1836 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 49, NO.

4, JULY/AUGUST 2013
Intelligent Speed Control of Interior Permanent
Magnet Motor Drives Using a Single
Untrained Articial Neuron
Casey B. Butt and M. Azizur Rahman, Life Fellow, IEEE
AbstractThis paper presents an intelligent speed controller for
the interior permanent magnet synchronous motor based on a sin-
gle articial neuron. Traditional articial neural network-based
motor controllers require extensive ofine training, which is both
time consuming and requires extensive knowledge of motor be-
havior for the specic drive system. In addition, drive behavior
is unpredictable when parameters outside the training set are en-
countered. The proposed drive system overcomes these limitations
by requiring no ofine training, is robust under varying operating
parameters, and is easily adaptable to various drive systems. Drive
efcacy is veried in simulation as well as experimentally.
Index TermsArticial neural network (ANN), intelligent
control, interior permanent magnet motor, interior permanent
magnet synchronous motor (IPMSM), motor control.
I. INTRODUCTION
T
HE INTERIOR permanent magnet synchronous motor
(IPMSM) in speed-controlled high performance drive ap-
plications is required to quickly achieve command speed and
maintain that operating point with maximum accuracy and
minimal perturbation despite the occurrence of sudden and un-
known disturbances. Conventionally, proportional-integral (PI)
and proportional-integral-derivative (PID) speed controllers
have been utilized to meet these control challenges. However,
the operation of the IPMSM is strongly affected by rotor
magnetic saliency, saturation, and armature reaction effects
[1], making the dq-axis reactance parameters change over the
operating range and rendering conventional xed-gain PI and
PID controllers very sensitive to parameter variations [2][4].
Researchers have used controllers based on fuzzy logic,
sliding mode, traditional articial neural network (ANN), and
genetic algorithm to address these issues [8][13]. Each of
these controllers, however, has its own performance and/or
implementation shortcomings.
This paper presents an ANN-based controller for the IPMSM
drive which precisely and accurately follows command speeds,
with insensitivity to load and parameter variations, yet requires
Manuscript received August 23, 2011; revised March 2, 2012, June 1, 2012,
and August 16, 2012; accepted October 18, 2012. Date of publication April 12,
2013; date of current version July 15, 2013. Paper 2011-IACC-465.R3, pre-
sented at the 2011 IEEE International Electric Machines and Drives Confer-
ence, Niagara Falls, ON, Canada, May 1518, and approved for publication
in the IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS by the Industrial
Automation and Control Committee of the IEEE Industry Applications Society.
The authors are with the Faculty of Engineering and Applied Science,
Memorial University of Newfoundland, St. Johns, NL A1B 3X5, Canada
(e-mail: casey.butt@gmail.com; arahman@mun.ca).
Digital Object Identier 10.1109/TIA.2013.2257973
minimal ofine training. The system is simplied to a single
articial neuron (SAN) to minimize complexity and computa-
tional capacity requirements. Speed error and comparison of the
SANs output command torque to a calculated reference torque
are used conditionally at each iteration to adaptively modify the
SAN parameters to minimize speed error and ensure stability.
An approximation of the maximum torque per ampere (MTPA)
mode of operation is incorporated, resulting in increased ef-
ciency up to base speed.
The proposed IPMSM drive system has been implemented
using a dSPACE digital signal processor (DSP) controller board
with a 1-hp IPMSM. The indirect vector control scheme of the
IPMSM drive has been employed [14]. Experimental results are
provided to conrm the efcacy of the drive system.
II. IPMSM DYNAMICS
Adopting the conventional practice of neglecting iron losses,
the mathematical model of an IPMSM drive can be described
by the following equations in a synchronously rotating rotor
dq reference frame as:

v
q
v
d

=r
s

i
q
i
d

pL
q
P
r
L
d
P
r
L
q
pL
d

i
q
i
d

P
r

f
0

(1)
T
e
=T
L
+ B
m

r
+ J
m
p
r
(2)
T
e
=
3P
2
[
f
i
q
+ (L
d
L
q
)i
d
i
q
] (3)
where
v
d
, v
q
d- and q-axis stator voltages;
i
d
, i
q
d- and q-axis stator currents;
L
d
, L
q
d- and q-axis stator inductances;
r
s
stator per-phase resistance;

