Towards a global control strategy for induction motor:

Speed regulation, ux optimization and power factor correction

A. El Fadili

, F. Giri, A. El Magri, R. Lajouad, F.Z. Chaoui

GREYC Lab., University of Caen, Caen, France
a r t i c l e i n f o
Article history:
Received 12 July 2011
Received in revised form 11 May 2012
Accepted 14 May 2012
Available online 21 June 2012
Keywords:
Induction machine
Magnetic circuit nonlinearity
AC/DC/AC converters
Speed regulation
Power factor correction
Backstepping technique
a b s t r a c t
A great deal of interest has been paid to induction machine control over the last years. However, most
previous works have focused on the speed/ux/torque regulation supposing the machine magnetic cir-
cuit to be linear and ignoring the machine power conversion equipments. The point is that speed regu-
lation cannot be ensured in optimal efciency conditions, for a wide range of speed-set-point and load
torque, unless the magnetic circuit nonlinearity is explicitly accounted for in the motor model. On the
other hand, the negligence of the power conversion equipments makes it impossible to deal properly
with the harmonic pollution issue due to motor power supply grid interaction. This paper presents
a theoretical framework for a global control strategy of the induction machine and related power equip-
ments. The proposed strategy involves a multi-loop nonlinear adaptive controller designed to meet the
three main control objectives, i.e. tight speed regulation for a wide range speed-reference variation, ux
optimization for energy consumption and power factor correction (PFC). Tools from the averaging theory
are resorted to formally describe the control performances.
Crown Copyright 2012 Published by Elsevier Ltd. All rights reserved.
1. Introduction
Induction motors are featured by their interesting power/mass
ratio, relatively low cost and simple maintenance (as they include
no mechanical commutator). It is widely agreed that these ma-
chines have promising perspectives in the industrial actuator eld.
This has motivated an intensive research activity on induction ma-
chine control especially over the last fteen years. The complexity
of this problem is threefold:
The multivariable and nonlinear nature of the machine
dynamics.
The multiform interaction with its environment: supply grid,
power converters, varying load, etc.
The multiplicity of control objectives: speed regulation, energy
consumption optimization, power factor correction, fault detec-
tion and diagnostic, etc.
Most previous works have focused only on speed/ux regula-
tion (with constant ux reference) following several control strat-
egies ranging from simple techniques, e.g. eld-oriented control
[11,14], to more sophisticated nonlinear approaches, e.g. feedback
linearization [1], direct torque control [21,4] or sliding mode
control [19]. A common point of these works is that the control de-
sign relies on a relatively simple machine model, next called stan-
dard model, assuming a linear representation of the magnetic
circuit (which of course is not true in real machines). Accordingly,
most previous control solutions involved ux regulation around
constant values (Fig. 1). Specically, the constant ux reference is
equal to its nominal value generally located at the elbow of the ma-
chine magnetic characteristic [12,2]. Then, energetic efciency is
actually maximal provided that the machine operates all time in
the neighborhood of its nominal point. But, this is not the case in
most practical applications because the machine load is generally
varying [8]. Indeed, in presence of small loads, the operation point
is below the nominal value causing useless energy stored in stator
inductances which reduces the machine efciency. In the case of
overloaded machine, this operates in the saturation zone of its
magnetic characteristic but, then, the standard model is no longer
valid and, consequently, the control performances are no longer
guaranteed. To overcome the above shortcomings, it is necessary
in speed control to let the ux reference be dependent on both
the speed reference and torque-load, i.e. the ux reference must
be state-dependent. Examples of speed/ux controllers involving
state-dependent ux reference (Fig. 2) have been developed in
[15]. The proposed controllers include optimal ux generators
the design of which relies on a machine model that takes into ac-
count the nonlinearity of the magnetic characteristic.
However, even in the preceding works the control problem has
been relatively simplied because the motor is viewed there as a
0142-0615/$ - see front matter Crown Copyright 2012 Published by Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ijepes.2012.05.022

