Experiment on a Permanent Magnet-Biased Magnetic Bearing

with a Cascade Digital Control

Mohamed S. Kandil *, Maxime R. Dubois *, Loicq S. Bakay t, Joiio P. Troviio *

*e- TESC Laboratory, Department of Electrical & Computer Engineering, Universite de Sherbrooke, Canada,
((Mohamed.Kandil; Maxime.Dubois; Joao. Trovao}@USherbrooke.ca)
tGE Renewable Energy, Canada ([email protected])

eliminate or reduce the fixed bias current have been proposed

Keywords: Magnetic Bearings, Digital Control, Energy in order to minimize power losses [4], [5]. However besides
Storage, Flywheels. the complexity of control schemes, lowering or eliminating
the bias current could lead to poor dynamic performance and
Abstract less robustness against disturbances so researchers studied a
variable bias current strategy as in [2] . The hardware solution
A permanent magnet-biased active magnetic bearing (PM­ is an alternative approach which depends on employing
AMB) is an efficient design of pure active magnetic bearing permanent magnets [3], [6], [7]. In PM-AMBs, permanent
systems. This paper presents a digital decentralized double­ magnets (PMs) provide the fixed bias flux while the control
loop controller for a radial homopolar PM-AMB. The inner currents are used only for stabilization and damping
loop is for current control while the outer loop is the master vibrations. One of the first contributions to the design of PM­
one and is devoted for regulation the shaft position around the AMBs is reported in [8] [9].
operating point despite the harmonic excitation forces due to
This paper discusses the digital implementation of a
the inherent rotor unbalance. The Spectrum Digital eZdsp
decentralized feedback control for a radial homopolar PM­
F2812 board is used for control implementation.
AMBs using the cascade control structure. In the paper, the
Experimental results are given to validate the performance of
PID controller is used in the outer loop for position control
the system followed by calculation of the power consumed by
while the PI controller is used for the inner loop to control the
the radial bearing.
current. The continuous time controller is then discretized to
obtain its digital counterpart. The Spectrum Digital eZdsp
1 Introduction F2812 board is used for control implementation. The main
An active magnetic bearing (AMB) is a mechatronic system contribution to this paper is the application of a radial
that have been widely used to achieve a contactless support homopolar PM-AMB to support a spinning shaft with a
for a spinning rotor. It is an inherently nonlinear and an open­ larger-scale size and heavier weight compared to the majority
loop unstable system, thus a feedback control loop is a must of reported cases in literature [3], [7] -[11] as well as the DSP
to attain stability besides achieving a satisfactory implementation of the cascade controller. The description of
performance. It is not surprising that most of the literature on the employed system as well as a simplified mathematical
AMBs concerns control schemes [1]. The commonly used model are given first. The discretization process is then
control scheme for AMBs adopt a fixed bias current supplied explained. Experimental results are given to validate the
to each coil to improve the linearity and dynamic effectiveness of the closed loop system under different
performance, and to enable a single power amplifier per axis operating conditions.
(less costly) if differential winding connection is adopted.
Another challenge connected to AMBs is their power 2 System Description
consumption. Although the losses are much lower than roller
The radial PM-AMB has two lamination stacks of four teeth
bearings, these losses can limit the operation in some
each. In between the two lamination stacks are located the
applications such as flywheel energy storage systems and
PM segments, as described in [6], see Fig. 1. Each tooth is
vacuum applications [2]. Copper losses which is proportional
wound with a coil of N turns (N 100) and these 8 stator
=
to the square of the current flowing in the coils have a key
coils are positioned along the vertical (4 coils) and horizontal
role in the power losses in magnetic bearings. Other losses
(4 coils) axes, thus forming 4 electromagnet (EM) poles. The
such as iron losses are also proportional to the coil current.
four EM coils for each of the two axis are connected in series,
Therefore enhancing the energy efficiency of magnetic
thus 4N coils per axis. Fig. 2 shows a picture of the test rig.
bearings depends on minimizing coil currents.
The driven terminal of the shaft is supported by a mechanical
There are two approaches to improve the energy efficiency of ball bearing while the non-driven terminal is supported by the
magnetic bearings [3]. For the first approach which adopts the radial homopolar PM-AMB. The rotation is realized through
software solution, various feedback control strategies which an induction motor which is connected to the shaft by means
 Coil Rotor

Stator

PM

Fig. I Radial homopolar PM-AMB.