r
rotor speed in angular frequency;
T
e
, T
L
electromagnetic and load torques;
J
m
moment of inertia of the motor and load;
B
m
friction coefcient of the motor;
P number of poles of the motor;
p differential operator (= d/dt);

f
rotor magnetic ux linking the stator.
III. CURRENT CONTROL ALGORITHM
The torque equation, (3), indicates that there exists a non-
linear relationship between the electrical torque and d- and
0093-9994/$31.00 2013 IEEE
BUTT AND RAHMAN: INTELLIGENT SPEED CONTROL OF INTERIOR PERMANENT MAGNET MOTOR DRIVES 1837
q-axis currents. In order to incorporate this nonlinearity in a
practical IPMSM drive, an approximation of the control tech-
nique known as the MTPA mode is derived. MTPA provides
the maximum motor torque with the minimum possible stator
current [15][19].
A. MTPA Mode
The maximum torque per unit current can be achieved by
differentiating (3) with respect to q-axis current i
q
and setting
the resulting equation to zero, which gives
i
d
=

f
2(L
q
L
d
)

2
f
4 (L
q
L
d
)
2
+ i
2
q
. (4)
Substituting (4) into (3), one can get a nonlinear relationship
between i
q
and T
e
as
T
e
=
3P
2

f
i
q

f
i
q
2
(L
d
L
q
)

2
f
i
2
q
4(L
q
L
d
)
2
+i
4
q

.
(5)
In real time, the implementation of the drive system becomes
potentially undened and computationally burdensome with
expressions (4) and (5). To address this, the d- and q-axis
currents are obtained by expanding the square root term of (4)
via a Taylor series expansion about zero, giving
i
d
= 0.11825i
2
q
. (6)
Numerical values of (6) are obtained by using the parameters
of Motor A found in the Appendix. By substituting (6) into (3),
the following relationship can be obtained:
i
q
= 1.06157T
e
. (7)
As this derivation involves a series approximation and the
assumption that the involved motor parameters remain constant,
(6) and (7) are used to achieve approximate MTPA control of
the IPMSM [20].
The block diagram in Fig. 1 shows the control scheme of the
motor drive. The command torque T

e
is obtained from a SAN-
based speed controller. Then, using (7), the reference q-axis
current i

q
is computed from T

e
. Subsequently, the reference
d-axis current i

d
is calculated using (6). The corresponding
three-phase currents are determined by vector rotation. A hys-
teresis current controller compares the reference three-phase
currents with actual currents and generates gate signals for the
transistorized inverter.
IV. DESIGN OF SAN CONTROLLER FOR IPMSM
A. SAN Control Principle
The dynamic model of the IPMSM can be rewritten from (1)
and (2) as
L
q
pi
q
+ P
r
L
d
i
d
=v
q
r
s
i
q
P
r

f
(8)
p
r
=(T
e
T
L
B
m

r
)/J
m
. (9)
Fig. 1. Block diagram of the SAN-based IPMSM drive.
The load will be considered as having unknown nonlinear
mechanical characteristics and can be modeled using the fol-
lowing equation as [2]:
T
L
= A
2
r
+ B
r
+ C (10)
where A, B, and C are arbitrary constants. To make the control
task easier, the equations of the IPMSM are expressed as a
single-input/single-output system by combining (9) and (10) in
continuous time domain form as
J
m
d
r
dt
= T
e
(B
m
+ B)
r
A
2
r
C. (11)
Asmall increment T
e
inT
e
causes a small increment
r
in
r
J
m
d(
r
+
r
)
dt
= (T
e
+ T
e
)
(B
m
+ B)(
r
+
r
) A(
r
+
r
)
2
C. (12)
Subtracting (11) from (12) gives
J
m
d(
r
)
dt
= T
e
(B
m
+ B + 2A
r
)(
r
) A(
r
)
2
.
(13)
By replacing all the continuous quantities of (13) by their
nite differences, the discrete-time small signal model of the
simplied IPMSM with nonlinear load can be given as
T
e
(n) =
J
m
t
s
e(n)
+(B
m
+ B + 2A
r
(n))
r
(n) + A{
r
(n)}
2
. (14)
Hence,
T

