/or software. A hardware approach is
Digital Object Identifier 10.1109/TMAG.2006.888731
A color version of Fig. 9 is available online at
described in [2], where the geometry is optimized for energy
efficiency in an advanced energy storage system for future military vehicles. The effect of various design parameters on hysteresis, eddy current, and windage losses is studied experimentally in [3], and the effect of magnetic pole arrangement on the
core loss of a magnetic bearing at high speed and high flux density operation is reported in [4]. A new type of magnetic bearing
is proposed in [5], which has smooth flux distribution to reduce
rotating loss. The use of permanent magnets to carry the static
load provides savings in energy consumption at the expense of
design complexity and cost. A survey of early work is given in
[6], where experimental results are presented.
A software approach to reduce power loss involves modifying
the bias current and the controllers. Normally a fixed electric
current is supplied to each electromagnet, which is referred as
the bias current. The control current is superimposed on the bias
current. Therefore, the bias current flows in the coils even if no
force is required, causing power loss and rotor heating. The purpose of this dc bias current is to linearize the force-current and
force-displacement relationships, to extend the dynamic force
capacity, and to improve the dynamic performance. There are
three commonly used operation modes of power amplifiers [7].
The most popular are Class-A power amplifiers, where the dc
bias current is set at half of the maximum allowable current. The
control current is added to the bias current in one coil and subtracted from the opposing coil according to the force direction.
This is highly inefficient in terms of the energy consumption,
but it provides maximum range for the dynamic force with good
linearity and dynamic performance. A gain scheduled adaptive
control algorithm is suggested in [8] to minimize the speed-synchronous control current.
Class-B power amplifiers are more energy efficient as the
bias current is set to a much lower value. The control current
is added to the bias current of one of the coils only according
to the force direction. This is appropriate for low bearing stiff-
0018-9464/$25.00 © 2007 IEEE
ness and low vibration applications because the force slew rate
and the dynamic performance and controllability are poor. Performance degradation of Class-B power amplifiers is studied in
[9], [10]. Large feedback gains are required to achieve an acceptable bearing stiffness, and hence the possibility of current
saturation is high. The design of nonlinear controllers to handle
current saturations and stability problems is not trivial [11].
In the case of zero-bias Class-C modes of operation, nonlinearities are severe and the control singularity has to be mitigated. Linear controllers do not function for this mode of operations, but several forms of nonlinear controllers are suggested
in the literature. A switching control strategy is introduced in
[12], and used with a nonlinear controller involving an integrator
backstepping approach to minimize power losses, to avoid singularities, and to ensure stability. Various other nonlinear control designs for a single one-degree-of-freedom active magnetic
bearing are discussed in [13] via numerical simulations. The application of H-infinity compensation based on a control current
switching rule is studied experimentally in [14]. Apart from controller complexity, the poor dynamic performance and lack of
robustness against changes in the operating conditions such as
changes in unbalance distribution and external excitation, are
the main drawbacks of that approach.
In this paper, a variable bias current approach is exploited
to achieve a required dynamic performance with minimum
power consumption under varying operating conditions. An
earlier study [15] applied to turbo-molecular pumps showed
that minimum bias current does not always produce minimum
power consumption under dynamic loading such as that caused
by unbalance disturbances. Magnetic bearings already suffer
from a limited force capacity, and low bias current operation
decreases the dynamic load capacity significantly and render
the bearings vulnerable to changes in operating conditions.
The forces in an active magnetic bearing are generated by applying control currents to the coils. Two opposing poles generate
a bidirectional force along the axis of pole centers as shown in
Fig. 1. Defining as the distance of the shaft center from the
bearing center towards the coil 1, and as the radial clearance,
the force generated in the -direction can be written as follows
Fig. 1. Schematic view of the opposing poles and the axis system.
are shown in Table I, where
is the maximum allowable
current. The normalized magnetic force generated in the -direction can now be written as
The coil currents and are normally generated from a control current in either the differential or unidirectional modes.
In both modes, a constant bias current is supplied to opposing
coils even if no force generation is required.
