You are on page 1of 10

1

DSP
Research on Sensorless Control Algorithms for BLDCMs and
Development of DSP Realization Techniques
Kuang-Yao Cheng () and Ying-Yu Tzou ()


Power Electronics and Motion Control Lab.,
Department of Electrical and Control Engineering, National Chiao Tung University

---

DSP(TMS320LF2407A)

PC-MATLAB


Abstract This dissertation presents the design and realization
of the sensorless control algorithms for brushless DC motors
(BLDCMs). Two sensorless methodologies are presented a
modified back-EMF measurement method, and a flux-based
rotor position estimation approach. By analyzing the
commutation characteristics, a modified back-EMF
measurement approach, which includes an adaptive delay
sampler, a hysteresis comparator, and a digital filter, is
presented for improving the accuracy of the zero-crossing
detection. A novel phase shifter is proposed to generate accurate
sensorless commutation signals over a wide speed range. The
phase shift errors with respect to the acceleration transient
operation and measurement noises are also analyzed to verify
the robustness and the constraints. Since the commutation
control is very critical for the BLDCM drive, a phase
compensation strategy is proposed to compensate the phase
error due to the low-pass filtering, measurement noises, and
nonideal trapezoidal-shape back-EMFs to optimize the
efficiency during the steady-state operations. Another proposed
sensorless control scheme is to present a general rotor position
estimation algorithm for both sinusoidal-type permanent
magnet synchronous motors (PMSMs) and trapezoidal-type
BLDCMs. Based on the estimated three-phase flux-linkage
incremental, the rotor position incremental can be estimated
with a weighted-sum approach. The integration drift and offset
problem for the flux linkage estimation can be avoided. Besides,
an internal closed-loop correction mechanism is presented to
improve the accuracy and robustness of the rotor position
estimation. The effect of the speed estimation accuracy for the
speed control loop is also addressed. Three kinds of rotor speed
estimators, which are based on the commutation signals, non-
excited phase back-EMF voltages, and the estimated rotor
position, are presented and compared. From the simulation
analyses, the proposed sensorless approach can achieve good
speed regulation during steady-state operations as well as
transient operations over a wide speed control range. The
comparison of the proposed two sensorless methodologies is also
given. For the speed control loop design of the sensorless
BLDCM drive, a novel fuzzy optimization strategy is proposed
for automatic tuning of the speed control parameters. By
combining the concept of the gradient optimization and the
fuzzy-logic linguistic description, expert knowledge can be
converted into a fuzzy step-size tuner to improve the convergent
rate of the overall optimization process. In comparison with
other existing tuning schemes, the proposed fuzzy optimization
strategy has advantages of robust performance and guaranteed
stability. This scheme is realized with a two-level hierarchical
architecture. The higher level is the fuzzy optimization
algorithm executed by PC-MATLAB, and the lower level is the
sensorless drive. In addition, various digital implementation
issues with a fixed-point DSP (TMS320LF2407A) are addressed,
including the synchronous sampling technique and the
quantization effect. Experimental results show that the
proposed sensorless control scheme can achieve the desired
speed control performance over a wide speed range including
the speed reversal under varying load torques.
Index TermsBrushless dc motors, sensorless algorithms,
commutation phase compensation, speed estimation, position
estimation, fuzzy optimization, DSP realization.
I. INTRODUCTION
Adjustable-speed motor drives are widely used in various
applications, such as manufacturing automation systems and
residential heating, ventilation, and air conditioning (HVAC)
equipments. In the past, DC motors were extensively used in
these applications. However, the commutators and brushes of
DC motors would reduce the reliability of the drive systems.
The permanent magnet AC motors (PMACMs) have emerged