e
(n) =

discrete
T
e
(n) = f (e(n),
r
(n),

r
(n)) (15)
where e(n) =
r
(n)
r
(n 1) is the change of speed
error,
r
(n) = e(n) =

r
(n)
r
(n) is the present sample
of speed error,
r
(n 1) is the past sample of speed error,

r
(n) is the present sample of actual speed,

r
(n) is the present
sample of command speed, t
s
is the sampling time interval, and
f denotes the nonlinear function. Equation (15) indicates that,
in order for the SAN-based speed controller to produce accurate
1838 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 49, NO. 4, JULY/AUGUST 2013
torque commands under nonlinear loading, its inputs need be
the speed command

r
(n), speed error
r
(n), and change of
speed error e(n).
Thus, the purpose of using the SAN speed controller is to
obtain the command torque by mapping the nonlinear func-
tional relationship between the command electrical torque T

e
and the rotor command speed

r
, speed error e(n), and change
in speed error e(n), with no reliance on knowledge of motor
parameters.
B. Reference Command Torque
Equation (2) can be rearranged and discretized to give
T
L
(n) = T
e
(n) J
m
p
r
(n) B
m

r
(n). (16)
J
m
and B
m
are known;
r
(n), the present sample of rotor
speed, can be obtained from a position sensor mounted to
the rotor shaft; and p
r
(n) = (
r
(n)
r
(n 1))/t
s
is the
change in rotor speed over the sampling period. This means
that load torque T
L
(n), at the present sample, can be calculated
directly from (16) if T
e
(n) is known at that sampling instant.
T
e
(n) can be calculated from (3) as
T
e
(n) =
3P
2
[
f
i
q
(n) + (L
d
L
q
)i
d
(n)i
q
(n)] . (17)
i
r
q
(n) and i
r
d
(n) can be calculated via Parks equation and
reference frame transformation from samples of i
a
(n), i
b
(n),
and i
c
(n) taken by current sensors. P and
f
are known, and
L
d
and L
q
must be assumed to have constant known values.
With the current sample value of T
L
now having been cal-
culated, (2) can be utilized to calculate the reference command
torque as follows:
T

ref
(n) = J
m
p
r
(n) + B
m

r
(n) + T
L
(n) (18)
where p
r
= (

r
(n)
r
(n))/t
s
is the current sample value
of speed error divided by the sampling time,

r
(n) is the current
sample of command speed, and
r
(n) is the current sample of
rotor speed.
This value of command torque is calculated with the as-
sumption that L
d
and L
q
have constant known values. That
assumption is, of course, incorrect. Using the actual inductance
curves of our test motor indicates that this T
ref
command
may vary from the correct torque value by up to 42% at the
most extreme cases. This command torque serves merely as a
reference for the real-time training of the SANand to set bounds
to assure drive stability, however, and for these purposes, it is
sufcient.
C. SAN-Based Speed Controller for IPMSM Drive
To minimize computational burden, a single feedforward
neuron structure is employed, with the command torque being
the only output of the SANcontroller. The inputs to the SANare
the present sample of command speed

r
(n), present sample of
speed error e(n) =

r
(n)
r
(n), and present sample of the
change of speed error e(n) =
r
(n)
r
(n 1). Neu-
ron output ranges from -1 to 1 via a tan-sigmoid transfer func-
tion, so this value is multiplied by the maximum peak torque
Fig. 2. SAN for IPMSM drive.
Fig. 3. Simulated response of SAN controller-based IPMSM drive to
188.5-rad/s speed command at no external load.
rating of the motor to obtain the actual torque command. Due
to space limitations, the equations of the ANN structure are not
included herethey are based directly on conventional feedfor-
ward ANN methods. The SAN structure is illustrated in Fig. 2.
From this command torque, the MTPA approximation is used
to obtain the appropriate q- and d-axis command currents, i

