2008 10th Intl. Conf.

on Control, Automation, Robotics and Vision

Hanoi, Vietnam, 1720 December 2008

Sensorless Speed Control of a Permanent Magnet

Type Axial Gap Self-Bearing Motor using
Sliding Mode Observer.
Dich Quang Nguyen

Satoshi Ueno

Dept. of Mechanical Engineering.

Ritsumeikan University.
1-1-1 Nojihigashi, Kusatsu, Shiga 525-8577, Japan.
E-mail: gr046070@se.ritsumei.ac.jp

Dept. of Mechanical Engineering.

Ritsumeikan University.
1-1-1 Nojihigashi, Kusatsu, Shiga 525-8577, Japan.
E-mail: sueno@se.ritsumei.ac.jp

AbstractThis paper presents an analytical and experimental

evaluation of sensorless speed vector control of an axial gap self
bearing motor, in which rotor speed and position are estimated
by using Sliding mode observer. The approach is based on the
estimation of the motor back-EMF (or induced voltage) through
a sliding mode observer with help of measured stator currents
and reference voltages. In order to achieve an accurate
estimation of the rotor speed and position, an adaptive gain of
sliding mode controller is proposed. In addition, during the
operation of system, low pass filters are used to smooth the
estimated variables. Experiment is implemented based on
dSpace1104 with two three phase inverters. Results confirm that
axial force and rotating torque can be controlled independently
and motor can get the good performance in steady state at the
average and high speed range, so further work is being continued
to improve the experiment for this proposal.
Keywords Combined motor-bearing, Self-bearing motor,
PM motor, Sensorless, Sliding mode observer.

I.

INTRODUCTION

Recently, magnetic bearing motor becomes more

interesting in many researches thanks to its advantages as high
speed range, less maintenance and working ability in harsh
environment, etc. Magnetic bearing motor usually has
structures like a rotary motor installed between two radial
magnetic bearings or combination of rotary motor and radial
magnetic bearing. However, it makes machines increase in size
and weight, which causes problems in some applications that
have limit space [1],[2]. For this reason, simpler and smaller
construction and less complex control system are desirable.
Satoshi Ueno has introduced an axial gap self-bearing motor
(AGBM) which is a combination of a disc motor and axial
magnetic bearing, it is quite simpler in structure and control
than conventional magnetic bearing-motor since hardware
component can be reduced [3],[4],[5],[6].
Fig. 1 shows the structure of an axial gap self-bearing
motor. The radial motions x, y, x, y of the rotor are dependent
on the control of radial magnetic bearings. Therefore, in case of
ideal radial magnetic bearings, these motions are constrained.
In this paper, only rotational and translational motions of rotor
along z axis are considered. The motor has two degrees of

c 2008 IEEE
978-1-4244-2287-6/08/$25.00 

Figure 1. Structure of PM type axial gap self bearing motor.

freedom. The rotor is a flat disc with permanent magnet (PM)

attached on two faces of disc, hence, it is a non-salient-pole
rotor. Two stators, one in each side of rotor, have three-phase
windings to generate the rotating magnetic flux in the air gap
that produces the motoring torque T to the rotor and generates
the attractive forces between the rotor and the stators F1 and F2.
The axial force F is the difference between two attractive
forces and is controlled by changing the magnitude and phase
of the currents in stator 1 and 2 with respect to rotor. Since the
attractive magnetic force is unstable, an axial position feedback
control is required to stabilize the axial translation.
In order to control its rotating torque independently from
the axial force, the vector control is a reasonable choice [5],[6].
But, for vector control, it is necessary to know the rotor
position for changing coordinate. Normally, the rotors position
can be obtained by a shaft-mounted encoder or a resolver.
These components cause the several disadvantages, such as
increase in the drive system cost and size, noise immunity and
reducing reliability. Therefore, the considerable sensorless
control strategies have been proposed to solve the problems.