Corresponding author.
E-mail address: elfadili_abderrahim@yahoo.fr (A. El Fadili).
Electrical Power and Energy Systems 43 (2012) 230244
Contents lists available at SciVerse ScienceDirect
Electrical Power and Energy Systems
j our nal homepage: www. el sevi er . com/ l ocat e/ i j epes
separated system directly controlled by acting on the stator volt-
ages. As a matter of fact, in practical applications, the motor is
physically controlled through a (three-phase) DC/AC PWM switch
inverter. Furthermore, the inverter is connected to a power supply
net through a AC/DC PWM rectier (Fig. 3). Ignoring the latter
amounts to suppose that the DC-link voltage (i.e. the AC/DC recti-
er output voltage) is perfectly regulated. The point is that perfect
regulation of such voltage cannot be ensured ignoring the rectier
load which is nothing other than the set DC/AC inverterMotor.
Moreover, the RectierInverterMotor association strongly inter-
acts with the AC power supply net (Fig. 3). Accordingly, the power
ow is in fact bidirectional: the circulation sense depends on the
load variation. Then, undesirable current harmonics are likely to
be generated in the AC line, due to the strongly nonlinear nature
of the association converterinvertermotor. This harmonic
pollution has several damaging effects on the quality of power
distribution along the AC line, e.g. electromagnetic compatibility
issues, voltage distortion, reactive power increase, larger power
losses, increased voltage drops, etc. In this respect, standards such as IEEE519-1992 and IEC 61000-3-2/IEC 6100-3-4 indicate the cur-
rent harmonic limits to be complied with, in terms of power factor
correction [9,18]. Of course, the power factor can be improved
using additional protection equipments (transformers, condensers,
etc.) and/or over-dimensioning the converter and net elements.
However, this solution is costly and may not be sufcient.
An attempt to simultaneously deal with speed/ux control and
PFC requirement has been done [16]. But, the ux optimization
requirement was not coped with there.
In the light of the above remarks, it becomes clear that a conve-
nient control strategy is one that consists in dealing with the con-
trol problem for the whole association rectierinvertermotor,
seeking simultaneous achievement of all relevant control objec-
tives, i.e. tight speed regulation for wide set-point variation range,
ux optimization despite large load change, satisfactory power
factor correction (PFC), fault detection and diagnostic, etc. Such a
global control strategy has still to be developed (Fig. 4). This study
Nomenclature
c
i
; d design parameters
f
e
voltage network frequency (Hz)
i
sd
; i
sq
d- and q-axis stator currents
i
e
rectier input current
k control action of the DC link voltage regulator
R
s
stator resistance
L
1
; 2C passive components of input converter
s PWM input signal controlling converter IGBTs
T
L
machine load torque
v
sa
; v
sb
a- and b-axis stator voltages
v
dcref
reference value of rectier output voltage v
dc
V
i
Lyapunov functions i 1. . . 6
~x the estimate error of x
x
1
average rectier input current, i.e. x
1

i
e
x
2
average rectier output voltage x
2
v
dc
x
4
average a-axis stator current x
4

i
sa

x
6
average a-axis rotor ux x
6


/
ra

y squared DC Link voltage y x

2
2
v
2
dc
z
1
input current tracking error z
1
x
1
kv
e
z
3
rotor speed tracking error z
3
X
ref
X
z
5
; z
6
interns errors
/
ra
; /
rb
a- and b-axis rotor ux
U
ref
average rotor ux reference U
ref
FI
s

X
ref
rotor speed reference (rd/s)
f combined rotor and load viscous friction
I
s
stator current norm (A)
i
sa
; i
sb
a- and b-axis stator currents
J combined rotor and load inertia
R
r
rotor resistance
L
seq
leakage equivalent inductance
p number of pole pairs
T
e
electromagnetic torque
u
i1;2;3
duty ratios
v
dc
rectier output voltage
v
e
t AC line voltage
^x the estimate of x
x average values over cutting periods of x
x

1
average input current reference x

1
kv
e
x
3
rotor speed x
3
X
x
5
average b-axis stator current x
5

i
sb

x
7
average b-axis rotor ux x
7


/
rb

y
ref
reference value of y, i.e. y
ref
v
2
dcref
z
2
DC Link voltage error z
2
y y
ref
z
4
rotor ux norm error z
4
U
2
ref
U
2
r
/
rd
; /
rq
d- and q-axis rotor ux
U
r
average rotor ux norm U
r

x
2
6
x
2
7
_
X machine rotor angular velocity
x
e
power supply net pulsation
a
1
R
r
a
2

RsRr
Lseq
; a
3

1
Lseq
; b
0

E
2
C
; b
1

dC
E
2
c
2
b
2

2
p
E
C
b
3

d
E
2
C;