q
and i

d
, respectively, to produce the desired motor speed. Phase
command currents, i

a
, i

b
, and i

c
, are then calculated via the
inverse Park transform and applied to the motor through the
voltage source inverter via a hysteresis controller.
In order to provide adaptive control, the weights and bias
of the SAN are updated, by back propagation, whenever the
speed error exceeds an appropriately chosen threshold. If this
threshold is not exceeded, the weights and bias will be used
again in the following iteration. The speed error itself, e(n) =

r
(n)
r
(n), is used for back-propagation updating.
In addition, at each sample, torque error e
T
(n) = T

ref
(n)
T

SAN
(n) is used for SAN retraining if e
T
(n) exceeds a tol-
erance value. If retraining is not successful within a specied
number of back-propagation iterations (to x a maximum limit
on the computing time that retraining can take), then T

ref
(n) is
used to generate motor currents at that sample. Bounding T

SAN
by comparison to an approximate torque command ensures
that the SAN does not cause the drive to become unstable,
and specifying the back-propagation iteration count allows the
control algorithm to be adjusted for the capabilities of the
processing power at hand. In addition, this also allows the SAN
to produce its own initial weights and bias, as well as adjust
those weights and bias on a per-sample basis, meaning that no
ofine training is required.
BUTT AND RAHMAN: INTELLIGENT SPEED CONTROL OF INTERIOR PERMANENT MAGNET MOTOR DRIVES 1839
Fig. 4. Simulated response of SAN drive, with T

ref
disabled, to 188.5-rad/s
start-up speed command and sudden application of full load (0 2 Nm)
at 2 s.
As an example of the training process, Fig. 3 shows the
simulated speed response of the SAN drive under no external
load to a rated command speed of 188.5 rad/s. The speed error
tolerance was set at 0.1 rad/s, the torque error tolerance was
set to 42% as explained earlier, and ten torque error retraining
iterations were allowed at each sample. Under these conditions,
the weights and bias of the SAN settled to W
1
= 0.6286,
W
2
= 1.5458, W
3
= 1.0276, and B = 0.4652. At the nal
sample, the command torque value was T

SAN
= 0.5098 N m.
The learning rates of the SAN were tuned manually so that the
SAN could train itself within the 42% torque error boundary for
all but four of the samples, where T

ref
was used as command
torque. As the sampling rate was 5 kHz, this indicates that the
SAN successfully provided the torque command for 4996 of the
5000 total torque commands issued.
To illustrate the contribution of the reference torque com-
mand T

ref
as a boundary for the SAN, Fig. 4 shows the
simulated response of the drive at start-up and under full load
(2 N m) application with T

ref
disabled. The drive does reach
and maintain command speed based on speed error training
alone, but with signicant oscillation and load sensitivity.
Each iteration of the combined speed- and T
ref
-based
training of the SAN-based controller imposes 50 mult./div.,
37 add./sub., and 16 exponential operations. This is a signicant
increase in computational burden as compared to a single
iteration of a conventional PID controller but only occurs when
retraining is necessary. During trained operation, a total of 17
operations are required by the SAN. In real-world implemen-
tation on the laboratory equipment, the SAN-based drive was
found to perform best at a maximum sampling rate of 5 kHz,
whereas a typical PID-based controller could be sampled at
10 kHz. A more complex four-neuron-based ANN controller
required the sampling frequency to be lowered to 1 kHz and
did not produce acceptable real-world results due to this.
V. LABORATORY IMPLEMENTATION
The complete drive system has been implemented in the
laboratory with a 1-hp IPMSM using a dSPACE DS1102 DSP
controller board. The actual motor currents are measured by
Hall-effect sensors and fed to the DSP board through an A/D
Fig. 5. Simulated response of SAN-based IPMSM drive to sudden application
of load (0 2 N m) at a command speed of 188.5 rad/s. (a) Speed,
(b) a phase current, (c) q-axis current, and (d) d-axis current.
1840 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 49, NO. 4, JULY/AUGUST 2013
Fig. 6. Simulated response of PID-based IPMSM drive to sudden application
of load (0 2 N m) at a command speed of 188.5 rad/s. (a) Speed,
(b) a phase current, (c) q-axis current, and (d) d-axis current.
Fig. 7. (a) Simulated SAN controller with sudden parameter change (R
s