1600

ICARCV 2008

The conventional solutions of non-salient-pole rotor

machines rely on the dependence of the rotor position from the
induced voltage in the stator windings [7]. These approaches
are quite simple and they use the motor voltage equations to
calculate the induced voltage from measured currents and
voltages. Unfortunately, due to the open loop calculation, the
estimation is very sensitive to system noises and parameters
inaccuracy. Thus, the inclusion of feedback capability in the
estimation strategy surely represents an improvement. State
observers as Luenberger [8], and Sliding Mode [9], extended
Kalman Filters [10] and Nonlinear Observer [11] are proposed,
in which the complete motor model is considered to estimate
both the speed and the rotor position. Among the different
approaches, the Sliding Mode Observer (SMO) represents an
attractive proposal because it is robust to measurement noise
and parameter deviations.

Figure 2.

This paper presents a sensorless speed vector control

system of the AGBM, in which the SMO is used to estimate
the induced voltage by using measured stator phase currents
and reference voltages and give rotor the rotational position
and speed. In order to confirm the proposed technique, some
experiments have been carried out.
II.

SENSORLESS CLOSED LOOP VECTOR CONTROL

3 Lsc0
 Lsl
2 g

where, K F

3 Lsc0
2 g

(2)

3Lsc0 / (4 g02 ) represents a force factor and

The rotor speed which is calculated by using state observer

is compared with reference speed, then the difference is input of
speed controller R, the output of speed controller is the q-axis
reference current, the q-axis reference currents for the two stator
windings are same with this current.

g: air gap length between stator and rotor

The voltage vector equation for one side stator in stationary
frame is as followed:
dis
 es
dt

(5)

The axial displacement from the equilibrium point along the

z-axis, z, can be detected by the gap sensor. The detected axial
position is compared with the axial position command zref, the
error is inserted in the axial position controller Rz, the output of
axial position controller is d-axis reference current. Position
command zref is always set to zero to make sure the rotor is right
in the midpoint between the two stators. The d-axis reference
currents for the two stator windings id1ref and id2ref can be
generated by using the offset current id0 subtracting and adding
idref respectively. The value of the offset current can be zero or a
small value around zero.

Lsl : leakage inductance

is Rs  Ls

2 KT i f iq

The motoring torque of AGBM can be controlled by q-axis

current (iq), while axial force can be controlled by d-axis current
(id), therefore the control scheme proposed for the sensorless
AGMB drive is shown in Fig. 2. It is based on the vector
control principle arranged in the d,q rotating frame, where daxis is aligned with the rotor flux vector.

with: Lsc0 : effective inductance per unit gap

Us

(4)

KT 3PLsc0 / (2 g 0 ) represents torque factor, if is equivalent to

the rotor current for permanent magnet, id0 is offset current and
g0 is the air gap at the equilibrium point.

(1)

and mutual induction: Lm

4 K F  id 0  i f  id

and motoring torque T is proportional to the quadrature axis

current (iq) as shown in (5).

A. Vector control scheme of AGBM

In this section, after a few preliminary remarks, the control
structure is derived and control methods of the axial force and
motoring torque of AGBM are repeated. Electrical model of
AGBM is based on equations of the general theory of
conventional AC electrical machines transformed to the
stationary frame (,) or field oriented frame (d,q). But, it is
very important to choose suitable transformation constants to
keep the invariance principle of powers. It has been chosen as
2 / 3 instead of 2/3 as in conventional three phase motor.
Furthermore, the self inductance of stator depends on air gap
between stator and rotor, so the stator phase inductance can be
approximated by:
Ls

Structure of PM type axial gap self bearing motor.