d CU
r
; e 1=x
e
Fig. 1. Control strategy involving constant ux reference (the controller is obtained
from the standard model).
Fig. 2. Control strategy involving state-dependent optimal ux (SDOF) reference.
A. El Fadili et al. / Electrical Power and Energy Systems 43 (2012) 230244 231
contributes to this purpose by proposing a theoretical framework
for the system modeling, multi-objective control design and con-
trol performance analysis. Presently, the focus is made on three
control objectives, i.e. wide range speed regulation, ux reference
optimization and PFC. For space limitation, the issues of state
observation and fault detection/diagnostic are not discussed in
the present paper (Fig. 4). The rst issue has already been coped
with in [3] where a ux observer was developed based on the same
model as presently. The second issue has been dealt with in many
places, see, e.g. [20].
The control aspects dealt with in the present paper present the
following features:
System modeling. The induction machine is represented by an
experimentally validated model that accounts for the nonlin-
earity of the magnetic characteristic [13,22]. The power con-
verters (AC/DC converter, the DC/AC inverter) are described
by their respective averaged instantaneous models. There, sig-
nal averaging is resorted to cope with the binary nature of the
converter switch control signals.
Control design. An adaptive multivariable controller is designed,
using nonlinear techniques, e.g. backstepping control. It con-
sists of three main components:
(i) A nonlinear bi-variable regulator is designed for the AC/DC
rectier so that the coupling between the power supply net
and the rectier operates with a unitary power factor and
the connection between the rectier and the inverter oper-
ates with a constant DC-link voltage, despite changes of the
operation conditions.
(ii) An adaptive nonlinear bi-variable regulator is developed to
make the motor velocity track its varying reference value
and the rotor ux norm track to its optimal state-depen-
dent reference. Parameter adaptation is resorted to cope
with parameter uncertainty that characterizes especially
the mechanicals parameters (load torque, rotor inertia,
and friction coefcient).
(iii) An online ux reference generator is constructed that pro-
vides the speed/ux regulator with the optimal ux refer-
ence trajectory. Optimality amounts to minimize the
necessary motor torque reducing thus the stator current
consumption. Based on the above motor model, the optimal
ux reference generator is analytically expressed in func-
tion of state variables, especially the stator currents.
Control performance analysis. The performances of the whole
nonlinear adaptive controller are formally analyzed making
adequate use of relevant tools from Lyapunov stability and
averaging theory [6]. It will be demonstrated that all control
objectives are achieved with a good accuracy. More specically,
the motor speed and rotor ux norm both track perfectly their
references. Furthermore, the steady-state tracking errors asso-
ciated with the rectier input current and output voltage are
shown to be harmonic signals with amplitudes depending,
Fig. 3. The whole system under study: induction motor and associated AC/DC/AC converters.
Fig. 4. Global control strategy (the dotted parts not covered in the present paper).
232 A. El Fadili et al. / Electrical Power and Energy Systems 43 (2012) 230244
among others, on the supply net frequency. The larger the net
frequency is, the smaller the tracking error amplitudes. There-
fore, if the net frequency is large enough the PFC requirement
will actually be guaranteed up to a harmonic error of insigni-
cant amplitude. This formally establishes the existence of the
so-called ripples (usually observed in similar practical applica-
tions) and proves why this phenomenon is generally
insignicant.
The paper is organized as follows: in Section 2, the whole asso-
ciation including the AC/DC/AC power conversion and induction
motor is modeled taking into account magnetic saturation; the
multi-objective controller is designed and analyzed in Section 3;
the control performances are illustrated through numerical simu-
lations in Section 4.
2. Modeling of the convertermotor association
The controlled system, illustrated by Fig. 3, includes an AC/DC
rectier, on one hand, and a combination inverterinduction mo-
tor, on the other hand. The rectier interfaces the inverterinduc-
tion motor set to the power supply net. The inverter is a DC/AC
converter operating (like the AC/DC rectier) according to the well
known Pulse Wide Modulation (PWM) principle.
2.1. AC/DC rectier modeling
The AC/DC rectier is an H-bridge boost converter consisting of
four IGBTs with anti-parallel diodes for bidirectional power ow
mode. This is expected to accomplish two main tasks: (i) providing
a constant DC link voltage and (ii) ensuring an almost unitary
power factor connection (PFC) to the power supply grid. Applying
Kirchhoffs laws, the rectier can be described by the following set
of differential equations:
di
e
dt

v
e
L
1

1
L
1
sv
dc
1a
dv
dc
dt

1
2C
si
e

1
2C
i
s
1b
where i
e
is the current in inductor L
1
; v
dc
denotes the voltage in
capacitor 2C; i
s
designates the input current inverter,
v
e

2
p
:E: cosx
e
t is the sinusoidal net voltage (with known con-
stants E; x
e
and s is the switch binary value:
s
1 if S is ON and S
0
is OFF
1 if S is OFF and S
0
is ON
_
1c
The above (instantaneous) model describes accurately the phys-
ical inverter. Then, it is based upon when constructing converter
simulators. However, it is not suitable for control design due to
the binary nature of the control input s. As a matter of fact, most
existing nonlinear control approaches apply to systems with con-
tinuous control inputs. Therefore, control design for the converter
will be performed using the following average version of the model
(1a) and (1b) [5,17]:
dx
1
dt

v
e
L
1

1
L
1
u
1
x
2
2a
dx
2
dt

1
2C
u
1
x
1

1
2C

i
s
2b
where:
x
1

i
e
; x
2
v
dc
; u
1
s 2c
are the average values over cutting periods of i
e
; v
dc
and s,
respectively.
Remark 1. In converter modeling the dead time between the two
switches in one leg is ignored for this study and not considered. In
other hand the standard three legs rectiers are most used in
industry. But, in traction and, more generally in the case of single-
phase supply, H-bridge rectiers are more suitable and most used.
2.2. Invertermotor modeling
As mentioned in Section 1, the achievement of speed regulation
and ux optimization in presence of wide range load variation
necessitates that the control design is based on a model that takes
into consideration the nonlinear nature of the machine magnetic
circuit. Fortunately, examples of such models do exist and have
been proved to be useful in control design, e.g. [15,13,22]. The
model proposed in the last reference is presently preferred as, on
the one hand, this model is established in the xed ab-frame
and, on the other, it was experimentally validated. The experimen-
tal validation, performed on a 7.5 KW machine, has conrmed the
nonlinear nature of the corresponding magnetic characteristic
(Fig. 5). The model developed in [22,13] is dened by the following
physical equations:
dX
dt

f
J
X
p
J
/
ra
i
sb
/
rb
i
sa

T
L
J
3a
di
sa
dt
a
2
i
sa
d/
ra
a
3
pX/
rb
a
3
v
sa
3b
di
sb
dt
a
2
i
sb
a
3
pX/
ra
d/
rb
a
3
v
sb
3c
d/
ra
dt
a
1
i
sa
L
seq
d/
ra
pX/
rb
3d
d/
rb
dt
a
1
i
sb
L
seq
d/
rb
pX/
ra
3e
where the various notations are dened as follows:
d is a varying parameter depending on the machine magnetic
state as shown by Fig. 6; this dependence has been given a poly-
nomial approximation:
d CU
r