2R
s
and L
d
0.5L
d
) under nonlinear load. (b) Simulated start-up responses
of SAN drive to rated speed command (188.5 rad/s) with inertia changes.
Dashed line = J
m
, and solid line = 1.5J
m
.
converter. As the motor neutral is isolated, only two phase
currents are fed back, and the other phase current is calculated
from them. Three-phase reference currents are generated uti-
lizing the reference q- and d-axis currents and rotor position
angle obtained through an encoder mounted on the shaft of the
motor. Computed three-phase reference currents are converted
to upper and lower hysteresis limits by adding and subtract-
ing a preselected percentage band. Actual motor currents are
compared to these hysteresis current limits, and pulse-width
modulated gate-drive signals are generated accordingly.
All computations involving the SAN, the generation of ref-
erence currents and, consequently, gate-drive signals for the
inverter, are done in the ANSI C programming language via
a PC. The program is compiled using the Texas Instruments C
compiler and downloaded to the DS1102 controller board via
dSPACE software. The sampling frequency is set at 5 kHz.
VI. RESULTS AND DISCUSSION
Prior to implementation, the complete SAN-based drive sys-
tem, including provisions for iron losses in the motor, was sim-
ulated using Matlab/Simulink software. A PID-based controller
was tuned and implemented to serve as a baseline comparison.
PID tuning was roughed out via the ZieglerNichols method
BUTT AND RAHMAN: INTELLIGENT SPEED CONTROL OF INTERIOR PERMANENT MAGNET MOTOR DRIVES 1841
Fig. 8. (a) Simulated PID controller with sudden parameter change (R
s