of a flexible coupling. Two single phase inverter circuits (one Fig. 2 Radial HMB test rig: (I ) Safety bearing, (2) Position
for each axis) are used to amplify the control current. The sensor,(3) HMB, (4) Shaft, (5) Ball bearing, (6) Flexible
inverter is realized with a full H-bridge configuration coupling, (7) Induction Motor
consisting of four IGBTs with 30Y DC-link because
bidirectional current is needed for homopolar PM-AMB where
applications unlike pure AMBs which can employ half H­ p center of gravity coordinate
bridge configuration. Sensors are a must to close the feedback M mass matrix
control loop. Two position sensors are used; one to monitor G gyroscopic matrix
the deviation along the horizontal axis while the other one to B coordinate transformation matrix
monitor the deviation along the vertical axis. The position Cbb damping matrix
sensors are of inductive type. Their linear span is from 0 to 2 Ks HMB stiffness matrix
mm with a resolution of 1 .um. Most off the shelf current Kbb ball bearing stiffness matrix
amplifiers of this type utilize an on-board analog PI control. Ki force to current factor matrix
In this work a digital current control is employed because of
Fun harmonic vibration forces matrix
the flexibility in implementation and tuning compared to
UJ rotation speed in radls
analog control. Hence two current sensors are also required
for closing the current control loop.
2.2 Electrical Dynamics

2.1 The Rotor-Bearing System Model It is assumed that the total resistances and inductances of
each of the two control windings are equal to Rand L
The net electromagnetic force of a homopolar PM-AMB
respectively. The electrical dynamics per axis in the system
generated per axis can be formulated as [12]
Ag can be described as

F =
f..Jo
{ (BplIl +Bc f - (BpnI -Bc f } (1) di
-
=
kg v
-
R.
-- / (4)
dt L L
where Ag is the air-gap area under one tooth, /10 is the where kg is the current amplifier gain, v is the control voltage.
permeability of free space, Bpm is the air-gap flux density due
to PMs while Be is the air-gap flux density due to control
3 Digital Controller
coils. It is common to linearize the force formula around the
operating point and use the following simplified expression A single radial magnetic bearing is usually represented as a
F = ks q + kc i (2) two mechanical DOFs system for rigid body rotor, where
each DOF describes the deviations of the shaft in the
where ks and kc are the positon stiffness and force to current horizontal and vertical directions. The dynamics of the two
gain respectively, q is the position signal deviation measured directions can be decoupled if the gyroscopic effect is not
from the equilibrium point, i is the control current . significant which is true here. Thus two separate feedback
It is assumed that the rotor is symmetric and rigid, and the control systems can be designed for each mechanical DOF,
axial motion is decoupled from the radial ones. Therefore the this is the so called decentralized control approach.
radial dynamics can be represented by 4 degrees of freedom
(DOF) while the axial dynamics is I -DOF which is not being
3.1 Cascade Control Structure
of particular interest here. The equation of motion for a rotor
suspended with a HMB and a ball bearing can be formulated The commonly used control structure for both electric drives
as [l3] and magnetic bearings is the cascade-loop control "or double­
(
Mp + wG + BebbBT p + BKsBT + BKbbBT P ) ( ) (3)
loop control" [1] [14] . The inner control loop is responsible
= BKji + Fun
for controlling the current. For AMBs, the outer control loop
is the primary or the master loop and is devoted for

2
 Double-loop digital controller

AID I+----------------'-----i

1 ______----------------------------------------

Fig. 3 Digital double-loop control structure

controlling the shaft position. The primary objective of the component to dampen oscillations around the equilibrium
feedback control system is to stabilize the inherently unstable point [1]. The simplest controller to achieve these
system first and then regulating the shaft position around the requirement is the well-known proportional-derivative (PD)
set-point. The input to the system is the position reference controller. For practical application, with a PD control there is
signal (usually the geometric center position of the bearing), always an offset error between the controlled variable and the
and the output signal is the position of the shaft. Fig. 3 set-point, no matter how high the P-gain. Addition of integral
illustrates the employed controller structure. action is the remedy to this situation since it is able to
eliminate this offset error. Hence the outer control loop for
real systems becomes the proportional-integral-derivative
3.2 Inner Control Loop
control (PID). The position error eq between the position
The function of the inner loop is to regulate the measured reference signal qr and the measured position signal q can be
current i to track the current reference signal ire! generated by defined as
the outer control loop. The current error is amplified and used (8)
to calculate the corresponding duty cycle to control the output
voltage of the H-bridge. By employing a high-gain controller, The PID controller for the outer loop in Laplace domain can
it is expected that the control current can follow accurately be described as
the current reference signal. It is unadvised to use a high
proportional feedback gain (P-controller) only, because the
current amplifier will be prone to saturation as well as the
( ;
Uo(S) = kp+ +kdS Eq(S) J (9)