In the differential mode (defined as Class-A mode with variable bias current), the control current is added to and subtracted
from the bias current to generate driving currents for both coils
as follows:
The bearing constant is a function of the permeability of free
space , the pole area , number of windings , and the half
angle between double pole faces
In order to generalize the analysis, normalized variables are
used throughout the paper. The normalized variables are denoted by capitalized symbols and the normalization constants
In the unidirectional mode (defined as Class-B mode with
variable bias current), only one of the coils receives the control
current depending on its sign
In the presence of a nonzero normalized static load , the bias
currents on the opposing poles are different to provide a constant
static force to counteract the static load. If is defined as the
mean bias current, i.e., the bias currents on two opposing coils
are set to
centrally positioned rotor
, (3) gives the following for a
The dynamic characteristics can be examined by assuming a
steady-state rotor synchronous response of the following form,
where is the normalized vibration amplitude in the -direction, and is the constant rotational speed of the rotor:
This would give a difference between bias currents in opposing
poles of
The magnetic force expression in (3) can be linearized for a
centered rotor around a zero control current
The negative bearing stiffness
and the current gain coeffiare functions of the bias current and can be obtained
by linearizing (3) as follows for both differential and unidirectional mode of operations:
For the unidirectional mode, due to discontinuity in (5), the current coefficient
is the average of current coefficients for positive and negative control currents. Because of their inherent negative stiffness properties, magnetic bearings are unstable and require a position feedback for stability
giving the following linearized bearing dynamics:
and must lie
It is important to note that the coil currents
between 0 and 1. This imposes restrictions on the bias current as
a function of the vibration amplitude and the bearing stiffness
In the differential mode, bias current limits are specified by
, and
, and can be written as follows after
substituting for
from (12):
For the unidirectional mode the bias current limits are
, which can be expressed in terms of the orbit
size and the desired bearing stiffness as follows after eliminating
Limits on the bias current settings can be interpreted as limitations on the achievable bearing stiffness as a function of the
maximum orbit size or vice versa. The maximum bearing stiff,
ness in the differential mode can be achieved when
i.e., a Class-A bearing. This gives the following expression for
the maximum bearing stiffness as a function of the orbit size :
The proportional feedback gain
produce positive bearing stiffness
should be large enough to
Alternatively, the following equation can be used to calculate
a position feedback gain in order to achieve a given bearing
as a function of the bias current and the static load:
For a unidirectional mode of operation, an expression for the
maximum bearing stiffness coefficient can be obtained from the
upper limit expression in (15). Differentiating this with respect
to gives the following optimum bias current setting as a function of the orbit size , and the corresponding maximum achievable value of the bearing stiffness:
To associate physical meaning to the normalized parameters,
consider a Class-A power amplifier (i.e.,
) with zero
static load. The normalized bearing characteristics are
The force capacity along one axis is
achieved by setting
and can be
as a funcFig. 2 shows the maximum achievable values of
tion of and the corresponding settings in order to achieve
the limiting performance. It can be seen from this figure that
the differential mode gives a better dynamic range for the mag, the
netic bearing. For example, for an orbit size of
maximum achievable bearing stiffness under differential mode
, but under the unidirectional mode,
. Thus, the differential mode
provides 2.24 times more stiffness than the unidirectional mode.
Fig. 2. Maximum achievable values of K and corresponding optimized bias
current settings as a function of orbit size A in both modes of operations.
Alternatively, for a given stiffness value of say
, the magnetic bearing will saturate when the orbit size reaches
in differential mode, but at
in unidirectional mode.
Fig. 3. Average power dissipation as a function of bias current for different
bearing stiffness and orbit sizes (solid: Differential, dashed: Unidirectional,
dashdot: Class-A).
), and the other for the positive control
) regions giving the following
current (i.e.,
average power dissipation:
A unified measure of ohmic losses (or
in normalized
form) is introduced in order to compare the energy efficiency
of different configurations by integrating the instantaneous
power dissipation of opposing coils along one axis over one
synchronous cycle and then averaging as follows:
For a Class-A power amplifier and zero static force, the above
for the maximum force (i.e.,
measure gives
for a steady-state centered position of the
case, and
rotor (i.e.,
Under variable bias current operations, the above cost measure can be minimized with respect to bias current as a func.
tion of the orbit size and the desired bearing stiffness
For a differential mode, if the current values under proportional
control are inserted into (18), this yields the following average
power dissipation expression as a function of the vibration orbit,
desired bearing stiffness and bias current:
Fig. 3 shows the power consumption as a function of for two
different values of bearing stiffness coefficients (
), and two different values of orbit sizes (
). The saturation and negative coil current regions
are not plotted. It is obvious from the figure that for the unidirectional mode, there is an optimum bias current setting, which
varies with the orbit size and the bearing stiffness. The smaller
bias current does not always provide better energy efficiency
under dynamic loading. Also in cases where higher dynamic
performance is required, this mode of operation is of limited
An expression for the optimum bias current settings to minimize energy consumption can be obtained by differentiating
and then equating it to zero.