2R
s
and L
d
0.5L
d
) under nonlinear load. (b) Simulated start-up responses
of PID drive to rated speed command (188.5 rad/s) with inertia changes.
Dashed line = J
m
, and solid line = 1.5J
m
.
and then adjusted manually for best performance. K
p
, K
i
, and
K
d
were tuned to balance the overshoot and loading response
with the settling time, and the back calculation method was
implemented as an antiwindup strategy. Simulated and exper-
imental results from both drives are presented in this section.
Fig. 5 shows the simulated speed, a phase, and q- and
d-axis current responses of the SAN-based IPMSM drive at
start-up and to a step increase in load from 0 to 2 N m for a
rated command speed of 188.5 rad/s. The speed disturbance due
to the load application is almost indiscernible at this resolution.
The increase in phase current draw due to the sudden torque
requirement increase is obvious. The q-axis current increases
with load to produce the required additional torque, and the
d-axis current increases slightly in the negative direction to take
advantage of the reluctance component of the torque equation,
as per the MTPA scheme.
Fig. 6 shows the simulated speed, a phase, and q- and
d-axis current responses of the PID controller-based IPMSM
drive at start-up and to a step increase in load from 0 to 2 N m
for a rated command speed of 188.5 rad/s. Unlike the SAN-
based controller case, the speed disturbance due to the load
application is more dramatic. The increase in current draw due
to the sudden torque requirement increase is similarly obvious,
although it ramps more gradually as compared to the SAN case.
Fig. 9. Experimental response of SAN-based IPMSMdrive to sudden changes
of command speed at light load (90 188.5 130 rad/). (a) Speed and
(b) a phase current.
The q-axis current increases with load to produce the required
additional torque, and the d-axis current increases slightly in the
negative direction, as per the MTPA scheme. These responses
show slower reaction times than the SAN case.
Fig. 7(a) and (b) shows the simulated SAN-based drives
speed responses to sudden parameter changes (R
s
to 2R
s
and
L
d
to 0.5L
d
) under nonlinear loading (T
L
=
2
/35532.25)
and to inertia changes under no load at start-up with rated
speed command (J
m
increased to 1.5J
m
). The SAN-based
drives speed response appears unaffected by these R
s
and L
d
variations under nonlinear load. The inertia increase reduces
the acceleration of the motor at start-up but has little effect
otherwise.
Fig. 8(a) and (b) shows the simulated PID-based drives
speed responses to sudden parameter changes (R
s
to 2R
s
and
L
d
to 0.5L
d
) under nonlinear loading (T
L
=
2
/35532.25) and
to inertia changes under no load at start-up with rated speed
command (J
m
increased to 1.5J
m
). The PID-based drives
speed response shows some sensitivity to these R
s
and L
d
variations under nonlinear load, although the command speed
is maintained. The inertia increase reduces the acceleration of
the motor at start-up and also increases overshoot.
Fig. 9(a) and (b) shows the experimental speed and current
responses of the SAN-based IPMSM drive to step changes
of the command speed at light load. The drive reaches and
maintains command speeds quickly and accurately during the
step increase and step decrease of the command speed. The
phase current magnitude does not appear to change signicantly
with motor speed (as load torque is constant). The current
frequency, however, increases and decreases appropriately with
changing command/motor speed.
1842 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 49, NO. 4, JULY/AUGUST 2013
Fig. 10. Experimental response of SAN-based IPMSM drive to sudden ap-
plication of full load (0 2 N m) at rated speed (188.5 rad/s). (a) Speed,
(b) a phase current, (c) q-axis current, and (d) d-axis current.
Fig. 10 shows the experimental speed, a phase, and q- and
d-axis current responses of the SAN drive to sudden application
of full load (0 2 N m) at rated speed (188.5 rad/s). Although
speed disturbance due to loading is not visible at this resolution,
the increased phase current amplitude with the application of
load is apparent. Likewise, q- and d-axis currents respond,
particularly i
q
, with increases in magnitude to meet the addi-
tional torque demand.
Fig. 11. Experimental start-up responses of SAN drive to rated speed com-
mand (235.6 rad/s) with inertia changes.
Fig. 12. Experimental response of PID-based IPMSM drive to (a) start-up
speed command of rated speed (188.5 rad/s) and (b) sudden application of full
load at rated speed.
Fig. 11 shows the start-up response of the SAN drive to rotor
inertia changes. Because the original laboratory motor did not
practically allow inertial adjustments, an alternate test motor
was used for these results (see Motor B in the Appendix).
The increased inertia decreases the drives acceleration to base
speed (235.6 rad/s) but has little visible effect otherwise. The
MTPA equations (6) and (7) and T
ref
were adjusted in accor-
dance with Motor B, but the SANs learning parameters were
not deliberately retuned to accommodate this motor.
For comparison and conrmation of simulation results,
Fig. 12 shows the experimental speed responses of the PID-
based drive (Motor A) to (a) start-up speed command of rated
speed (188.5 rad/s) and (b) sudden application of full load
at rated speed. As in the simulation, the PID-based drive
accurately achieves command speed with moderate overshoot
but characteristically experiences temporary disturbance when
presented with sudden load incursion.
BUTT AND RAHMAN: INTELLIGENT SPEED CONTROL OF INTERIOR PERMANENT MAGNET MOTOR DRIVES 1843
VII. CONCLUSION
These results indicate that the SAN-based controller with
real-time training offers excellent speed response and load han-
dling for high-performance IPMSMdrives. Ahigh performance
standard is achieved despite a minimum level of complexity
and computational burden as compared to other intelligence-
based adaptive controllers. A new approach to the use of
articial intelligence for motor drives has been presented, in
which the often cumbersome task of ofine training has been
sidestepped by enabling the drive to train itself in real time
during operation.
The SAN-based real-time control strategy can be a good
choice to provide adaptive control of the IPMSM drive without
necessitating high processor power or extensive ofine training.
On the very modest PC-based system used for experimental
work (dSPACE ds1102 controller board in a PII 300-MHz
PC with 128-MB RAM), the SAN-based controller achieved
a sampling frequency of 5 kHz. With greater computing power,
better accuracy and precision of speed and current responses
would be expected.
APPENDIX
MOTOR PARAMETERS
Motor A: 3 , 1 hp, 208 V, 60 Hz, P = 2, L
d
= 0.04244 H,
L
q
= 0.07957 H, r
s
= 1.93 , J
m
= 0.003 kg m
2
,
B
m
= 0.0008 N m/rad/s, and
f
= 0.314 V/rad/s.
Motor B: 3 , 2 hp, 245 V, 112.5 Hz, P = 6, L
d
=
0.005432 H, L
q
= 0.008582 H, r
s
= 2.90 , J
m
=
0.000449 kg m
2
, B
m
= 0.000119 N m/rad/s, and