high tendency of noise generation. The typical approach is to

use a proportional-integral (PI) for current control. The
kp ki
where is the proportional gain, is the integral gain, and kd
is the derivative gain for the outer control loop. The
current error signal e, can be described as expression of the controller given in Equation (9) cannot be
implemented in practice. First of all because the transfer
e, = ire!-i (5)
function of the controller is not proper'. The other problem is
The continuous-time PI controller for the inner loop in the so­ that a pure derivative action is not allowable in practice
called parallel form is formulated as because of the amplification of the high frequency

�
measurement noise. The commonly used solution to these
u,(t) = cp.ei (t)+Ci. J e, (r)dr (6) problems is to cascade the derivative term with a low pass
filter. The practical form of PID controller can be written as
Cp Ci
where is the proportional gain and is the integral gain for
the inner control loop. And the Laplace equivalent form is
(10)

(7)
where T is the time constant of the low pass filter.

3.4 Controller Discretization

3.3 Outer Control Loop

The current controller and the position controller were
The outer loop, which is also known as position control loop, described by Laplace transform in Equations (7) and (10)
is the main controller. Since AMBs are inherently open loop respectively. Since we want to implement these controllers in
unstable systems, the first goal of the main controller is to digital form, it is necessary to represent them in discrete-time.
stabilize the system to the equilibrium point. The intuitive The commonly used approach for designing a digital
reasoning is to imitate the mass-spring-damper (MSD)
system. Hence the controller is required to provide a restoring , In control theory, a transfer function is proper if its numerator does not
force similar to the mechanical spring as well as a damping exceed its denominator.

3
controller for a continuous-time system is to first design an
x 10.4 Steady state condition
analog controller and then convert it into an equivalent
discrete-time system that have a close approximate behavior
[15]. There are various techniques from the signal processing
field that can be used to convert an analog controller to its
digital form counterpart. The method we are going to use is
the approximation of differential equations by numerical 0.05 0.1 0.15 02 0.25 0.3 0.35 0.4 0.45 0.5
integration. There are commonly three methods for this Time (sec)
purpose, Forward Euler (FE), Backward Euler (BE), and

H"� : : : : : : ' --�"l

Trapezoidal method or Tustin's method. In this section we
present the discretization of the position controller and the
current controller can be done in a similar manner.
Discretizing a continuous-time system by Tustin method can
simply be done by replacing the Laplace variable s with 0.05 0.1 0.15 02 0.25 0.3 0.35 0.4 0.45 0.5
Time (sec)

(11) Fig. 4 Rotor horizontal and vertical displacements at 0 rpm

Staedy state with no-rotation

where Ts is the sampling time and z is the variable of Z­
transform. By following this rule, the integral and derivatives

( Z +I
terms can be discretized respectively as follows

I(z) kiT,
z-l J
=

(12)
2 0.05 0.1 0.15 02 0.25 0.3 0.35 0.4 0. 45 0.5
Time (sec)
and

(13)

0.05 0.1 0.15 02 0.25 0.3 0. 35 0.4 0.45 0.5

Since the proportional part is static, no discretization is Time (sec)

needed and thus Fig. 5 Rotor horizontal and vertical currents at 0 rpm
(14)
radial hybrid magnetic bearing. The control parameters for
and the complete outer-loop control signal would be the inner and outer loops were initially selected in simulation

(Z) {P(z)+I(z)+ D(z)} E( z)

=
using the system model described by (3) built in
Va (15) Matlab/Simulink environment and then finely tuned after
extensive experimentation to