(19) and (20) with respect to
For the differential mode this gives
The same procedure can be followed for the unidirectional
mode, where, due to discontinuity, the integration should be
broken into two regions; one for the negative control current
It can be shown however that this optimum value is always less
than the minimum allowable bias current given by (14), hence
it has no practical significance. This is consistant with findings
Fig. 4. Percentage energy saving when using optimum variable bias current in
differential mode compared with constant bias current of I = 0:5 (Class-A).
K values increase in the direction of the arrows from 0.1 to 1 with an increment
of 0.1, and then from 1.5 to 5 with an increment of 0.5.
reported in [16]. Therefore, the optimum bias current setting for
the differential mode is
Fig. 5. Optimum bias current settings and the corresponding average power
dissipation as a function of orbit size.
This gives the minimum average power dissipation of
Fig. 4 shows the percentage energy saving, as a function of
the orbit size and the bearing stiffness, by using the optimum
variable bias current in (22) compared with the conventional
constant bias current mode of operation.
As expected, the savings are significant at low vibration levels.
The value at which the constant bearing stiffness curves intersect with the zero energy saving line (horizontal axis) indicates
the limiting case where the optimum bias current approaches
the 0.5 value. The control currents would saturate for orbit sizes
greater than those at the crossing points. This is not an issue for
) settings. In many applications, the norsoft bearing (low
malized retainer bearing clearance is set to below 0.6; hence,
the use of a normalized bearing stiffness of 0.7 or lower would
ensure that no saturation would occur for vibrations within the
clearance of the retainer bearings. The “best” value for bearing
stiffness is a tradeoff between transmitted forces and the rotor
For the unidirectional mode, the optimum setting of the bias
current is
by . Therefore, all regions in the figure may not be usable dedue to current
pending on the value of the bearing stiffness
limitations as set out by (14) and (15). The results clearly show
the advantages of using differential mode variable bias current
controllers in terms of the energy consumption.
As a numerical example, consider a radial bearing for a ver. If
tical rotor with a desired bearing stiffness of
(i.e., 10% of the air gap) at a
the shaft orbit size is
steady-state operating running speed, under Class-A mode with
, a position feedback gain of
is required
from (12). This gives an average power dissipation of
However, under variable bias control, the minimum energy con, and
sumption can be achieved with
giving the same desired bearing stiffness and hence the same
orbit size but with a much lower power dissipation figure of
. That is, an energy saving of approximately 92% that
results in quieter operation and less heat generation. If the orbit
size changes due to external disturbances or transient effects, the
to stay on the optimum energy
controller can adjust and
consumption conditions without affecting the system dynamics.
If the controller is adaptive and selects the bearing stiffness in
accordance with the operating conditions, such as fuzzy logic
can easily be achieved
controllers [15], then this desired
by using the proposed optimized bias current, and the corre. A discussion of control strategies is outside the
scope of this paper, but the proposed control of the bias current
can be used with any open-loop, closed-loop, or nonlinear controllers with slight modification.
This gives a minimum average power dissipation of
Fig. 5 shows the optimum bias current settings and the corresponding average power dissipation curves as a function of the
orbit size for both modes of operation. This figure is normalized
In this section, the effect of the static load on the bearing
dynamic properties, bias current optimization and energy consumption is studied. The static load carrying capacity is provided by the difference in bias current in the opposing coils as
discussed earlier. The bias current in coils opposite to the direc, and
for the
tion of the static load is set to
Fig. 6. Bearing stiffness capacity as a function of the orbit size for different
static load levels for both mode of operations. The static load increases in the
direction of arrows from 0 to 0.5 with increments of 0.1.
opposing coils. The bias current denotes the average (mean)
bias current, and
can be calculated from (6). As shown in
(8), the static load increases the negative bearing stiffness by
, and an increase in position feedback gain is required
in order to achieve the desired bearing stiffness as given by (12).