f
= 0.042935 V/rad/s.
REFERENCES
[1] M. A. Rahman and P. Zhou, Field circuit analysis of brushless permanent
magnet synchronous motors, IEEE Trans. Ind. Electron., vol. 43, no. 2,
pp. 256267, Apr. 1996.
[2] M. A. Rahman and M. A. Hoque, On-line self-tuning ANN based speed
control of a PMDC motor, IEEE/ASME Trans. Mechatronics, vol. 2,
no. 3, pp. 169178, Sep. 1997.
[3] M. R. Emami, I. B. Turksen, and A. A. Goldenburg, Development of
a systematic methodology of fuzzy logic modeling, IEEE Trans. Fuzzy
Syst., vol. 6, no. 3, pp. 346361, Aug. 1998.
[4] S. Y. Yi and M. J. Chung, Robustness of fuzzy logic control for an uncer-
tain dynamic system, IEEE Trans. Fuzzy Syst., vol. 6, no. 2, pp. 216225,
May 1998.
[5] K. Erenay, I. Ciprut, L. Tezduyar, and Y. Istefanopulos, Application of
fuzzy algorithms to the speed control of washing machines with brushless
dc motors, in Proc. Int. Conf. Elect. Mach., Istanbul, Turkey, 1998,
pp. 12311236.
[6] C. B. Butt, M. A. Hoque, and M. A. Rahman, Simplied fuzzy logic
based MTPA speed control of IPMSM drive, IEEE Trans. Ind. Appl.,
vol. 40, no. 6, pp. 15291535, Nov./Dec. 2004.
[7] C. Namudri and P. C. Sen, A servo-control system using a self controlled
synchronous motor (SCSM) with sliding mode controller, IEEE Trans.
Ind. Appl., vol. IA-23, no. 2, pp. 283295, Mar. 1987.
[8] A. Consoli and C. Antonio, A DSP based sliding mode eld oriented
control of an interior permanent magnet synchronous motor drive, in
IPEC Conf. Rec., Tokyo, Japan, 1990, pp. 296303.
[9] M. A. El-Sarkawi, A. A. El-Samahy, and M. L. El-Syed, High perfor-
mance drive of dc brushless motors using neural network, IEEE Trans.
Energy Convers., vol. 9, no. 2, pp. 317322, Jun. 1994.
[10] M. A. Rahman and M. A. Hoque, On-line adaptive articial neural
network based vector control of permanent magnet synchronous motors,
IEEE Trans. Energy Convers., vol. 13, no. 4, pp. 311318, Dec. 1998.
[11] Y. Yi, D. M. Vilathgamuwa, and M. A. Rahman, Implementation of
articial neural network based controller for an interior permanent mag-
net motor drive, IEEE Trans. Ind. Appl., vol. 39, no. 1, pp. 96104,
Jan./Feb. 2003.
[12] M. A. Rahman, M. A. Hoque, C. B. Butt, M. N. Uddin, and M. A. Abido,
Testing of genetic algorithm based PI controller for IPMSM drive,
in Proc. IEEE Int. Conf. Ind. Technol., Bangkok, Thailand, Dec. 2002,
pp. 119124.
[13] M. N. Uddin, M. A. Abido, and M. A. Rahman, Laboratory implementa-
tion of an articial neural network for online tuning of a genetic algorithm
based PI controller for IPMSM drive, in Proc. Int. Conf. Model., Simul.
Elect. Mach., Convert. Syst., Montreal, QC, Canada, Aug. 2002.
[14] S. Morimoto, Y. Takeda, and T. Hirasa, Current phase control meth-
ods for permanent magnet synchronous motors, IEEE Trans. Power
Electron., vol. 5, no. 2, pp. 133139, Apr. 1990.
[15] P. Pillay and R. Krishnan, Control characteristics and speed controller
design for a high performance permanent magnet synchronous motor
drive, in Proc. IEEE Power Electron. Spec. Conf., 1987, pp. 598606.
[16] T. M. Jahns, G. B. Kliman, and T. W. Neumann, Interior permanent
magnet synchronous motors for adjustable speed drives, in Conf. Rec.