def
q
0
q
1
U
r
q
m
U
m
r
3f
where the involved coefcients have been identied (based on
spline approximation) using the experimental magnetic charac-
teristic of Fig. 5.
0 5 10 15 20
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Magnetic current (A)
R
o
t
o
r

f
l
u
x

c
u
r
r
e
n
t

(
W
b
)
Fig. 5. Magnetic characteristic experimentally built up in [13] for a 7.5 KW
induction motor: rotor ux norm Ur (Wb) versus magnetic current Il (A).
A. El Fadili et al. / Electrical Power and Energy Systems 43 (2012) 230244 233
U
r
denotes the amplitude of the (instantaneous) rotor ux,
denoted /
r
. Consequently, one has:
U
r

/
2
ra
/
2
rb
_
3g
where /
ra
; /
rb
denote the rotor ux ab-components.
i
sa
; i
sb
are the ab-components of the stator current and stator
voltage, respectively.
X represents the motor speed.
R
s
; R
r
denote the stator and rotor resistances.
f ; J; T
L
are respectively the friction coefcient, rotor inertia and
load torque; all three parameters are unknown.
p is the number of pole pairs.
L
seq
is the equivalent inductance of both stator and rotor leakage
brought to the stator side.
a
1
R
r
; a
2
R
s
R
r
=L
seq
; a
3
1=L
seq
.
The numerical values of the model parameters are given by Ta-
ble 1. As already mentioned, the numerical values correspond to an
induction motor of 7.5 kW.
The couple v
sa
; v
sb
denote the stator voltage in ab-coordinate
(Parks transformation of the three-phase stator voltages). The in-
verter is featured by the fact that the stator a- and b-voltage can
be controlled independently. To this end, these voltages are ex-
pressed in function of the corresponding control action (see e.g.
[10]):
v
sb
v
dc
u
3
4a
v
sa
v
dc
u
2
4b
where u
2
; u
3
represent the average a- and b-axis components (in
Parks transformation) of the three-phase duty ratio system
s
1
; s
2
; s
3
. The latter are dened by (1c) replacing there S and S
0
by S
i
and S
0
i
i 1; 2; 3. Now, let us introduce the state variables:
x
3
X; x
4

i
sa
; x
5

i
sb
; x
6


/
ra
; x
7


/
rb
4c
where x denotes the average value over cutting periods of x. Then,
the power absorbed by the DC/AC inverter takes the well known
expression P
ai
x
2

i
s
. On the other hand, the power released by
the inverter is given by P
rm
x
2
u
2
x
4
u
3
x
5
. Invoking the power
balance one has P
ai
P
rm
which yields:

i
s
u
2
x
4
u
3
x
5
4d
Substituting (4a)(4c) in (3a)(3g) gives the following state-
space representation of the association invertermotor:
dx
3
dt

f
J
x
3

p
J
x
6
x
5
x
7
x
4

T
L
J
5a
dx
4
dt
a
2
x
4
dx
6
a
3
px
3
x
7
a
3
u
2
x
2
5b
dx
5
dt
a
2
x
5
a
3
px
3
x
6
dx
7
a
3
u
3
x
2
5c
dx
6
dt
a
1
x
4
L
seq
dx
6
px
3
x
7
5d
dx
7
dt
a
1
x
5
L
seq
dx
7
px
3
x
6
5e
U
r

x
2
6
x
2
7
_
5f

d CU
r
q
0
q
1
U
r
q
m
U
m
r
5g
where U
r
and

d denote the average values over cutting periods of U
r
and d respectively. The equations obtained up to now are brought
together to form the state-space model of the whole system includ-
ing the AC/DC/AC converters combined with the induction motor:
dx
1
dt

ve
L
1

1
L
1
u
1
x
2
6a
dx
2
dt

1
2C
u
1
x
1

1
2C
u
2
x
4
u
3
x
5
6b
dx
3
dt

f
J
x
3

p
J
x
6
x
5
x
7
x
4

T
L
J
6c
dx
4
dt
a
2
x
4


dx
6
a
3
px
3
x
7
a
3
u
2
x
2
6d
dx
5
dt
a
2
x
5
a
3
px
3
x
6


dx
7
a
3
u
3
x
2
6e
dx
6
dt
a
1
x
4
L
seq

dx
6
px
3
x
7
6f
dx
7
dt
a
1
x
5
L
seq

dx
7
px
3
x
6
6g
U
r

x
2
6
x
2
7
_ _
_
6h

d CU
r
q
0
q
1
U
r
q
m
U
m
r
6i
Recall that the load torque T
L
, the rotor inertia J and the friction
coefcient f are all unknown parameters.
3. Controller design
3.1. Control objective
Based on the whole system model (6a)(6i), we aim at develop-
ing a controller that would be able to achieve the following objec-
tives, despite the uncertainty that prevails on the mechanical
parameters f ; J; T
L
:
(i) Speed regulation: the machine speed X must track, as clo-
sely as possible, a given time-varying reference signal X
ref
.
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
500
1000
1500
2000
2500
3000
3500
4000
Rotor flux norm (Wb)
D
e
l
t
a