This also reduces the stiffness capacity of the magnetic bearing.
Fig. 6 shows the reduction of bearing stiffness capacity with
increasing static load for both differential and unidirectional
modes of operations. The zero static load curves are the same
as shown in Fig. 2. The unidirectional mode again gives inferior bearing stiffness capacity, and saturates at a lower orbit size
compared with the differential mode. Obviously, the design of
a magnetic bearing should take into account the static load as
discussed in [17].
An average power dissipation expression can be obtained by
in (4) for the differential mode, and coil current
expressions into (18). This gives the following term to be added
to (19):
Fig. 7. Optimum bias current settings under static loading and the corresponding average power dissipation as a function of the orbit size A for
K = 0:5 (The static load increases in the direction of arrows starting from 0
with increments of 0.1).
Fig. 8. Percentage energy saving of using the optimized bias current compared
with Class-A mode for K = 0:5. The static load increases in the direction of
arrows starting from 0 with increments of 0.1.
with respect to gives the following
optimum bias current setting, which includes the effect of static
This converges to (22) for
. Fig. 7 shows the optimum
bias current settings for a typical bearing stiffness of
and the corresponding average power dissipation figures as a
function of the orbit size . An increase in static load requires
a higher bias current setting, and hence there is higher energy
consumption. The percentage energy saving compared with the
Class-A mode is shown in Fig. 8. An increase in static load
reduces the percentage energy savings.
As a numerical example consider a horizontal rotor supported
by two radial magnetic bearings. For a desired bearing stiffness
, and a static load at each axis of the magnetic bearof
to the vertical) of
, the bias currents
ings (which are
at the lower and upper coils should be 0.4 and 0.6 respectively
) for Class-A mode of operation. If the orbit
size is
at the steady-state operating speed of the rotor,
Fig. 9. A view of the experimental rotor bearing system used in the simulations.
then the average power dissipation is
. However, under
variable bias current differential mode, the optimum bias current
, giving the lower and upper coil bias currents of apis
) with
proximately 0.09 and 0.45, respectively (i.e.,
an average power dissipation of
. This corresponds to
an energy saving of approximately 58% without changing the
bearing stiffness, and hence the same orbit size.
The experimental magnetic/bearing system used to validate
the theoretical analysis consists of a uniform flexible steel shaft
of length 2 m and radius 0.025 m, with four 10-kg disks of radii
0.125 m. The total rotor mass is 100 kg and it is mounted horizontally on two radial magnetic bearings. The control current
of the magnetic coils was supplied through eight amplifiers. The
magnetic bearings have an air gap of 1.3 mm and each is protected by a retainer bearing having a 0.75 mm clearance. The
magnetic bearing data are provided in Table II. Eight eddy-current displacement sensors are placed at four planes with a 45
angle with the vertical line. A schematic view of the experimental setup showing the rotor, magnetic bearings, retainer
bearings and sensors is provided in Fig. 9.
A dSPACE digital signal processor (DSP), together with
Simulink and Realtime Workshop software, was used to implement the control algorithm. A rigid rotor assumptions were
made because the experiments were conducted at 10 Hz, which
is well below the first flexural critical speed of 30 Hz. The orbit
size was measured by a recursive calculation of synchronous
Fourier coefficients along the axis of the magnetic forces. This
approach enables changes in orbit size to be estimated without
excessive delay. A PID controller was used to control the rotor,
and the proportional gain set to achieve a bearing stiffness
N/m (i.e.,
). Derivative and integral
controller gains were set to achieve effective bearing damping
N s/m, and integral action of
N/m/s. These
physical values were divided by
before being supplied to
the PID controller in order to reflect the effect of changing bias
current. The optimum bias current control system to achieve a
specified bearing stiffness is shown in Fig. 10.
Fig. 10. Bias current control to minimize energy consumption for a specified
bearing stiffness k (RFT: recursive Fourier transform).