IEEE IAS Annu. Meeting, 1985, pp. 814823.
[17] B. K. Bose, A high performance inverter-fed drive system of an inte-
rior permanent magnet synchronous machine, IEEE Trans. Ind. Appl.,
vol. IA-24, no. 6, pp. 987997, Nov./Dec. 1988.
[18] S. Morimoto, M. Sanda, and Y. Taketa, Wide-speed operation of inte-
rior permanent magnet synchronous motors with high-performance cur-
rent regulator, IEEE Trans. Ind. Appl., vol. 30, no. 4, pp. 920926,
Jul./Aug. 1994.
[19] M. F. Rahman, L. Zhong, and K. W. Lim, A direct torque-controlled
interior permanent magnet synchronous motor drive incorporating eld
weakening, IEEE Trans. Ind. Appl., vol. 34, no. 6, pp. 12461253,
Nov./Dec. 1998.
[20] M. A. Hoque, C. Butt, and M. A. Rahman, A novel approach for MTPA
speed control of IPMSM drive, in Proc. 2nd IEEE Int. Conf. Elect.
Comput. Eng., Dhaka, Bangladesh, Dec. 2627, 2002, pp. 336339.
Casey B. Butt received the B.Sc. degree in physics
and the B.Eng., M.Eng., and Ph.D. degrees in elec-
trical engineering from the Memorial University of
Newfoundland, St. Johns, NL, Canada, in 1995,
2002, 2003, and 2007, respectively.
In 2007, he joined the hybrid electric vehicle divi-
sion of General Motors Company, Detroit, MI, USA.
His research interests are in the areas of electric
machines and intelligent control systems.
M. Azizur Rahman (M66SM73F88LF07)
was born in Santahar, Bangladesh, on January 9,
1941. He received the B.Sc. degree in electrical engi-
neering from the Bangladesh University of Engineer-
ing and Technology, Dhaka, Bangladesh, in 1962,
the M.A.Sc. degree in electrical engineering from
the University of Toronto, Toronto, ON, Canada, in
1965, and the Ph.D. degree in electrical engineer-
ing from Carleton University, Ottawa, ON, Canada,
in 1968.
He is currently a Professor and University Re-
search Professor at the Memorial University of Newfoundland, St. Johns,
NL, Canada. He has published over 700 papers, 178 IEEE TRANSACTIONS
papers, and two books, is the holder of 11 patents, and contributed ten chapters
for research and academic publications. His current research interests are
in machines, intelligent controls, power systems, digital protection, power
electronics, and wireless communication.
Prof. Rahman is a Registered Professional Engineer in the Provinces of
Newfoundland and Labrador, Canada. He is a Fellow of the Institution of
Engineering and Technology, U.K., a Fellow of the Engineering Institute
of Canada, a Fellow of the Canadian Academy of Engineering, and a Life
Fellow of the Institution of Engineers, Bangladesh. He was the recipient of
the following awards: the IEEE Notable Service Award in 1987, IEEE Industry
Application Societys Outstanding Achievement Award in 1992, IEEE Canada
Outstanding Engineering Educators Medal in 1996, IEEE Third Millennium
Medal in 2000, IEEE Cyril Veinott Electromechanical Energy Conversion
Award in 2003, IEEE William E. Newell Power Electronics Award in 2004,
IEEE Dr.-Ing. Eugene Mittelmann Achievement Award in 2007, IEEE Richard
H. Kaufmann Technical Field Award in 2007, and IEEE Power and Energy
Society Distinguished Service Award in 2008.