2
as viable candidates for high-performance servo drive
applications to solve the maintenance problems of the DC
motors. The PMACMs are classified on the basis of the wave
shape of the induced back-EMFs, i.e. sinusoidal and
trapezoidal [3]. The sinusoidal type is known as PM
synchronous motor (PMSM), and the trapezoidal type is
brushless DC motor (BLDCM). The BLDCMs have 15%
more power density than PMSMs, due to the higher ratio of
the rms value to the peak value of the flux density [4].
Besides, since only two phases are excited at any instant of
the BLDCM control, the commutation control of the
BLDCMs is much simpler than the PMSMs, which require
continuous and instantaneous absolute rotor position.
Therefore BLDCMs are becoming more attractive for many
industrial applications, such as compressors, electrical
vehicles, and DVD players etc. Because BLDCMs use
permanent magnets for excitation, rotor position sensors are
needed to perform electrical commutation. Usually, three
Hall-effect sensors are used as rotor position sensors for a
BLDCM. However, the rotor position sensors present several
disadvantages from the standpoint of total system cost, size,
and reliability. For this reason, many research working on
controlling the motor speed of BLDCMs without rotor
position sensors have been reported in the literatures [5]-[47],
which can be generally classified into three types.
The first method is directly or indirectly measure
position-dependent variables, such as back-EMFs and third-
harmonic voltages [8]-[17]. Though this method is easily to
be implemented to provide rotor position information,
directly access to the power converter or machine terminals is
usually necessary and may introduce noise. Since the back-
EMF is very small at low speed and is zero at standstill, the
direct measurement method can not work at low-speed
operations. In addition, auxiliary hardware circuits are
required for the indirect measurement method. The second
method is to derive the rotor position and speed information
with a mathematic motor model by measuring the phase
voltages and currents. This is so called observer-based
method [18]-[37]. Since the motor model is adopted to
calculate the flux vectors, this method is generally heavily
dependent on the model parameters. Some on-line adaptive
tuning methods have been proposed to compensate the
parameter uncertainty. However, more computational efforts
are required for the adaptation, which implies that a more
powerful microprocessor is necessary. The third sensorless
method is to utilize the rotor saliency or magnetic saturation.
The rotor position information is extracted by exciting a
high-frequency signal to the motor. Several signal injection
patterns and modified pulse-width-modulation (PWM)
techniques have been reported [38]-[47]. Since this method is
independent of the rotor speed and motor parameters, this
method can be used at low speeds and standstill. However,
the high-frequency signal injections may cause some
undesirable effects to the drives, such as torque pulsations.
Although there are many sensorless approaches in the
literature, few of them can be both applied to the PMSMs and
BLDCMs [21]-[24]. Since the flux distribution in a BLDCM
is trapezoidal, therefore, the well-known d-q rotor reference
frames model developed for the PMSM is not applicable, and
some properties of the two-axis stationary reference frame
can not be used either.
The research objectives of this dissertation is thus in
three-folds. The first one is to develop an effective and
practical sensorless control technique for the BLDCMs to
reduce the hardware complexity and cost, and to increase the
mechanical robustness and reliability compared to the
previous existing methods. A modified back-EMF
measurement based sensorless approach is presented. With
considering the realization constraints, the proposed
algorithm has been successfully realized with integrated-
circuit (IC) architecture for many low-cost applications [15].
The second objective of this research is to develop a general
rotor position and speed estimation algorithm for PMACM
drive system over a wide speed control range. In this
approach, the flux linkage increments are calculated to derive
the rotor position. Without integrating the flux linkage
increments, the integration drift or offset problems can be
avoided with this approach. The detailed realization issues of
the sensorless algorithm will be introduced by using a
commercial Texas Instrument digital signal processor (DSP)
with fixed-point arithmetic (TMS320LF2407A). The third
objective of this study is to develop a control parameter
tuning strategy for the high-performance motor drives. The
developed PC-based auto-tuning software can be used for the
on-line tuning of the control parameters of a digital motor
drive through the RS-232/485 serial interface. The selected
control parameters can be off-line tuned according to a
selected performance index. In a practical servo drive, there
exist many nonlinear characteristics and physical constraints
such as voltage and current limits, detuning effect of vector
control, parameter variations due to temperature variations,
quantization error due to limited sampling effect, etc. In order
to overcome uncertainties due to these effects, a fuzzy-logic-
based optimization scheme is developed for a practical digital
motor drive. The proposed fuzzy optimization algorithm will
be verified on a PC-MATLAB environment.
II. SENSORLESS COMMUTATION CONTROL
A. Back-EMFs Measurement and Zero-Crossing Detection
From the mathematical modeling of the BLDCM, the sum
of three terminal voltages three terminal voltages, v
a
, v
b
, and
v
c,
can be derived as follows
( )
n cn bn an c b a
v v v v v v v 3 + + + = + + (1)
For analyzing the effectiveness of the back-EMF estimation,
the commutation from phase a-b to phase a-c is considered as
an example. During the two-phase conducting period, two
conducting phase currents are opposite and another one is
zero, that is i
b
=-i
a
, i
c
=0. Therefore, from (1), the back-EMF of
the non-excited phase c is
( ) ( ) | | ) ( ) ( ) (
3
1
) (
e c e b e a c b a c n c e c
e e e v v v v v v e u u u u + + + + = = (2)
By simplifying (2), e
c
can be estimated as follows
( ) ( ) | | ) ( ) ( 3
2
1
) (
e b e a c b a c e c
e e v v v v e u u u + + + + = (3)
For an ideal trapezoidal back-EMF, e
a
(
e
)+e
b
(
e
) is equal to
zero. Hence, the non-excited phase back-EMF can be fully
estimated by (3) with only three terminal voltages. For the
nonideal or distorted back-EMFs [48], however, e
a
(
e
) is
different from -e
b
(
e
), that is
0 ) ( ) ( ) ( = = +
e n e b e a
e e e u u u (4)
Due to this nonzero term, the back-EMF estimation
algorithm in (3) would produce an electrical angle-dependent
error, and cause a phase shift to the zero-crossing point of the

3
Time
a
i
b
i
c
i
I
I
T
fw
b
e
E
(1) (2)
(1): Two-phase conducting period
(2): Three-phase conducting period
E
c
e
Time
(1)
2
dc
V

Fig. 1. Back-EMF measurement during two-phase and three-phase
conducting period.
0 60 180 240 360 300 120
u
e
Commutation
Zero-crossing
Delaycount
th
D
Select
k
e
'
k e
(after
delaysampling)
(before
delay sampling)

Fig. 2. Typical waveforms of non-excited phase back-EMFs and zeror-
crossing signals.
real back-EMF. If the shapes of the back-EMFs can be
known, this error can be compensated in advance. However,
for unknown back-EMFs, this error would cause inaccurate
commutation control. Therefore, some phase compensation
strategies have been proposed to improve the accuracy of the
estimation [12]-[14]. Fig. 1 shows the back-EMF estimation
during the two-phase conducting period.
During the commutation period, three phases are
conducted until the decaying current reduces to zero. It
should be noted that the estimation algorithm in (3) is only
valid by assuming the non-excited phase current is zero.
Hence (3) would become a wrong estimation for back-EMFs
during the three-phase conducting period. In this example,
during the commutation from phase a-b to phase a-c, the
terminal voltage v
b
is equal to half dc-link voltage ( 2
dc
V )
due to the free-wheeling diode clamping effect until phase
current i
b
vanishes as shown in Fig. 1. Hence, if (3) is applied
for the estimation, then the estimated back-EMF of non-
excited phase a would be derived as follows
( ) | |
2
) (
) ( 3
2
1
) (
e n dc
e n c b a b e b
e V
e v v v v e
u
u u
+
= + + + = (5)
Obviously, ) (
e b
e u differs from the real back-EMF ) (
e b
e u .
Since the back-EMF estimation can no be achieved during
the three-phase conducting period with only three terminal
voltages, a delayed sampling technique is required to mask
the estimation for avoiding wrong zero-crossing detection
during this period. Fig. 2 illustrates the typical waveforms of
the estimated non-excited phase back-EMF and the
corresponding zero-crossing signal during two-phase
conducting period and three-phase conducting periods. It
should be noted that the duration of the three-phase
conducting period would depend on the operating conditions
[49]. Hence the delayed-sampler should be adjusted
according to the different operations. In this paper, an
adaptive delay controller is presented to improve the back-
EMF estimation. Fig. 3 shows the block diagram of the
modified back-EMF estimation and zero-crossing detection.
The threshold value D
th
for the delay controller can be
adaptively calculated as