The average power dissipation per Ohm of coil resistance was
calculated by integrating the square of the measured coil currents
over one synchronous period, and dividing the results by the syn. An average over 10 periods was taken to
chronous period
minimize the effect of measurement noise. The unknown unbalance distribution along the rotor was the only external forcing to
generate a steady-state orbit. The rotor weight was supported by
the two magnetic bearings. The coils were placed with a 45 angle
to the vertical, hence the static load at each axis was
. The synchronous vibration amplitudes at each
axis is calculated by a recursive Fourier transform algorithm. The
maximum of the amplitude values along two axes was used as the
orbit size for each magnetic bearings. Under the Class-A mode of
A), bias currents for the upper and lower coils
operation (
were set to 4.3 A and 5.7 A, respectively, giving vibration amplitudes of 86.4 m and 100.1 m along the two axes for MB-2
). The predicted average power dissi(i.e., the orbit size
pation for MB-2 was 102.63 W/ , compared with the measured
consumption of 102.44 W/ . When the system was switched to
variable bias current mode, the optimum bias current for MB-2
A. The measured average power
settled at 2.04 A with
dissipation was 25.80 W/ , which corresponds to 74.8% energy
The top plot in Fig. 11 shows the predicted and measured
average power dissipation when the bias current was manually
decreased from 5 A to 2.2 A with an increment of 0.2 A. The
middle plot shows the predicted and measured percentage energy savings compared with the Class-A mode of operation. The
significant energy saving introduced by the variable bias current
The optimum bias current expression as a function of the orbit
size and the required bearing stiffness are obtained analytically
for both differential and unidirectional mode of operations, and
these formed the basis of the suggested adaptive control of the
bias current. The effect of static loading on the energy consumption and the maximum achievable bearing stiffness is also
The suggested bias current control technique was applied
to an experimental rotor horizontally supported by two radial
magnetic bearings. Experimental results showed the validity of
the analytical predictions. The measured energy savings agreed
well with the analytical predictions. The proposed method
was able to achieve 74.8% energy saving compared with the
standard Class-A mode of operation.
Fig. 11. Experimental and predicted results for MB-2 when the bias current
is manually decreased. Top plot: average power dissipation, Middle plot: percentage energy saving compared with Class-A mode of operation, Bottom plot:
i for static force generation.
method can clearly be seen. The bottom plot shows the predicted
, which provides the static lift force. Integral
and measured
action ensures zero steady-state errors adjusting the bias currents at opposing coils. The results suggest that the theoretical
values were slightly higher than the measured difference. This
may be due to errors in the linearized model and also misalignment errors of opposing poles.
As described above, the optimum bias current was set in accordance with (27), which aims to achieve zero minimum current at lower coils. However, due to measurement and modeling errors, and also due to inaccuracies in the recursive Fourier
transform especially under transient conditions, the control current may drive the coil current temporarily to negative values.
Although negative currents are prevented by a zero saturation
block, this saturation may not be desirable. This can be prevented by adding a safety margin to (27). Because such transient saturation was experienced during the experiments, further
tests were carried out with safety margins of 0.1, 0.2, and 0.3 A.
The power savings were slightly reduced to 73.9%, 72.8%, and
71.5%, respectively, compared with the 74.8% saving with zero
safety margin. Because of the small cost of introducing a safety
margin, it may be advisable to implement it if zero saturation
may be a potential problem for the power electronics or for the
system dynamics. It was observed that a safety value of 0.2 A
was sufficient to ensure that no saturation was occurred during
the experiments.
Magnetic bearings offer contact free support of rotating machinery with losses less than that of the oil-film bearings. However, there is still a scope to achieve significant energy savings through implementing adaptive control of the bias current.
[1] G. Schweitzer, “Active magnetic bearings—Chances and limitations,”
in Proc. IFTOMM Sixth Int. Conf. Rotor Dynamics, Sydney, Australia,
2002, vol. 1, pp. 1–14.
[2] M. A. Pichot and M. D. Driga, “Loss reduction strategies in design
of magnetic bearing actuators for vehicle applications,” IEEE Trans.
Magn., vol. 41, no. 1, pp. 492–496, Jan. 2005.
[3] P. E. Allaire, “Rotor power losses in planar radial magnetic bearings—Effects of number of stator poles, air gap thickness, and
magnetic flux density,” Trans. ASME, J. Eng. Gas Turbines Power,
vol. 121, no. 4, pp. 691–696, 1999.
[4] L. S. Stephens and C. R. Knospe, “Effect of magnetic pole arrangement
on core loss in laminated high-speed magnetic journal bearings,” IEEE
Trans. Magn., vol. 32, no. 4, pp. 3246–3252, Jul. 1996.