Fig. 6. Characteristic d; Ur : directly computed points (++) and polynomial
interpolation (solid). Unities: dXH
2
; Ur (Wb).
Table 1
Numerical values of considered motor characteristics.
Characteristic Symbol Value Unit
Nominal power P
N
7.5 KW
Nominal voltage Usn 380 V
Nominal ux Urn 0.56 Wb
Stator resistance Rs 0.63 X
Rotor resistance Rr 0.52 X
Inertia moment J 0.22 kg m
2
Friction coefcient f 0.001
N m s rd
1
Number of pole pairs p 2
Leakage equivalent inductance Lseq 7 mH
234 A. El Fadili et al. / Electrical Power and Energy Systems 43 (2012) 230244
(ii) Flux optimization: the rotor ux norm U
r
must track as
accurately as possible a state-dependent ux reference
U
ref
FI
s
where I
s
denotes the stator current norm and
the function F: has yet to be determined so that U
r
U
ref
entails a minimal stator current consumption.
(iii) PFC requirement: the rectier input current i
e
must be sinu-
soidal and in phase or opposed phase with the AC supply
voltage v
e
.
As there are three control inputs at hand, namely u
1
; u
2
and
u
3
, we will seek an additional control objective:
(iv) Controlling the continuous voltage v
dc
making it track a
given reference signal v
dcref
. The latter is generally set to a
constant value equal to the nominal voltage entering the
inverter.
3.2. AC/DC rectier control design
3.2.1. Controlling rectier input current to meet PFC
The PFC objective means that the input current rectier should
be sinusoidal and in phase (or opposite phase) with the AC supply
voltage. We therefore seek a regulator that enforces the current x
1
to track a reference signal x

1
of the form:
x

1
k v
e
7
At this point k is any real parameter that is allowed to be
time-varying. Typically, k is positive when the induction machine
absorbs energy and negative when the machine restores energy
(generation mode). Introduce the current tracking error:
z
1
x
1
x

1
8
In view of (6a), the above error undergoes the following
equation:
_
z
1

v
e
L
1

1
L
1
u
1
x
2
_ x

1
9
To get a stabilizing control law for this rst-order system, con-
sider the quadratic Lyapunov function V
1
0:5z
2
1
. It can be easily
checked that the time-derivative
_
V
1
is a negative denite function
of z
1
if the control input is chosen to be:
u
1

L
1
c
1
z
1
v
e
=L
1
_ x

x
2
10
with c
1
> 0 a design parameter. The properties of such control law
are summarized in the following proposition.
Proposition 1. Consider the system, next called current- or inner-
loop, composed of the current Eq. (6a) and the control law (10) where
c
1
> 0 is arbitrarily chosen by the user. If the reference x

1
k v
e
and
its rst time derivative are available then one has the following
properties:
(1) The current loop undergoes the equation _ z
1
c
1
z
1
where
z
1
x
1
x

1
. As c
1
is positive this equation is globally exponen-
tially stable, i.e. z
1
vanishes exponentially, whatever the initial
conditions.
(2) If in addition k converges (to a nite value), then the PFC
requirement is asymptotically fullled in average, i.e. the (aver-
age) input current x
1
tends (exponentially fast) to its reference
kv
e
as t ! 1.
3.2.2. DC link voltage regulation
The aim is now to design a tuning law for the ratio k in (7) so
that the rectier output voltage x
2
v
dc
is regulated around a gi-
ven reference value v
dcref
. As mentioned above, v
dcref
is generally
(not mandatory) chosen to be constant equal to the nominal inver-
ter input voltage amplitude (i.e. nominal stator voltage).
a. Relationship between k and x
2
.
The rst step in designing such a loop is to establish the relation
between the ratio k (control input) and the output voltage x
2
. This
is the subject of the following proposition.
Proposition 2. Consider the power rectier described by (6a) and
(6b) together with the control law (10). Under the same assumptions
as in Proposition 1, the output voltage x
2
varies, in response to the
tuning ratio k, according to the equation:
dy
dt

1
C
kv
2
e

1
C
z
1
v
e
vx; t 11
with y x
2
2
and
vx; t
1
C
x
2
u
2
x
4
u
3
x
5
12
Proof. The power absorbed by the AC/DC rectier is given by the
well known expression P
absorbed
x
1
v
e
. On the other hand, the
power released by the rectier (toward the load including the
capacity and the inverter) is given by P
released
u
1
x
1
x
2
. Power bal-
ance entails P
absorbed
P
released
or, equivalently:
x
1
v
e
u
1
x
1
x
2
13
Also, from (7) and (8) one immediately gets that x
1
kv
e
z
1
which together with (13) yields u
1
x
1
kv
2
e
z
1
v
e
=x
2
. This,
together with (6b), gives:
dx
2
dt

1
2Cx
2
kv
2
e
z
1
v
e

1
2C
u
2
x
4
u
3
x
5
14
Now, deriving y x
2
2
with respect to time and using (14) yields
the relation (11) and ends the proof. h
b. Squared DC-link voltage regulation.
The ratio k stands up as a control signal in the rst-order system
dened by (11). As said before, the reference signal y
ref

def
v
2
dcref
(of
the squared DC-link voltage x
2
v
dc
is generally given as a con-
stant value, namely the nominal value of the inverter input voltage.
Then, it follows from (11) that the tracking error z
2
y y
ref
undergoes the following equation:
_
z
2