E V
I L
D
dc
s
th
2
3
+
= (6)
Since the estimated back-EMF signal must contain noises, a
hysteresis comparator with the hysteresis band Z
h
is used for
avoiding wrong zero-crossing detection. Besides, a digital
filter is also used to filter out the possible noise.
B. Commutation Phase Shifter
After detecting the zero-crossing signal of the estimated
non-excited phase back-EMF, an additional 30-degree phase
shift is required to perform correct commutation.
Conventionally, this phase shift is generated using an analog
filter [8]. However, the phase lag resulted by the filter varies
with motor speed; hence the accuracy of the sensorless
commutation control depends on the rotor speed. In [11], a
novel frequency-independent phase shifter (FIPS) has been
proposed as shown in Fig. 4(a). The algorithm of this phase
shifter is very simple and useful for sensorless commutation
control. Since the count values of c
p
(k) and c
n
(k) depend on
the rotating speed, and the value of scaling factor is less
than one, a long bit-length divider is required at low speed
operation to compute c
p
(k) and c
n
(k), i.e. the commutation
instants, precisely. However, the computation effort of the
long bit-length divider is quite large for the real-time
applications. In this paper, a digital simplified-type FIPS,
which only needs a simple multiplier instead of a long bit-
length computation unit, is presented as shown in Fig. 4(b) to
reduce the computation effort of the original FIPS. To
describe the basic operation of the proposed digital phase
shifter, an ideal periodic input signal is considered. Fig. 5
illustrates the operational waveforms of the proposed phase
shifter. Assume that the input x(k) in Fig. 4(b) is the zero-
crossing signal z(k) of the non-excited phase back-EMF, and
the output y(k) is the corresponding commutation signal h(k).
h
*
(k) is the desired commutation signal. Also,
d
is set as two-
times of the increasing increment
i
to make the phase shift
equal to 30 degrees from z(k). k
zn
denotes the time when the
nth zero-crossing of input signal z(k) occurs, and k
cn
denotes
the time when the nth commutation occurs. In order to
analyze the effectiveness and robustness of the proposed
phase shifter, two kinds of input signals are considered: i) the
frequency of the input signal is accelerating and no
measurement noise (error due to acceleration); ii) the input
Non-excited
Phase
Back-EMF
Estimation
Delay Sampler
Adaptive
Delay
Controller
Digital
Filter
a
i
b
i
c
i
k
e
a
v
b
v
c
v
'
k
e
a
H

b
H

c
H

th
D
h
Z
k
z
'
k
z

Fig. 3. Block diagram of modified back-EMF estimation and zero-
crossing detection.

4
signal is periodic and includes measurement noises (error due
to noise). First, consider the case (i) and assume the
acceleration rate is
0
o . Fig. 6(a) shows the typical waveform
of an accelerating input signal is depicted. The phase shift
error
1
c due to acceleration can be calculated as follows

( )
( )
( ) ( ) ( )
( )
2

) (
2 1
3
2
1 1
) (
2 1
3
2
1 1 3 1
2 1 3
1

6
) ( ) (
6
) (
2
1
0
2
1
0
1
0
1
0
1
1
1
t
u
o

t
u
o

t
t
u u
t
u c

+
(

+ + +
+
+
=
= =

c
c
z
k
c
k
c
k
z
k
k
k
k k k


(7)
From (7), it can be seen that the phase shift error
1
c due to
acceleration is a function of ,
0
o , and ) (
1 c
k u

. A higher
rotating speed ) (k u

will produce a lower phase shift error


based on (7), which implies that better performance can be
achieved at higher speed. Since ) (
2 c
k u

is faster than ) (
1 c
k u

,
the phase shift error
2
c is lower than
1
c as shown in Fig.
6(a). It should be noted that the phase shift error will not be
accumulated. Hence the accurate commutation signal can be
estimated at the instant k
c3.
Next, the case (ii) is considered.
Fig. 6(b) illustrates a periodic input signal with the frequency
0
e , and two kinds of noises are inserted. One wrong zero-
crossing signal with time interval m
1
is generated at k
z1
for
simulating the error due to free-wheeling diode-clamping
effect, and the other wrong zero-crossing signal with time
interval m
2
is generated at k
z4
for simulating the measurement
error due to flatness around zero. The phase shift error
3
c
due to the noises can be described by

| |
) )( 1 (
) )( 1 ( ) 1 ( ) (
1 2 0
1 2 3 3 0 1
*
1 0 3
m m
m m k k k k
z z c c
+ =
+ + + = =
e
e e c
(8)
where
*
1 c
k denotes the desired commutation instant. From
(8), it can be seen that the phase shift error
3
c due to input
noises is a function of
0
e , , m
1
, and m
2
. It should be noted
that the phase shift error
3
c increases with increasing rotor
speed, which implies that the input noises have significant
effect at high-speed operation. Also, note that when
2 1
m m = ,
the phase shift error
3
c becomes zero, that is, the effect of
noises can be cancelled by the integral action of the proposed
phase shifter. Besides, phase-advanced control, i.e. 5 . 0 < ,
can also reduce the phase shift errors. Similar analysis can be
done for the next zero-crossing instant k
z6
, the phase shift
error
4
c due to the noises can be derived as

2 0 2
*
2 0 4
) 1 ( ) ( m k k
c c
e e c + = = (9)
A phase shift error
4
c

would occur at k
c2
, but would be
vanished at next commutation instant k
c3.

1
0
1
0
n
x
p
x
+
+
d

-
-
) (k x
) (k c
n
L
L
) (k c
p
1
0
1
0

=
k
k m
p
P
m x ) (
n
x
p
x
L
L
L
L
+
1
0
1
0
+

+
-
-
-

=
k
k m
n
n
m x ) (
) (k x
) (k c
p
) (k y
(a)
(b)
) (k c
n
1
0
1
0
) (k y
Commutation
Logic
i

=
k
k m
n
n
m x ) (

=
k
k m
p
p
m x ) (

Fig. 4. Block diagramof digital phase shifters.
k
k
k
k
0 =
n
k
1 c p
k k =
0 =
n
k
1 c p
k k =
2 c n
k k =
) (k c
p
) (k c
n
) (k h
) (
*
k h
0
0
0
1
1
1
1
L
1 z
k
1 c
k
2 c
k 3 c
k
k
) (k z
2 z
k
3 z
k
L
0 =
p
k

Fig. 5. Ideal operational waveforms of the proposed phase shifter.

k
k
k
k
0 =
p
k
0 =
n
k
1 c p
k k =
0 =
n
k
1 c p
k k =
2 c n
k k =
) (k c
p
) (k c
n
) (k h
) (
*
k h
0
0
0
1
1
1
1
L
1 z
k
1 c
k
2 c
k
3 c
k
k
) (k z
2 z
k
3 z
k
L
1 c p
k k =
2 c n
k k =
1
c
2
c

(a)
k
k
k
k
0 =
n
k
1 c p
k k =
0 =
n
k
1 c p
k k =
2 c n
k k =
) (k c
p
) (k c
n
) (k h
) (
*
k h
0
0
0
1
1
1
1
L
1 z
k
1 c
k
2 c
k
2 z
k
3 z
k
k
) (k Z
4 z
k
5 z
k
L
0 =
p
k
1
m 2
m
6 z
k
7 z
k
3
c
4
c
1 c
k

(b)
Fig. 6. Error analyses of the proposed digital phase shifter.