4 Experimental Results Cp = 3, Cj= 10, kp = 3300, kj = 20000, kd= 8, FO.OOI

The control algorithm is implemented using the Spectrum
Table 1 presents the parameters of the system. Two
Digital eZdsp F2812 board. The board employs
decentralized feedback controllers are used to control the
TMS320F2812 DSP which is a 32-bit DSP with fixed-point
Rotor mass (m) 61.9 kg arithmetic and includes six dual PWM channels and 16
Rotor transverse moment of inertia 4.79 kg m2 ADCs. Analog anti-alias low pass filters with 1.5 kHz cut-off
frequency are used to attenuate the effect of high frequency
Rotor polar moment of inertia 0.086 kg m2
measurement noise. A sampling rate of 10 kHz is used.
Force to current factor (kcx = key) 609 N/A
HMB stiffness for horizontal motion (ksx) 28.05 N/mm
4.1 System Performance at 0 RPM
HMB stiffuess for vertical motion (ksy) 47.8 N/mm
Nominal PM-AMB air-gap length Imm Fig. 4 shows the horizontal and vertical displacements of the
Nominal safety bearing clearance length 0.5mm levitated shaft at steady state with 0 rpm rotation speed. The
peak-to-peak (P2P) displacement for the horizontal axis is
Coil resistance (R) l.l37 D
19.84 flITl while for the vertical direction it is 7.92 flITl. The
Coil inductance (L) 0.136 H
corresponding control efforts for the non-rotation steady state
DC voltage supply 30 V condition are given in Fig. 5.
Table 1 MODEL DATA FOR ROTOR-BEARING SYSTEM

4
 10.4 Rotation test

I VvvvVVV\Ol
x

·1 L-�--�--�--�--�--�--��L-�--�
o 0.05 0.1 0.15 02 0.25 0.3 0.35 0.4 0.45 0.5

Time (sec)

I " hfYV0i0!9
x 10.4
1

·1
o 0.05 0.1 0.15 02 0.25 0.3 0.35 0.4 0.45 0.5

 Time (sec)
 6   : '   
Fig. 8 Rotor horizontal and vertical displacements at 1000 rpm

rE� f :�
Rotation test

· L-�--�--�--�--�--��L-�--�--�
1
0.05 0.1 0.15 02 0.25 0.3 0.35 0.4 0.45 0.5
0.05 0.1 0.15 02 0.25 0.3 0.35 0.4 0.45 0.5
Time (sec)
Time (sec)

i�j � o 0.05 0.1 0.15 02 0.25

Time (sec)
0.3 0.35 0.4 0.45 0.5
i :� 0.05 0.1 0.15 02 0.25

Time (sec)
0.3 0.35 0.4 0.45 0.5

Fig. 7 Step response - vertical direction

Fig. 9 Rotor horizontal and vertical currents at 1000 rpm

The second case is for the system rotating at 1800 rpm. The
4.2 Step Response horizontal and vertical vibration amplitudes are shown in Fig.
10. The P2P horizontal and vertical displacements are 0.0877
Fig. 6 shows the transient response for a 0.05 mm step
mm and 0.113 mm respectively. The corresponding control
applied to the input of the horizontal direction closed loop.
efforts are shown in Fig. 11. The maximum vibration
The upper part of the figure illustrates the horizontal
amplitude for both cases is approximately 0.068 mm which
displacement while the corresponding control current is
represents 13.6% of the 0.5 mm safety bearing clearance.
presented in the lower part. The step response for the vertical
direction is presented in Fig. 7. The maximum horizontal

r��
displacement is 0.178 mm with 255% peak overshoot (PO)
10.4 Rotation test
while the maximum vertical displacement is 0.166 mm with x

232% PO. The time required for the vertical direction to settle
within ±50/0 of the equilibrium position is 0.1158 seconds.
For horizontal direction, it takes 0.1162 seconds to reduce
oscillations to amplitudes of 30% of the steady state value �
is
·1 L-�--�--�--�--��--�--�--��
before the system settles after approximately OA1 seconds. o 0.05 0.1 0.15 02 0.25 0.3 0.35 0.4 0.45 0.5

Time (sec)

i fjf!#!#!!!Jv1�
4.3 System Performance under Rotation

Two cases are given to address the behavior of the system

during rotation. Fig. 8 shows the vibration level of the shaft
spinning at 1000 rpm. The P2P horizontal displacement is
·1
0.1325 mm while the P2P vertical vibration amplitude is o 0.05 0.1 0.15 02 0.25 0.3 0.35 0.4 0.45 0.5

0.1058 mm. The control currents for the horizontal and Time (sec)
vertical axes for the 1000 rpm case are illustrated in Fig. 9. Fig. to Rotor horizontal and vertical displacements at 1800 rpm