[5] N. Kurita, R. Kondo, and Y. Okada, “Lossless magnetic bearing by
means of smoothed flux distribution,” in Ninth Int. Symp. Magnetic
Bearings (ISMB9), Lexington, KY, 2005.
[6] E. H. Maslen, P. E. Allaire, M. D. Noh, and C. K. Sortore, “Magnetic bearing design for reduced power consumption,” Trans. ASME,
J. Tribol., vol. 118, no. 4, pp. 839–846, 1996.
[7] Mechanical vibration. Vibration of rotating machinery equipped with
active magnetic bearings, Part 1: Vocabulary, BS-ISO 14839-1, 2002.
[8] F. Betschon, “Reducing magnetic bearing currents via gain scheduled
adaptive control,” IEEE/ASME Trans. Mechatronics, vol. 6, no. 4, pp.
437–443, Dec. 2001.
[9] K. R. Bornstein, “Dynamic load capabilities of active electromagnetic
bearings,” Trans. ASME, J. Tribol., vol. 113, no. 3, pp. 598–603, 1991.
[10] T. S. Hu, “Reducing power loss in magnetic bearings by optimizing
current allocation,” IEEE Trans. Magn., vol. 40, no. 3, pp. 1625–1635,
May 2004.
[11] P. Tsiotras and M. Arcak, “Low-bias control of AMB subject to voltage
saturation: State-feedback and observer designs,” IEEE Trans. Control
Syst. Technol., vol. 13, no. 2, pp. 262–273, Mar. 2005.
[12] N. Motee and M. S. de Queiroz, “A switching control strategy for magnetic bearings with a state-dependent bias,” in Proc. 42nd IEEE Conf.
Decision and Control, Hawaii, 2003, vol. 1–6, pp. 245–250.
[13] P. Tsiotras and B. C. Wilson, “Zero- and low-bias control designs for
active magnetic bearings,” IEEE Trans. Control Syst. Technol., vol. 11,
no. 6, pp. 889–904, Nov. 2003.
[14] S. Sivrioglu, K. Nonami, and M. Saigo, “Low power consumption nonlinear control with
compensator for a zero-bias flywheel AMB
system,” J. Vib. Control, vol. 10, no. 8, pp. 1151–1166, 2004.
[15] M. N. Sahinkaya, A. E. Hartavi, C. R. Burrows, and R. N. Tuncay,
“Bias current optimization and fuzzy controllers for magnetic bearings in turbo molecular pumps,” in Ninth Int. Symp. Magnetic Bearings
(ISMB9), Lexington, KY, 2004.
[16] D. Johnson, G. V. Brown, and D. J. Inman, “Adaptive variable bias
magnetic bearing control,” in Proc. Amer. Control Conf., Philadelphia,
PA, 1998, pp. 2217–2223.
[17] D. K. Rao, “Stiffness of magnetic bearings subjected to combined
static and dynamic loads,” Trans. ASME, J. Tribol., vol. 114, no. 4, pp.
785–789, 1992.
Manuscript received July 25, 2006; revised November 17, 2006. Corresponding author: M. N. Sahinkaya (e-mail: [email protected]).
M. Necip Sahinkaya received the B.Sc. degree from the Istanbul Technical
University, Istanbul, Turkey, in 1974, and the M.Sc. and Ph.D. degrees from the
University of Sussex, Sussex, U.K., in 1976 and 1979, respectively.
Following periods as an academic at the universities of Sussex, Strathclyde
and Bath, he joined an international shipping company as a general manager
in late 1988. He rejoined the University of Bath in May 2000, and is now a
Reader in the Department of Mechanical Engineering. His research interests
include active control of magnetic bearings, control and dynamic analysis of
multi-physics systems. He was the Chair of the Organizing Committee for the
UKACC Control 2004 conference.
Ahu E. Hartavi received the B.Sc., M.Sc., and Ph.D. degrees in electrical engineering from the Istanbul Technical University, Istanbul, Turkey, in 1997, 2000,
and 2005, respectively.
She is currently working as a Post Doctoral Researcher in the Automotive
Controls and Mechatronics Research Center (AUTOCOM) at the Istanbul Technical University. Her research interests include modeling, design, and control of
active magnetic bearings, power optimization, electrical machines and drives,
modeling and control of powertrain, hybrid electric vehicles, solar and fuel cells.
She is the Workshops Chair of the 2007 IEEE Intelligent Vehicle Symposium.