5
e
k
T
c
1 c
k
2 c
k
Back-EMF
p
e
p
e
PhaseA
(Non-excited)
PhaseB
(Non-excited)
PhaseC
(Non-excited)
3 c
k 4 c
k
) (k h
k
k
Sampling for
speed control
T
s

Fig. 7. Concatenate waveformof ideal non-excited phase back-EMFs.

Fig. 8. Simulation result of sensorless ramp-speed control performance with
different acceleration rates.
C. Commutation Phase Compensation
In order to produce maximum torque at low-speed
operation for BLDCMs, the phase currents are required to be
aligned in phase with the back-EMF waveforms, i.e. phase
shift between zero-crossing of the non-excited phase back-
EMF and real commutation signal is 30 degrees [50].
However, a commutation phase error may exist due to the
phase-lag of low-pass filtering, noises, transient operations,
and nonideal effect of the estimated back-EMF. In this paper,
the proposed digital phase shifter provides a phase
compensation capability for adjusting excitation angles for
the sensorless commutation control of BLDCMs. The phase
compensation can be easily implemented by fixing the
decreasing slew rate
d
, and changing the increasing slew rate

i
with the following equation:

d
ad f
i

u u

60
30
*

= (9)
where u
f
is the phase lag due to the low-pass filtering of the
terminal voltages, and
*
ad
u is the desired advanced angle.
D. Rotor Speed Estimation
The accuracy of speed estimation is very crucial for
closed-loop speed control. Since the commutation signal h(k)
can be estimated with the presented algorithm, the time
interval T
c
between two commutations can be easily
calculated as shown in Fig. 7. Hence the rotating speed can
be estimated as follows [8],[13]:

c c
e r
PT PT
f
P
20
6
120 120
~
= = = e (10)
where
r
e
~
is the estimated rotating speed from commutation
signals, P denotes the number of rotor poles, and f
e
is the
electrical rotating frequency. However, this method can only
update the estimated speed when commutation occurs, and
works poorly at low-speed operations because of no
information between two commutations. In order to improve
the speed estimation at low-speed operation, this paper
proposes a novel speed estimation technique, which can
estimate the instantaneous speed at each sampling instant of
speed control loop. Fig. 7 also presents a concatenation
waveform of the ideal non-excited phase back-EMF signal
e(k). The rotating speed can be also estimated from the peak
value e
p
of the back-EMF as follows:

E
p
r
K
e
= e (11)
where K
E
is the back-EMF constant of a BLDCM. The
estimation algorithm in (11) suffers from a similar problem
as (10), because the peak value e
p
can be only actually known
at each commutation instant. However, e
p
can be predicted
from the sampled back-EMF signal e(k) with sampling period
T
s
between two commutations. When the sampling instant is
between two commutations, i.e. k
c1
<k<k
c2
, the speed
estimation algorithm is derived as follows:
) ( ) (
) (
~
20
) (
1 c r
r s E
r
k k e
k PT K
k e
e
e A = (12)
where ) (
~
k
r
e is equal to ) (
~
1 c r
k e , because no commutation
occurs during this period. From the observation of (12) the
estimated speed ) ( k
r
e can be updated at each sampling
instant of the speed control loop by differentiating the non-
excited phase back-EMF.
E. Simulation Analyses
Fig. 8 shows the comparison of the ramp speed responses
with different acceleration rates. It is clear that the sensorless
commutation phase error due to acceleration increases with
increasing acceleration rates. Fig. 9 shows the comparison of
the closed-loop sensorless control performance at low-speed
operations with two different speed estimation techniques. It
is clear that the control performance with estimated speed
from commutation signals is worse than the one with
estimated speed from back-EMF voltages, especially for the
disturbance rejection ability. Therefore, the proposed rotor
speed estimation algorithm can extend the sensorless speed
control range to lower speeds compared to the conventional
method [8],[13]. From the simulation analyses, accurate
sensorless commutation control can be achieved for the
BLDCM drive at low-speed and high-speed steady-state
operations as well as transient operations. Therefore, this
approach can be effectively used for many low-cost
applications. However, for very-low speed operations (below
100 rpm), this sensorless approach is problematic due to a
low signal-to-noise ratio (SNR) in voltage signals, which are
the major variables for estimating the commutation signal
and rotor speed. Besides, the proposed sensorless approach
can be only applied to the BLDCMs with trapezoidal back-

6

(a)

(b)
Fig. 9. Simulation result of closed-loop sensorless speed control performance
at low-speed operations with different speed estimations: (a) from
commutations, (b) fromback-EMFs.
EMFs. Therefore, a more general and robust sensorless
control algorithm will be developed in the following section.
III. ROTOR POSITION ESTIMATION ALGORITHM
For a PMACM, the induced back-EMF voltages are
functions of the rotor position, and are proportional to the
angular speed. Hence, the three-phase back-EMF voltages
can be expressed as

(
(
(

=
(
(
(

) (
) (
) (
3
2
1
e r E
e r E
e r E
c
b
a
e K
e K
e K
e
e
e
u e
u e
u e
(13)
where e
1
, e
2
, and e
3
are normalized trapezoidal or sinusoidal
flux-distribution functions. From (13), it can be seen that if
the back-EMF constant K
E
is assumed constant, the rotor
position can be estimated from the back-EMF voltages and
the rotor speed. However, if only the relationship between the
rotor position and the back-EMF voltages is used, the
position estimation scheme may not operate accurately at low
speeds due to the low signal-to-noise ratio of the back-EMF
voltages. Besides, only the non-excited phase back-EMF
voltage can be directly measured for the BLDCMs as
discussed in the previous chapter. Hence, in order to improve
the accuracy of the rotor position estimation, the relationship
between the rotor position and the flux linkage of the
permanent magnets is also utilized in the rotor position
estimation algorithm. The total flux linkages in the motor can
be obtained by integrating the three-phase voltages and
currents as follows:

t
t
t
t
t
t
t
t
t
t
t

d
e
e
e
t i
t i
t i
L
L
L
d
i
i
i
R
R
R
v
v
v
t
t
t
t
c
b
a
c
b
a
s
s
s
t
c
b
a
s
s
s
cn
bn
an
c
b
a
}
}