(3)

here, is is stator current vector, Us is stator voltage vector and es

is induced voltage vector.
It is known in [6] that, the axial force F is proportional to
the direct axis current (id), as shown in (4):

The motor currents in the two-phase stator reference frame

, are calculated by the measurements of two actual phase
currents. Therefore, the d,q components are obtained using the
estimated rotor position from observer. The quadrature
components are controlled to the reference value which is given
by the speed controller, while the direct components are
controlled to the reference value which is given by the axial

1601

position controller. The outputs of the current controllers,

representing the voltage references, are afterward directed to
the motor using the Pulse Width Modulation technique
(PWM), once an inverse transformation from the rotating to
the three phase stator reference frame has been performed. All
controllers are standard PI regulators except axial position
controller (PID).

permanent magnets of the rotor in the stator phases, Ze is the

electrical rotor angular speed and T is the rotor position or
magnet flux vector position.
It is obvious that, mechanical speed changes more slowly
than other electrical variables, so it can be assumed that in
short period the speed is constant, we can get:

The rotor position observer is arranged in the two phase

fixed , reference frame. The rotor position is calculated by
means of the estimation of the induced voltage components in a
state observer, using the instantaneous values of the motor
phase currents and reference voltages.
B. Sliding mode observer
Sliding mode observer is proposed in Fig. 3, in which the
complete motor model is considered to estimate both the
speed and the rotor position. The sliding mode observer has
the same structure of the classical Luenberger observer for
linear system. However, the difference here is the use of sign
of the output error instead of its actual value. The difference
between measured and estimated current is used for getting
sign of error, which multiply with sliding controller gain, the
output of controller is correction factor , which is added to
the mathematical model and the process repeats every control
cycle until error of two kinds of currents is zero.
From (3), the mathematical model of the AGBM in the
two phases stator reference frame (,) can be expressed in the
following equations:
diD
dt
di
E
dt

Rs
1
1
iD  U D  eD
Ls
Ls
Ls
Rs
1
1
 iE  U E  eE
Ls
Ls
Ls


(6)

where U, U, i, i and e, e are the - components of the

stator voltage, current and induced voltage, respectively. For
an ideal sinusoidal PM type AGBM, the induced voltage
components in stationary frame can be described as:
eD

eE

OmZe sin T

OmZe cos T

deD
dt

deE
dt

(8)

Ze eD

From (7), rotor position and speed can be calculated as:

T
Ze
or Ze

arctan( eD / eE )
eD2  eE2

Om
dT
dt

(10a)
(10b)

From (3), the vector equation for sliding mode observer

can be obtained when the induced voltage components can be
considered as disturbances:
Us

di
is Rs  Ls s  e s  '
dt

(11)

or in (,) frame , obtaining the following:

di 1
RS
1
i  e +i
= U 
dt
L
L
L
S
S
S

di
R
1
1
= U  S i  e +i
dt LS
LS
LS

de
dt =e

de
dt =e

Figure 3.

(9)

The estimated velocity is calculated at a rate more slowly

than the rotor position estimation. The advantage of the
formula (10a) is its ability to obtain reasonable accurate speed
estimation. Its disadvantage is obvious that the result of
velocity computation will depart because of the effects of
permanent magnet flux linkages Om , which may be changed as
the temperature varies. The strength of formula (10b) is the
accuracy of speed estimation throughout the medium to high
speed range when rotor angle estimation is stable. The
drawback to this method is the noise which results from the
differentiation process. Especially, it will degrade the system
response in the transient state. However, in this paper, the
equation (10b) is used to calculate the rotor speed.

(7)

where Om is the amplitude of the flux induced by the

Ze eE

Structure of Sliding mode observer.