1
C
E
2
k
1
C
E
2
k cos2x
e
t

2
p
E
C
z
1
cosx
e
t vx; t _ y
ref
15
where we have used the fact that v
e
t

2
p
:E: cosx
e
t and
v
2
e
t E
2
1 cos2x
e
t. To get a stabilizing control law for the
system (15), consider the following quadratic Lyapunov function:
V
2

1
2
z
2
2
16
It is easily checked that the time-derivative
_
V
2
can be made
negative denite in the state z
2
by letting:
kE
2
kE
2
cos2x
e
t

2
p
E z
1
cosx
e
t
C c
2
z
2
vx; t C _ y
ref
17a
where c
2
> 0 is a design parameter. The point is that such equation
involves a periodic singularity due to the mutual neutralization of
A. El Fadili et al. / Electrical Power and Energy Systems 43 (2012) 230244 235
the rst two terms on the left side of (17a). To get off this singularity
and, besides, to avoid an excessive control chattering, we just ignore
the two terms in cos on the left side of (17a). Therefore, we con-
sider the following approximate and simple solution:
k
C
E
2
c
2
z
2
vx; t
C
E
2
_ y
ref
17b
Bearing in mind the fact that the rst derivative of the control
ratio k must be available (Proposition 1), the following ltered ver-
sion of the above solution is considered:
_
k dk d
C
E
2
c
2
z
2
vx; t d
C
E
2
_ y
ref
18
At this point, the regulator parameters, d and c
2
, are any posi-
tive real constants. The forthcoming development (especially the
proof of theorem 2) will make it clear how these parameters
should be chosen for the control objectives to be achieved. For
now, we summarize our main ndings in the following
proposition:
Proposition 3. Consider the control system consisting of the AC/DC
rectier described by (6a) and (6b) together with the control law
dened by (10) and (18). Using Proposition 1 (Part 1), it follows that
the resulting closed-loop undergoes, in the z
1
; z
2
; k-coordinates, the
following equation where z
1
x
1
x

1
and z
2
y y
ref
:
_
Z
1
A
1
Z
1
f Z
1
; t gvx; t h_ y

ref
19a
where
Z
1
z
1
z
2
k
T
19b
f Z
1
; t 0 b
0
k cos2x
e
t b
2
z
1
cosx
e
t 0
T
2 IR
3
19c
A
1

c
1
0 0
0 0 b
0
0 b
1
d
_
_
_
_
_
_; g 0 1 b
3

T
; h 0 1 b
3

T
19d
b
0

E
2
C
; b
1

dC
E
2
c
2
; b
2

2
p
E
C
; b
3

d
E
2
C 19e
Proof. The ltering (18) makes k and its derivative
_
k both avail-
able. Then, Part 1 of Proposition 1 ensures that _ z
1
c
1
z
1
. This,
together with (15) and (18), gives the state-space equation
(19a). h
3.3. Rotor ux reference optimization
Most existing induction machine speed controllers involve a
ux regulation loop (Fig. 1). Furthermore, this loop is generally de-
signed based upon a machine standard model (ignoring the ma-
chine magnetic saturation). Then, for coherency, the ux
reference signal is generally given a constant value coinciding with
the machine nominal ux value [11,14,1,21,19].
Of course, such controllers are unable to achieve optimal ma-
chine performances (power efciency, power factor, maximal tor-
que, etc.) in presence of small loads. In [15], new controllers
were proposed involving a state-dependent optimal ux reference.
Optimality was achieved in the sense of stator current consump-
tion, using models that account for the magnetic circuit saturation.
Presently, a similar ux reference optimality is sought but, unlike
the previous references, the machine mechanical parameters and
load torque are subject to uncertainty and the control strategy ac-
counts for the presence of power conversion elements (Fig. 4).
In this subsection, the aim is to nd a relationship between the
rotor ux norm and the stator current norm. In this respect, let us
recall that the norms of electrical quantities are invariant under
frame Park change. Then, it makes sense, for simplicity reasons,
to perform all this section computations in the Park oriented dq
reference frame. Indeed, within this reference frame, the ux
q-component is (or can be considered to be) null and all state
variables are constant in steady-state. Then, the machine electro-
magnetic torque T
e
is expressed as follows:
T
e
p/
rd
i
sq
pU
r
i
sq
20
On the other hand, Eq. (3d) becomes in the Park oriented dq
reference frame:
dU
r
dt
a
1
i
sd
L
seq
dU
r
21a
and that the steady-state current i
sd
can be given the following
expression:
i
sd

L
seq
a
1
dU
r
21b
In (20), (21a) and (21b), i
sd
and i
sq
denote the d- and q-compo-
nents of the stator current. Also, the stator current norm expres-
sion simplies to:
I
2
s
i
2
sd
i
2
sq
22
Then, using (20)(22), one gets the following expression of the
electromagnetic torque:
T
e
pU
r