(
(
(

+
(
(
(

(
(
(

(
(
(

(
(
(

(
(
(

=
(
(
(

0
0
) (
) (
) (
) (
) (
) (
0 0
0 0
0 0

) (
) (
) (
0 0
0 0
0 0
) (
) (
) (
) (
) (
) (
(14)
From (13) and (14), the flux linkages are functions of phase
currents and the rotor position. Therefore, the rotor position
can be deduced from the estimated flux linkage. However,
the integration computation in (14) suffers from the
integration drift due to motor parameter variations and
measurement errors. This drift induces a significant problem
in the estimation of the flux linkages, especially at low
speeds [18]-[27]. In order to overcome the drawbacks of the
drift problem, a novel rotor position estimation algorithm is
proposed by utilizing flux linkage increments. Substituting
(13) into (14), the flux linkage equations can be rewritten as

(
(
(

(
(
(

(
(
(

(
(
(

) (
) (
) (
) (
) (
) (
0 0
0 0
0 0
) (
) (
) (
3
2
1
e
e
e
e E
c
b
a
c
b
a
e
e
e
dt
d
P
K
t i
t i
t i
L
L
L
t
t
t
dt
d
u
u
u
u

(15)
For digital implementation, if the sampling frequency is high
enough, the rotor position increment within one sample
interval can be estimated from (15) as follows

(
(
(
(
(
(
(

A A
A A
A A
=
(
(
(

A
A
A
)

(
)

(
)

3
2
1
e
c c
e
b b
e
a a
E
ce
be
ae
e
i L
e
i L
e
i L
K
P
u

u
u
u
(16)
where i
a
, i
b
, and i
c
are the phase current increments, and

a
,
b
, and
c
are the flux linkage increments within one
sampling interval of each phase. Ideally, each phase should
produce identical rotor position increment, that is

ce be ae
u u u u

A = A = A = A (17)
where u

A is the estimated rotor position increment. By


using a weighted average method [24] to estimate the rotor
position increment, (16) can be rewritten as follows:

( )
( )
( )
(
(
(

A A
A A
A A
=
(
(
(

A
A
A
)

(
)

(
)

(
)

( )

( )

( )

1
3
2
1 3
3 2
2 1
e c c
e b b
e a a
E
e e ce
e e be
e e ae
e i L
e i L
e i L
K
P
e e
e e
e e
u
u
u
u u u
u u u
u u u
(18)
By summing each row in (18), the general rotor position
Flux Linkage
Increments
Calculation
Rotor Position
Increment
Estimation
Back-EMF
Function
Generation
an
v
a
i
a a
i LA A
1
e
1
z
+
+
b b
i LA A
c c
i LA A
2
e
3
e
bn
v
cn
v
b
i
c
i
u

A
e
u

(Internal Cl osed-Loop Correction)



Fig. 10. Block diagram of the internal closed-loop rotor position
correction mechanism.

7
estimation algorithm can be derived as
| |
( ) ( ) ( ) | | )

( )

( )

(
)

( )

( )

( )

( )

( )

(
) 1 (

) (

1 3 2
1 3 3 2 2 1
e c c e b b e a a
e e e e e e E
e e
e i L e i L e i L
e e e e e e K
P
k k
u u u
u u u u u u
u u
A A + A A + A A
+ +
+ =
(19)
With the presented method in (19), the integration
computation is moved from the flux linkage estimation to the
rotor position estimation. Besides, the internal closed-loop
correction mechanism can correct the estimated rotor position
at each sampling instant to stabilize the position integration
as shown in Fig. 10. This is one important feature of this
algorithm. Furthermore, this internal correction loop can also
improve the robustness of the rotor position estimation with
respect to the parameter variations and measurement noises.
Fig. 11(a) shows the sensorless speed control
performance of the sensorless BLDCM drive over a wide
speed range. From the point of view of the practical
application, the proposed sensorless approach has good
performance over a wide speed control range even during the
speed reversal as shown in this figure. Besides, the sensorless
estimation algorithm is also applied to a PMSM with
sinusoidal back-EMFs to verify the generality of this
approach. Fig. 11(b) shows the similar simulation result of
the sensorless speed control performance. The performance is
consistent with the BLDCM shown in Fig. 11(a).
IV. FUZZY OPTIMIZATION ALGORITHM
In order to specify the relationship between the servo
control parameters and servo performance, a performance
evaluation method is presented. Two objective functions are
defined for evaluating the speed dynamic response as follows

( )
) ( ) (
2
1

for , ) ( ) (
2
1
) (
2
*
N N
T
t
k m m N J
T
E E
r
k
p
Nk
p
Nk m s
r
r r r TR
O O =
A
= =

+
=
e e
(20)
) ( ) (
2
1
) (
2
1
) (
1 ) 1 (
2
*
N I N I m i N J
T
Q Q
p
k N
p
Nk m
q
CE
= =

+
=
(21)
where J
TR
(N) and J
CE
(N) denote the objective functions of the
speed transient response and the corresponding control effort
at the N cycle, respectively. The desired rise time t
r
of speed
response can be determined when the rotor speed is greater
than
*
9 . 0
d
e . With small gains, the control effort is small and
the speed response is very slow. By increasing the velocity
loop gains, the transient speed error may decrease to a
minimum value with increasing the control effort.
After defining these objective functions, the design
problem for high-performance servo drives becomes as a
multi-objective optimization problem, which implies to
minimize all these four objective functions simultaneously.
Since there are some tradeoffs between these objective
functions, the multi-objective problem is difficult to be
solved. In order to solve this problem, a weighted-sum
method [51] is presented for representing the relative
importance of each objective function. By properly
combining the weighted objective functions, a convex
objective function can be formulated for determining the
optimal values of servo control parameters as follows

T
V E
J W J = (22)
where ] , [
CE TR
J J J = , and W
V
=[w
TR
,w
CE
] is a weighting
vector, which is used to specify the performance
specifications. Hence the optimization problem can be re-
formulated as follows:

T
V
K
J W min
w.r.t.

subject to
(max) (min)
K K K < < (23)
where ] , [
vi vp
K K K = is the control parameter vector.
(max)
K and
(min)
K are the maximum and minimum vectors of
K. The combined objective function for evaluating velocity
control performance can be rewritten as

( )
CE TR CE
CE TR
r d
a q
CE
CE CE TR TR V
J J w w
J J
t
t i
w
J w J w J
+ =
|
|
.
|

\
|
+
A
A
=
+ =
2
2 *
2 *
(max)

) (
) (


e
(24)
where w
CE
can be arbitrarily assigned. Since the combined
objective functions are convex functions, the gradient method
can be applied to find the optimal control parameters for
minimizing the objective functions [52]. The update
equations for each control parameter are

(a)

(b)
Fig. 11. Simulation result of closed-loop sensorless speed control
performance over a wide speed range for: (a) a BLDCM, (b) a PMSM.