1602

(12)

 i =K1 sign (i  i )=K1 sign(V D )

i =K1 sign (i  i )=K1 sign(V E )

with
are correction factors.
e =K 2 sign (i  i )=K 2 sign(V D )
=K sign (i  i )=K sign(V )

2
E
e 2

As the result, we have:

For estimating Ke, we will assume that, the second order

derivative of difference between measured and estimated
currents is zero, therefore:

1 when V > 0

and sign-operator defined as: sign(V ) = 0 when V = 0

 1 when V < 0

and the symbol ^ indicates estimated variables, with reference

to the extended state format we have:
x Ax  Bu  K . sign( s )
(13)

y Cx
where K is a gain matrix, for the PM type AGBM, the gain
matrix can assume as the following:
Ki 0
(14)
0 K
e

according to the symmetry of the equations respect to the ,

components, it will be received:
0
0
K
K
Ki 1
and K e 2
(15)

0 K1
0 K2
K

Accordingly with the sliding mode theory [14], the sliding

surface will be defined upon the stator current error as:
s

V D

V E

(16)

When the estimated error trajectories reach the sliding

surface, i.e., s 0 , it is clear that the observed currents will
converge to the actual ones. The choice of gain factors must
satisfy the necessary conditions for sliding mode convergence.
Let's design the Lyapunov function to find such condition of
sliding mode existence:
V

1 T
s s
2

1 2
(V D  V E2 )
2

(17)

For designing Ki, The stator current error is obtained by

subtracting (11) from (3), we get:
RS
1

 
VD iD  iD - L V D  L ( eD  eD )  ziD

S
S
(18)

R
1

S
V
  i - V  ( e  e )  z
i
iE
E
E
E
E E E
LS
LS

de s des
di di
= -Rs ( s - s ) = -Rs  s
dt dt
dt dt

sT  s

0

in order to assure that system is convergent

Further, the choice of the gain factor also results from the
compromise of assuring the sliding mode operation all over
the operating range (large gains) with the minimum ripple
(small gains). In fact, the large switching gain causes a great
amount of ripple on the observed variables, which become
unusable for control. The solution proposed in this paper
consists in adaptive gains proportional to the motor frequency,
i.e. speed:
K1

sT  s

R
1
 S (V D2  V E2 ) 
( eD  eD )V D
LS
LS

1

(eE  eE )V E  V D ziD  V E ziE  0
LS

K10Ze

K2

K 20Ze

(24)

esf ( n ) esf ( n  1)  T 2S f c ( es ( n )  esf ( n ))

(25)
where es presents the unfiltered signal, which is output of the
sliding controller, and esf is the filtered signal. Constant
T

1 / f PWM presents the PWM frequency.

Filter design must be accurately considered, the value of

the cutoff frequency fc depends on the selection of the sliding
mode controller gain and is set experimentally. In this paper,
digital filter is used for both the back-EMF components.
In fact, due to the low-pass characteristic, the filtered
signals are reduced in amplitude and phase delayed respect to
the inputs. In order to avoid the angular error which would
arise because of using the filtered back-EMF signals, a proper
compensation mechanism must also be implemented. The
phase can be corrected by using equation:
esf .e jG

(26)

or in , frame we can get:

eDsm

eEsm

(20)

and

Unfortunately, the unavoidable ripple, which is caused by

axial position controller and sliding mode controller, affecting
the sliding mode signals prevents the implementation of the
vector control. Low pass filters should be used in order to take
off the noise and get the actual signal. The equation of filter is
obtained by the discretization of a continuous one-pole filter
using a ZOH (Zero Order Hold) as:

By substituting (16) and (18) into (19) we can get:

V

(23)

K e   max( eD , eE ) / Rs

esm

(19)

(22)

By the same way with above, the value of Ke must satisfy:

Then, the sliding mode exists when:

V

(21)

Ki ! max( eD , eE )

eDf cos G  eEf sin G

eDf sin G  eEf cos G

(27)

From a theoretical point of view, the compensating angle

G can be obtained directly by the phase/frequency response

of the adopted filter. In fact, a further correction is needed, due

1603

to the delay between the sliding mode estimates and the actual
variables.
The compensation value depends on motor speed, so the
value of rotor speed is divided into thirty ranges from zero
value to norm value, each speed range has its own slope and
constant phase compensation component. The compensating
angle is fetched experimentally.
III.