I
2
s

L
seq
a
1
dU
r
_ _
2

23
Fig. 7 shows the curves representing the electromagnetic torque
T
e
versus the ux U
r
, for various values of the stator current I
s
. It is
clearly seen that for a given torque there are many operation
points that only differ by the value of the ux U
r
and the current
I
s
. For instance, a torque T
e
20 Nm can be produced, for the ma-
chine of Table 1, with:
A ux U
r
0:50 Wb and a current I
s
8:5 A.
A ux U
r
0:37 Wb and a current I
s
10 A.
A ux U
r
0:21 Wb and a current I
s
16 A.
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
0
10
20
30
40
50
60
Rotor flux norm (Wb)
E
l
e
c
t
r
o
m
a
g
n
e
t
i
c

t
o
r
q
u
e

(
N
m
)
Is=8.5 A
Is=10 A
Is=16 A
Fig. 7. Electromagnetic torque Te (Nm) in function of rotor ux Ur , for different
stator currents norm Is.
236 A. El Fadili et al. / Electrical Power and Energy Systems 43 (2012) 230244
From an energetic viewpoint, the best operation point is one
that involves the least current consumption. Let T
ei
i 1; . . . ; r
be a sufciently large sample of relevant torque values. It is readily
seen from Fig. 7 that for any T
ei
, there is a unique couple U
ri
; I
si

that involves the least possible stator current. A set of couples

U
ri
; I
si
has thus been obtained using (23) and interpolated to get
a polynomial function F such that U
ri
FI
si
. The polynomial
interpolation thus constructed is denoted:
FI
s
h
n
I
n
s
h
n1
I
n1
s
h
1
I

s
h
0
24
Such polynomial function is referred to optimal current-ux
(OCF) characteristic and is illustrated by Fig. 8. Recall that optimal-
ity is understood in the sense of minimal absorbed stator current I
s
for a given torque T
e
.
Remark 2. The polynomial interpolation yielding the function F
has been obtained the Matlab functions POLYDER, POLYVAL,
SPLINE, and POLYFIT.
3.4. Speed and ux adaptive control design and analysis
In this section, a speed and ux adaptive regulator is designed
for the saturated induction machine described by (6c)(6i). The
speed reference X
ref
is any bounded and derivable function of time
and its two rst derivatives are available and bounded. These
requirements can always be met by ltering the reference through
second-order linear lters. The ux reference U
ref
is obtained on-
line from the average value of the stator current so that, if
U
r
U
ref
with U
r

x
2
6
x
2
7
_ _ _
, the machine will operate in opti-
mal conditions (i.e. with minimal absorbed stator current). In the
light of Section 3.3, it follows that such optimality will be guaran-
teed if the ux reference is computed online as follows:
U
ref
FI
s
h
n
I
n
s
h
n1
I
n1
s
h
1
I
s
h
0
25
In fact, the relation (25) is nothing other than (24) replacing
there the variables by their average values (because the system
model (6a)(6i) involves average variables).
The adaptive speed/ux controller design will now be per-
formed in two steps using the backstepping technique [7]. First,
introduce the tracking errors:
z
3
X
ref
x
3
26a
z
4
U
2
ref
x
2
6
x
2
7
26b
Step 1 It follows from (5a), (5d), and (5e) that the errors z
3
and z
4
undergo the differential equations:
_
z
3

_
X
ref

p
J
x
6
x
5
x
7
x
4

T
L
J

f
J
x
3
27
_ z
4
2U
ref
_
U
ref
2x
6
_ x
6
2x
7
_ x
7
2U
ref
_
U
ref
2x
6
a
1
x
4
L
seq

dx
6
px
3
x
7
2x
7
a
1
x
5
L
seq

dx
7
px
3
x
6

2U
ref
_
U
ref
2a
1
x
6
x
4
x
7
x
5
2L
seq

dx
2
6
x
2
7
28a
_
z
4
2U
ref
_
U
ref
2a
1
x
6
x
4
x
7
x
5
2L
seq

dU
2
ref
z
4
28b
In (27) and (28), the quantities px
6
x
5
x
7
x
4
and
2a
1
x
6
x
4
x
7
x
5
stand up as virtual control signals. Let us tempo-
rarily suppose these to be the actual control signals and consider
the Lyapunov function candidate:
V
3

1
2
z
2
3
z
2
4
29
It can be easily checked that the time derivative of (29) can be
made negative denite function of z
3
; z
4
, i.e.
_
V
3
c
3
z
2
3
c
4
z
2
4
30
by letting px
6
x
5
x
7
x
4
l
1
and 2a
1
x
6
x
4
x
7
x
5
m
1
with:
l
1

def
Jc
3
z
3

_
X
ref
T
L
f X
ref
z
3
31a
and
m
1

def
c
4
z
4
2U
ref
_
U
ref
2L
seq

dU
2
ref
z
4
31b
where c
3
and c
4
are any positive design parameters. Since J; T
L
and f
are unknown, the rst equation in (31) is replaced by its certainty
equivalence from, yielding the following adaptive control laws:
l
1

def

Jc
3
z
3

_
X
ref


T
L

^
f X
ref
z
3
32a
m
1

def
c
4
z
4
2U
ref
_
U
ref
2L
seq

dU
2
ref
z
4
32b
and

J;