8
z
z 1
Fuzzification
Inference Engine
V
J
V
J A
V J
K
V J
K
A
Knowledge Base
(Fuzzy Rul es)
Defuzzification
vp
K
q
vp
q
vi
K
q
vi
q

Fig. 12. Block diagram of the fuzzy step-size tuner.
+

Servo Motor
and
Load Dynamics
Servo Controller
Servo
Command
Servo
Response
Gradient-Based
Optimization Algorithm
Performance
Evaluation
Update
Controllers
(Supervisor Level Batch Processing)
(Hardware Level Real-Ti me Processing)
Servo
Response
Fuzzy-Logic Based
Step-Size Tuner

Fig. 13. Hierarchical fuzzy optimization scheme for the sensorless BLDCM
drive.

) (
) (
) ( ) 1 (
N K
N J
N K N K
vp
V
vp vp vp
c
c
= + q (25)

) (
) (
) ( ) 1 (
N K
N J
N K N K
vi
V
vi vi vi
c
c
= + q (26)
where N denotes the current iteration number, and q
vp
, and
q
vi
are the step-sizes for the optimization equations. In
general, small step-sizes may lead to an inefficient search
process. On the other hand, large step-sizes allow the search
process to approach the minimum efficiently. However, large
step-sizes may not guarantee the descent direction for
optimization and may cause oscillation around the local
minimum point. Several methods have been proposed to
adaptively adjust the step-sizes during the optimization
process [53]. For example, the steepest descent algorithm can
be applied to choose the optimal step-size for the maximum
decreasing amount of the objective function at each
increment. However, this method needs to calculate the
objective values with many different step sizes and may slow
the overall optimization process. In this study, a fuzzy step-
size tuning strategy is proposed to adaptively update the step
sizes for achieving both fast convergent rate and precise
optimum results of the optimization process. Since fuzzy-
logic can simply transfer the expert-knowledge into an
algorithm by using linguistic descriptions, the proposed
strategy can best extract the experts knowledge. Fig. 12
shows the detailed block diagram of the fuzzy step-size tuner
for optimization algorithms. The inputs of the fuzzy step-size
tuners are the combined servo loop objective function J
V
(N),
and the change of the combined objective functions AJ
V
(N)=
J
V
(N)- J
V
(N-1). The output variables are the step sizes for
updating velocity control parameters. Fig. 13 shows the block
diagram of the proposed fuzzy optimization scheme for the
speed controller. This scheme is composed of a two-level
hierarchical architecture. The higher level is the proposed
fuzzy optimization algorithm, which executes in a batch
processing mode with lower priority, and the lower level is
the servo control system, which operates in a real-time
processing mode with higher priority. By using the
hierarchical structure in realizing the fuzzy-logic based
optimization techniques for digital servo drives, the benefits
of computational efficiency and learning capabilities of
intelligent control strategy can be combined [54].
V. EXPERIMENT RESULTS
The ratings and parameters of the BLDCM are listed in
Table 1. Fig. 14 shows the speed control performance with a
wide speed control range (100 to 3000 rpm). From this figure,
the proposed sensorless algorithm has a wide-speed
regulation capability and good transient response under full-
speed operations. Fig. 15 shows the steady-state response at
60 rpm with the novel sensorless algorithm under the 0.75
p.u. load condition. The minimum controllable speed is
successfully improved with comparison to the back-EMF
measurement method. Fig. 16 shows the sensorless startup
performance from standstill to 300 rpm. A constant current-
controlled startup strategy is proposed in this study. Stable
and smooth startup performance can be achieved with the
proposed algorithm. Fig. 17 shows the full-speed sensorless
control performance. Fig. 18 shows the torque-speed
characteristics of the proposed sensorless BLDCM drive. The
controllable speed ratio is 50:1, and the speed regulation
performance is 0.6%.
VI. CONCLUSIONS
This work has presented the design and analysis of a
sensorless drive for the BLDCM. The proposed sensorless
drive is implemented in a real-time motor drive system, and
is shown through extensive simulations and experiments to
be robust to system parameters and disturbances while
providing accurate rotor position estimation. The feature of
this sensorless control scheme can be summarized as follows.
First, the scheme does not rely on the motor having salient
poles, and it does not involve physical modification such as
placement of search coils. Second, the scheme is
computationally less intensive than many other schemes
proposed. The most important feature is that it does not
require the knowledge of the mechanical load or the moment
of inertia. Only three motor electrical parameters are required
for the estimation: the stator resistance, the stator inductance
and the back-EMF constant. The parameter variations in the
sensorless algorithm can be successfully corrected with the
internal closed-loop. The general property of this sensorless
approach has also been demonstrated with simulations for a
sinusoidal-type PMSM. Various implementation issues are
discussed, including the acquisition of PWM current in a
pulse width modulated inverter system and the quantization
effect of the digital implementation. Intensive experimental
results were given to demonstrate the sensorless control
performance. Accurate rotor position and speed estimation,
robustness to parameter uncertainty, and good disturbance
rejection, can be obtained.
In summary, the results of this study suggest that the
developed sensorless strategy is a viable alternative for
electric drive applications that require the high efficiency,
high performance, and robust sensorless BLDCM drives. The
sensorless drive design, experimental results and
implementation issues addressed in this study will provide
valuable information for any future sensorless drive designs.
On the other hand, the presented fuzzed optimization
technique for control parameter tuning is proved the potential
for intelligent servo drives.

9

Fig. 14. Experimental result of sensorless speed control performance over a
wide speed range with the proposed sensorless commutation control.

Fig. 15. Steady-state response at 60 rpm with the rotor position estimation
algorithm.

Fig. 16. Sensorless startup performance fromstandstill to 300 rpm.