Inverter
board

IMLEMENTATION AND RESULTS

Radial
bearing

A. Control hardware
In order to confirm the proposed control method for PM
type AGBM, an experimental setup was set up which is shown
schematically in Fig. 4. The rotor disc, as shown in Fig. 5, has
a diameter of 55 mm, and four neodymium iron magnets with
the thickness of 0.8mm for each side are attached to its surface
to create two pole pairs. For experimental simplicity, the rotor
is supported by two radial ball bearings in order to restrict the
radial motion.
The stator, as shown in Fig. 6, has a diameter of 55 mm and
six concentrated wound poles each with 200 coil turns. The
stators can slide on linear guide to ensure the same desired air
gap between rotor and two stators. A DC generator (Sanyo
T402) is installed to give the load torque which is adjusted by
changing the resistance, which is connected to outputs of
generator. In order to measure the rotor speed and the axial
position, a rotary encoder (Copal RE30D) and an eddy-currenttype displacement sensor (Sinkawa Co. Ltd. VC-202N) are
installed, respectively.
The control hardware of sensorless AGBM drive is shown
in Fig. 7. It is based on a dSpace1104 board dedicated to
control of electrical drives, which includes PWM units, general
purpose input/output units (8 ADC and 8 DAC) and encoder
interface. The DSP reads the displacement signal from the
displacement sensor via an A/D converter, and the rotor angle

Figure 4.

Schematic of the experiment setup.

Figure 5.
Figure 6.

Stator of AGBM

Rotor of AGBM.

DC
Generator

Stator

Rotor

Figure 7.

Linear
guide

Displacement
sensor

Overview of control hardware of AGBM.

position and speed from the encoder via an encoder interface.

Two motor phase currents are sensed, rescaled, and converted
to digital values via an A/D converter. Then, the dSpace1104
calculates reference currents using the rotation control and
axial position control algorithms and send its commands to
three phase inverter board. The AGBM is supplied by two
three phase PWM inverters with a switching frequency of 20
kHz.
B. Experimental and Results
First, parameter estimation is carried out. The phase
resistor is 2.6. The static phase inductances are measured by
measuring frequency response of the voltage versus current of
c 11 u 106 Hm and
the stator coil. We have La0
Lal

5 u 103 H . Then, the inductances of (1) are obtained by

calculating, we have Ls 17 u 103 H . The air gap between the

stator and the rotor is adjusted to 1.5 mm, including the
magnet. Amplitude of the flux induced by the permanent
magnets of the rotor in the stator phases can be estimated by
measuring no load-voltage of AGBM in generator mode,
which is m 0.022Wb . Rotor mass is 0.28 kg and rotor
inertia is 0.000106 kgm2.
As the induced voltages amplitude is proportional to the
speed, moreover, at standstill the position information is not
available at all, leading to an uncontrolled movement of the
rotor at start-up. Thus, the proposed observer does not allow
the speed and position detection at very low speed and
standstill. So in this paper, first, an encoder start-up procedure
is implemented, in order to accelerate the motor up to a speed
value where the estimates are sufficiently reliable to guarantee
the proper operation of the observer, then observer will
replace the encoder.
Fig. 8 presents the estimated back-EMF components by
SMO after using low pass filter with cutoff frequency 101 Hz
at 1500 rpm. Its sinusoidal shape shows good estimation, and
can support theoretically for the proposed estimated method.
Fig.9 shows the axial displacement of rotor, first, when
the displacement is set to 0.25 mm, at time of 0.115 s, the

1604

axial position controller starts to work. In transient state, the

maximum error is 0.1 mm, that is considerable smaller than air
gap length at the equilibrium point (g0 =1.5mm) and settling
time is about 0.01 s. After that, the displacement is almost

Figure 8. Estimated back-EMF components.

zero in steady state.