5
 i :�
Current in Magnetic Bearings for Energy Optimization,"
Rotation test
iEEE Trans. Magn., vol. 43, no. 3, pp. 1052-1060, Mar.
2007.
[3] E. H. Maslen, P. E. Allaire, M. D. Noh, and C. K.
Sortore, "Magnetic bearing design for reduced power
consumption," J Tribol., vol. 118, no. 4, pp. 839-846,
1996.
0.05 0.1 0.15 02 0.25 0.3 0.35 0.4 0.45 0.5
[4] P. Tsiotras and B. Wilson, "Zero- and low-bias control
Time (sec)

i :�
designs for active magnetic bearings," iEEE Trans.
Control Syst. Technol., vol. 11, no. 6, pp. 889-904,Nov.
2003.
[5] S. Deno and M. N. Sahinkaya, "Adaptive Bias Current
Control in Active Magnetic Bearings for Energy
Optimization," Volume 1: 23rd Biennial Conference on
0.05 0.1 0.15 02 0.25 0.3 0.35 0.4 0.45 0.5
Mechanical Vibration and Noise, Parts A and B. 2011.
Time (sec) [6] L. Bakay, M. Dubois, P. Viarouge, and J. Ruel, "Losses
Fig. 11 Rotor horizontal and vertical currents at 1800 rpm in hybrid and active magnetic bearings applied to Long
Term Flywheel Energy Storage," Power Electronics,
Machines and Drives (PEMD 2010), 5th lET
4.4 Power Loss Evaluation lnternational Conference on. pp. 1-6,2010.
[7] S. Cheng and S. W. Day, "Design and control of hybrid
The power consumption of the radial homopolar PM-AMB is magnetic bearings for maglev axial flow blood pump,"
evaluated for two operating conditions 1000 rpm, and 1800 2010 lEEEIASME Int. Conf Adv. Intel!. Mechatronics,
rpm by calculating the ohmic loss according to pp. 187-192,Jul. 2010.
[8] C. Sortore, P. Allaire, and E. Maslen, "Permanent
(16) magnet biased magnetic bearings, design, construction
and testing," Second into Symp. Magn. Bear., 1990.
where Ix and Iy are the root-mean-square (RMS) value of the [9] Y. Fan, A. Lee, and F. Hsiao, "Design of a
control currents flowing in the electromagnetic coil of the permanent/electromagnetic magnetic bearing-controlled
horizontal and vertical directions respectively. Rx and Ry rotor system," J Franklin inst., vol. 334, no. 3,1997.
represents the total resistance for the horizontal and vertical [10] S. Fukata and K. Yutani, "Dynamics of Permanent­
coils respectively. The total ohmic losses for the radial Magnet Biased Active Magnetic Bearings," Kyushu
bearing are 0.182 watts and 0.205 watts for the first and Univ, Third Int. Symp. Magn. Suspenso Techno!., pp.
721-736,1996.
second cases respectively.
[11] A.-C. Lee, F.-Z. Hsiao, and D. Ko, "Analysis and
Testing of Magnetic Bearing with Permanent Magnets
5 Conclusion for Bias," JSME Int. journal. Ser. C, Dyn. Control.
Robot. Des. Manuj, vol. 37, no. 4, pp. 774-782,1994.
This paper presents the digital implementation of a
[12] Y. Zhilichev, "Analysis of a magnetic bearing pair with
decentralized double-loop controller for a radial PM-AMB of a permanent magnet excitation," Magn. iEEE Trans.,
the homopolar configuration. The description of the employed vol. 36, no. 5, pp. 3690-3692,2000.
system as well as a simplified mathematical model are given [13] M. S. Kandil, M. R. Dubois, J. P. Trovao, and and L. S.
first. The employed controller is then discussed. PI controller Bakay, "A Sliding Mode Control of a Hybrid Magnetic
is used for the inner loop while PID control is used for the Bearing for Wayside Flywheel Energy Storage
outer loop. Tustin method is used to discretize the feedback Systems," in Vehicle Power and Propulsion Conference
controller. Experimental results are given to validate the (VPPC), 2015 iEEE, 2015.
satisfactory performance of the closed loop system under [14] A. Hughes, Electric Motors and Drives: Fundamentals,
different operating conditions. The maximum vibration Types and Applications. Newnes,2005.
amplitude for the studied cases is approximately 0.068 mm [15] M. Santina and A. Stubberud, "Discrete-Time
which represents 13.6% only of the 0.5 mm safety bearing Equivalents of Continuous-Time Systems," in The
clearance. The total ohmic losses consumed by one radial Control Handbook, Second Edition, CRC Press, 2010,
PM-AMB is approximately no more than 0.21 watts at 1800 pp. 12-34.
RPM rotation speed.

References

[1] H. Bleuler, M. Cole, P. Keogh, R. Larsonneur, E.

Maslen, Y. Okada, G. Schweitzer, A. Traxler, and E. H.
Maslen, Magnetic Bearings: Theory, Design, and
Application to Rotating Machinery. Berlin, Heidelberg:
Springer Berlin Heidelberg,2009.
[2] M. N. Sahinkaya and A. E. Hartavi, "Variable Bias