Fig. 17. Full-speed sensorless control performance.
0 0.2 0.4 0.6 0.8 1
0
500
1000
1500
2000
2500
3000
Torque-Speed Characteristics
Torque (p.u.)
S
p
e
e
d

(
r
p
m
)
100
60 30

Fig. 18. Torque-speed characteristics of the proposed sensorless algorithm
for a BLDCM drive.
TABLE 1
RATINGS AND PARAMETERS OF BLDCM DRIVE
3-phase brushless DC motor
Type Y-connection, 4 poles
Rated power 400 W
Rated speed 2400 rpm
Rated voltage 220 V
Rated stator current 3.2 A
Stator resistance 2 O
Stator inductance 8 mH
Torque constant 0.5 N-m/A
REFERENCES
[1] D. W. Novotny and T. A. Lipo, Vector Control and Dynamics of AC
Drives, New York: Oxford Univ. Press, 1996.
[2] P. Vas, Vector Control of AC Machines, London, U.K.,: Oxford Univ.
Press, 1990.
[3] T. M. J ahns, Motion control with permanent-magnet AC machines,
IEEE Proc., vol. 82, no. 8, pp. 1241-1252, Aug. 1994.
[4] R. Krishnan, Electric Motor Drives: Modeling, Analysis, and Control.
Upper Saddle River, NJ : Prentice Hall, 2001.
[5] K. Rajashekara, A. Kawamura, and K. Matsuse, (Editors) Sensorless
Control of AC Motor Drive, IEEE Press, 1996.
[6] P. Vas, Sensorless Vector and Direct Torque Control, Clarendon Press,
Oxford. 1998.
[7] J . P. J ohnson, M. Ehsani, and Y. Guzelgunler, Review of sensorless
methods for brushless DC, IEEE IAS Annual Meeting Conf. Rec., vol.
2, pp. 1033-1040, 1999.
[8] K. Iizaka, H. Uzuhashi, M. Kano, T. Endo, and K. Mohri,
Microcomputer control for sensorless brushless DC motor, IEEE
Trans. Ind. Applicat., vol. IA-21, no. 4, pp. 595-601, May/J une 1985.
[9] T. M. J ahns, R. C. Becerra, and M. Ehsani, Integrated current
regulation for a brushless ECM drive, IEEE Trans. Power Electron.,
vol. 6, no. 1, pp. 118-126, J an. 1991.
[10] R. C. Becerra, T. M. J ahns, and M. Ehsani, Four-quadrant sensorless
brushless ECM drive, in Proc. IEEE-APEC Conf., 1991, pp. 202-209.
[11] D. H. J ung and I. J . Ha, Low-cost sensorless control of brushless DC
motors using frequency-independent phase shifter, IEEE Trans.
Power Electron., vol. 15, no. 4, pp. 744-752, J uly 2000.
[12] H. C. Chen and C. M. Liaw, Sensorless control via intelligent
commutation tuning for brushless DC motor, IEE Proc. Electron.
Power Applicat., vol. 146, no. 6, pp. 678-684, Nov. 1999.
[13] H. C. Chen and C. M. Liaw, Current-mode control for sensorless
BDCM drive with intelligent commutation tuning, IEEE Trans.
Power Electron., vol. 17, no. 5, pp. 747-756, Sept. 2002.
[14] J . X. Shen and K. J . Tseng, Analyses and compensation of rotor
position detection error in sensorless PM brushless DC motor drives,
IEEE Trans. Energy Conver., vol. 18, no. 1, pp. 87-93, March 2003.
[15] K. Y. Cheng, Y. T. Lin, C. H. Tso, and Y. Y. Tzou, Design of a