Fig.10 indicates the speed error between measured and
estimated speed at steady state of 500 rpm, the maximum error
is about 2 rpm, that is about 0.4%, it is acceptable for speed
control. Fig.11 and Fig.12 presents the estimated and
measured rotor position and the error between them at the
steady state with rotating speed is 500 rpm, the maximum
error is about 0.5 rad.
IV. CONCLUSION
Sensorless speed control method for the AGBM is
proposed and developed. Results from experiment confirm the
good performance of the closed loop speed control. The axial
position and rotating speed can be controlled independently.
The performance of AGBM sensorless drives is good at high
speed range (over 10% norm speed) and in steady state. It is
suitable for applications where low speed and standstill
operations are not required. The fist problem in experiment is
represented by the ripple, which is caused by axial position
controller, affecting the estimated variables. The second
problem is that the proposed observer can not work in zero and
low speed. Further work is being continued to improve the
experiment for this proposal.
REFERENCES
[1]

Figure 9. Axial displacement.

Figure 10. Speed error at 500 rpm.

Figure 11.

Measure and estimated rotor position (rad) at 500 rpm.

Figure 12.

A. Chiba, T. Deido, T. Fukao and M. A. Rahman, An analysis of

bearingless AC motors, IEEE Trans. Energy Conversion, vol. 9, pp. 6167, Mar. 1994.
[2] Y. Okada, K. Dejima and T. Ohishi, Analysis and comparison of PM
synchronous motor and induction motor type magnetic bearing, IEEE
Trans. Industry Applications, vol. 32, pp. 1047-1053, Sept./Oct. 1995.
[3] Y. Okada, S. Ueno, T. Ohishi, T. Yamane and C. C. Tan, Magnetically
levitated motor for rotary blood pumps, Artif. Organs, vol. 21, no. 7,
pp. 739-745, 1997.
[4] Y. Okada, T. Masuzawa, K. Matsuda, K. Ohmori, T. Yamane, Y.
Konishi, S. Fukahori, S. Ueno and S. J. Kim, Axial type self bearing
motor for axial flow blood pump, International Society for Artificial
Organs vol. 27, pp. 887-891, 2003.
[5] S. Ueno and Y. Okada, Vector control of an induction type axial gap
combined motor-bearing, in Proc. of the IEEE Int. Conf. on Advanced
Intelligent Mechatronics, Sept. 19-23, 1999, Atlanta, USA, pp. 794-799.
[6] S. Ueno and Y. Okada, Characteristics and control of a bidirectional
axial gap combined motor-bearing, IEEE Transactions on
Mechatronics, Vol. 5, No. 3, Sept. 2000, pp. 310-318.
[7] R. Wu, G.R. Slemon "A Permanent Magnet Motor Drive without a Shaft
Sensor", IEEE Trans. Ind. Applications, vol. 27, no. 5, Sept./Oct. 1991,
pp. 1005-1011.
[8] Dich Nguyen Quang and Satoshi Ueno, Sensorless speed control of a
permanent magnet type axial gap self bearing motor, in Proc. of the 11th
Int. Symposium on Magnetic Bearings, August 2008, Nara, Japan,
pp.108-113.
[9] F. Parasiliti, R. Petrella and M. Tursini, Sensorless speed control of a
PM synchronous motor by sliding mode observer, Proc. ISIE 97, Vol.
3, pp. 1106-1111, Guimares, 1997.
[10] A. Germano, F. Parasiliti and M. Tursini, Sensorless speed control of a
PM synchronous motor by Kalman filter, Proc. of ICEM , Vol. 2, pp.
540-544, Paris, September 1994.
[11] Cristian De Angelo, Guillermo Bossio, Jorge Solsona, Guillermo O.
Garca, and Mara Ins Valla, A Rotor Position and Speed Observer for
Permanent-Magnet Motors With Nonsinusoidal EMF Waveform, IEEE
Trans. Industrial Electronics, vol. 52, no. 3, pp. 807-813, June 2005.
[12] A. Chiba, et. al., Magnetic Bearings and Bearingless Drives, 1st edition,
Elservier, Great Britain, 2005.

Position error (rad) at 500 rpm.

1605