10
sensorless commutation IC for BLDC motors, in Proc. IEEE-PESC
Conf., 2002, pp. 295-300.
[16] S. Ogasawara and H. Akagi, An approach to position sensorless drive
for brushless DC motor, IEEE Trans. Ind. Applicat., vol. IA-27, no. 5,
pp. 928-933, Oct. 1991.
[17] J . C. Moreira, Indirect sensing for rotor flux position of permanent
magnet AC motors operating in a wide speed range, IEEE Trans. Ind.
Applicat., vol. 32, no. 6, pp. 1394-1401, Nov./Dec. 1996.
[18] R. Wu and G. R. Slemon, A permanent magnet motor drive without a
shaft sensor, IEEE Trans. Ind. Applicat., vol. 27, no. 5, pp. 1005-
1011, Sept./Oct. 1991.
[19] T. H. Liu and C. P. Cheng, Adaptive control for a sensorless
permanent-magnet synchronous motor drive, IEEE Trans. Aerospace
and Electron. Syst., vol. 30, no. 3, pp. 900-909, J uly 1994.
[20] J . J iang and J . Holtz, High dynamic speed sensorless AC drive with
on-line model parameter tuning for steady-state accuracy, IEEE
Trans. Ind. Electron., vol. 44, no. 2, pp. 240-246, April 1997.
[21] N. Ertugrul and P. Acarnley, A new algorithm for sensorless
operation of permanent magnet motors, IEEE Trans. Ind. Applicat.,
vol. 30, no. 1, pp. 126-133, J an./Feb. 1994.
[22] C. French and P. Acarnley, Control of permanent magnet motor
drives using a new position estimation technique, IEEE Trans. Ind.
Applicat., vol. 32, no. 5, pp. 1089-1097, Sept./Oct. 1996.
[23] N. Ertugrul, and P. Acarnley, Indirect rotor position sensing in real
time for brushless permanent magnet motor drives, IEEE Trans.
Power Electron., vol. 13, no. 4, pp. 608-616, J uly 1998.
[24] Y. Li and N. Ertugrul, A novel, robust DSP-based indirect rotor
position estimation for permanent magnet AC motors without rotor
saliency, IEEE Trans. Power Electron., vol. 18, no. 2, pp. 539-546,
March 2003.
[25] S. Ostlund and M. Brokemper, Sensorless rotor-position detection
fromzero to rated speed for an integrated PM synchronous motor
drive, IEEE Trans. Ind. Applicat., vol. 32, no. 5, pp. 1158-1165,
Sept./Oct. 1996.
[26] J . Hu, and B. Wu, New integration algorithms for estimating motor
flux over a wide speed range, IEEE Trans. Power Electron., vol. 13,
no. 5, pp. 969-977, Sept. 1998.
[27] A. Consoli, S. Musumeci, A. Raciti, and A. Testa, Sensorless vector
and speed control of brushless motor drives, IEEE Trans. Ind.
Electron., vol. 41, no. 1, pp. 91-96, Feb. 1994.
[28] J . S. Kimand S. K. Sul, New approach for high-performance PMSM
drives without rotational position sensors, IEEE Trans. Power
Electron., vol. 12, no. 5, pp. 904-911, Sept. 1997.
[29] L. A. J ones and J . H. Lang, A state observer for the permanent-
magnet synchronous motor, IEEE Trans. Ind. Electron., vol. 36, no.
3, pp. 374-382, Aug. 1989.
[30] R. B. Sepe and J . H. Lang, Real-time observer-based (adaptive)
control of a permanent-magnet synchronous motor without
mechanical sensors, IEEE Trans. Ind. Applicat., vol. 28, no. 6, pp.
1345-1352, Nov./Dec. 1992.
[31] J . S. Kim and S. K. Sul, High-performance PMSM drives without
rotational position sensors using reduced order observer, IEEE IAS
Annual Meeting Conf. Rec., pp. 75-82, 1995.
[32] J . Solsona, M. I. Valla, and C. Muravchik, A nonlinear reduced order
observer for permanent-magnet synchronous motors, IEEE Trans.
Ind. Electron., vol. 43, no. 4, pp. 492-497, Aug. 1996.
[33] Z. Guchuan, A. Kaddouri, L. A. Dessaint, and O. Akhrif, A nonlinear
state observer for the sensorless control of a permanent-magnet AC
machine, IEEE Trans. Ind. Electron., vol. 48, no. 6, pp. 1098-1108,
Dec. 2001.
[34] R. Dhaouadi, N. Mohan, and I. Norum, Design and implementation
of an extended Kalman filter for the state estimation of a permanent
magnet synchronous motor, IEEE Trans. Power Electron., vol. 6, no.
3, pp. 491-497, J uly 1991.
[35] S. Bolognani, R. Oboe, and M. Zigliotto, Sensorless full-digital
PMSM drive with EKF estimation of speed and rotor position, IEEE
Trans. Ind. Electron., vol. 46, no. 1, pp. 184-191, Feb. 1999.
[36] B. Terzic and M. J adric, Design and implementation of the extended
Kalman filter for the speed and rotor position estimation of brushless
DC motor, IEEE Trans. Ind. Electron., vol. 48, no. 6, pp. 1065-1073,
Dec. 2001.
[37] S. Bolognani, M. Zigliotto, and M. Zordan, Extended-range PMSM
sensorless speed drive based on stochastic filtering, IEEE Trans.
Power Electron., vol. 16, no. 1, pp. 110-117, J an. 2001.
[38] A. B. Kulkarni and M. Ehsani, A novel position sensor elimination
technique for the interior permanent-magnet synchronous motor
drive, IEEE Trans. Ind. Applicat., vol. 28, no. 1, pp. 144-150,
J an./Feb. 1992.
[39] R. Mizutani, T. Takeshita, and N. Matsui, Current model-based
sensorless drives of salient-pole PMSM at low speed and standstill,
IEEE Trans. Ind. Applicat., vol. 34, no. 4, pp. 841-846, J uly/Aug.
1998.
[40] M. J . Corley and R. D. Lorenz, Rotor position and velocity estimation
for a salient-pole permanent magnet synchronous machine at standstill
and high speeds, IEEE Trans. Ind. Applicat., vol. 34, no. 4, pp. 784-
789, J uly/Aug. 1998.
[41] M. W. Degner and R. D. Lorenz, Using multiple saliencies for the
estimation of flux, position, and velocity in AC machines, IEEE
Trans. Ind. Applicat., vol. 34, no. 5, pp. 1097-1104, Sept./Oct. 1998.
[42] S. Ogasawara and H. Akagi, An approach to real-time position
estimation at zero and low speed for a PM motor based on saliency,
IEEE Trans. Ind. Applicat., vol. 34, no. 1, pp. 163-168, J an./Feb. 1998.
[43] S. Ogasawara and H. Akagi, Implementation and position control
performance of a position-sensorless IPM motor drive system based on
magnetic saliency, IEEE Trans. Ind. Applicat., vol. 34, no. 4, pp. 806-
812, J uly/Aug. 1998.
[44] S. Ogasawara and H. Akagi, Rotor position estimation based on
magnetic saliency of an IPM motor, IEEE IAS Annual Meeting Conf.
Rec., pp. 460-466, 1998.
[45] T. Aihara, A. Toba, T. Yanase, A. Mashimo, and K. Endo, Sensorless
torque control of salient-pole synchronous motor at zero-speed
operation, IEEE Trans. Power Electron., vol. 14, no. 1, pp. 202-208,
J an. 1999.
[46] N. Kasa and H. Watanabe, A mechanical sensorless control system
for salient-pole brushless DC motor with autocalibration of estimated
position angles, IEEE Trans. Ind. Electron., vol. 47, no. 2, pp. 389-
395, April 2000.
[47] S. Morimoto, K. Kawamoto, M. Sanada, and Y. Takeda, Sensorless
control strategy for salient-pole PMSM based on extended EMF in
rotating reference frame, IEEE Trans. Ind. Applicat., vol. 38, no. 4,
pp. 1054-1061, J uly/Aug. 2002.
[48] S. J . Kang and S. K. Sul, Direct torque control of brushless DC
motor with nonideal trapezoidal back EMF, IEEE Trans. Power
Electron., vol. 10, no. 6, pp. 796-802, Nov. 1995.
[49] C. S. Beredsen, G. Champenois, and A. Bolopion, Commutation
strategies for brushless DC motors: Influence on instant torque, IEEE
Trans. Power Electron., vol. 8, no. 2, pp. 231-236, April 1993.
[50] T. M. J ahns, Torque production in permanent magnet synchronous
motor drives with rectangular current excitation, IEEE Trans. Ind.
Applicat., vol. 20, no. 3, pp. 803-813, J uly/Aug. 1984.
[51] Carlos M. Fonseca and Peter J . Fleming, Multiobjective optimization
and multiple constraint handling with evolutionary algorithms-part I:
A unified formulation, IEEE Trans. Syst., Man, and Cybern., vol. 28,
pp. 26-37, J an. 1998.
[52] Edwin K. P. Chong and Stanislaw H. Zak, An Introduction to
Optimization. New York: Wiley, 1996.
[53] J . S. R. J ang, C. T. Sun, and E. Mizutani, Neuro-Fuzzy and Soft
Computing. Englewood Cliffs, NJ : Prentice Hall, 1997.
[54] F. Karray, W. Gueaieb, and S. Al-Sharhan, The hierarchical expert
tuning of PID controllers using tools of soft computing, IEEE Trans.
Syst., Man, and Cybern., vol. 32, pp. 77-90, Feb. 2002.

You might also like