T
L
and
^
f are estimates (yet to be determined) of J; T
L
and f,
respectively. As the quantities px
6
x
5
x
7
x
4
and 2a
1
x
6
x
4
x
7
x
5

are not the actual control signals, they cannot be let equal to l
1
and m
1
, respectively. Nevertheless, we retain the expressions of l
1
and m
1
as the rst stabilizing functions and introduce the new
errors:
z
5
l
1
px
6
x
5
x
7
x
4
33
z
6
m
1
2a
1
x
6
x
4
x
7
x
5
34
Then, using the notations (29)(34), the dynamics of the errors
z
3
and z
4
, initially described by (27)(30), can be rewritten as
follows:
_
z
3

_
X
ref

1
J
l
1
z
5

T
L
J

f
J
x
3
_
z
3

_
X
ref

1
J

Jc
3
z
3

_
X
ref


T
L

^
f x
3
z
5
_ _

T
L
J

f
J
x
3
_
z
3
c
3
z
3

J
J
c
3
z
3

_
X
ref

T
L
J

~
f
J
x
3

1
J
z
5
35a
_
z
4
c
4
z
4
z
6
35b
where

J J

J;

T
L
T
L


T
L
and
~
f f
^
f 35c
0 10 20 30 40 50
0
0.8
Stator current norm (A)
r
o
r
o
r

f
l
u
x

n
o
r
m

r
e
f
e
r
e
n
c
e

(
W
b
)
0.7
0.6
0.5
0.4
0.3
0.2
0.1
Fig. 8. Optimal current-ux characteristic obtained from the polynomial interpo-
lation of the experimental points U

Te
j
; I

Te
j
_ _
for the induction machine with
physical characteristics of Table 1.
A. El Fadili et al. / Electrical Power and Energy Systems 43 (2012) 230244 237
Similarly, the time-derivative of V
3
can be expressed in function
of the new errors as follows:
_
V
3
c
3
z
2
3
c
4
z
2
4
z
4
z
6
z
3

J
J
c
3
z
3

_
X
ref

T
L
J

~
f
J
x
3

1
J
z
5
_ _
36
Step 2: The second design step consists in choosing the actual
control signals, u
2
and u
3
, so that all errors
z
3
; z
4
; z
5
; z
6
converge to zero. To this end, it must be
made clear how these errors depend on the actual control
signals u
2
; u
3
. First, focusing on z
5
, it follows from (33)
that:
_
z
5
_ l
1
p _ x
6
x
5
x
6
_ x
5
_ x
7
x
4
x
7
_ x
4
37
Using (5a)(5e), (35c) and (31), one gets from (37):
_
z
5


J c
3
_
z
3


X
ref
_ _

J c
3
z
3

_
X
ref
_ _

T
L

_
~
f x
3

^
f _ x
3
_ _
p _ x
6
x
5
_ x
7
x
4
p x
6
_ x
5
x
7
_ x
4

_
z
5
c
3

J
^
f c
3
z
3

J
J
c
3
z
3

_
X
ref

T
L
J

~
f
J
x
3

1
J
z
5
_ _

X
ref

^
f
_
X
ref
pa
1
x
4
L
seq

dx
6
px
3
x
7
x
5
a
1
x
5
L
seq

dx
7
px
3
x
6
x
4

_

Jc
3
z
3

_
X
ref

_

T
L

_
~
f x
3
_ _
px
6
a
2
x
5
a
3
px
3
x
6


dx
7
a
3
u
3
x
2
px
7
a
2
x
4

dx
6
a
3
px
3
x
7
a
3
u
2
x
2
38
For convenience, the above equation is given the following com-
pact form:
_ z
5
l
2

c
3

J
^
f
J
z
5
pa
3
x
2
x
7
u
2
x
6
u
3

c
3

J
^
f
_ _

J
J
c
3
z
3

_
X
ref

T
L
J

~
f
J
x
3
_ _

Jc
3
z
3

_
X
ref

_

T
L

_
~
f x
3
_ _
39
with
l
2
c
3
z
3
c
3

J
^
f
_ _

X
ref

^
f
_
X
ref
_ _
p
2
a
3
x
3
x
2
6
x
2
7

p L
seq

d a
2
_ _
x
6
x
5
x
7
x
4
p
2
x
3
x
7
x
5
x
6
x
4
40
Similarly, it follows from (34) that, z
6
undergoes the following
differential equation:
_
z
6
_ m
1
2a
1
_ x
6
x
4
x
6
_ x
4
_ x
7
x
5
x
7
_ x
5
41
Using (5a)(5e) and (32b), it follows from (41):
_
z
6
c
4
_
z
4
2U
ref

U
ref
2
_
U
2
ref
2L
seq

d 2U
ref
_
U
ref

_
z
4
_ _
2L
seq
_

d U
2
ref
z
4
_ _
2a
1
x
4
a
1
x
4
L
seq

dx
6
px
3
x
7
_ _
2a
1
x
6
a
2
x
4


dx
6
a
3
px
3
x
7
a
3
u
2
x
2
_ _
2a
1
x
5
a
1
x
5
L
seq

dx
7
px
3
x
6
_ _
2a
1
x
7
a
2
x
5
a
3
px
3
x
6


dx
7
a
3
u
3
x
2
_ _
42
where the derivative of

d is obtained from (6i):
_

d
d

d
dU
r
dU
r
dt

d

d
dU
r
x
6
U
r
_ x
6

x
7
U
r
_ x
7
_ _
43
Eq. (34) is in turn given the following compact form:
_
z
6
m
2
2a
1
a
3
x
2
x
6
u
2
x
7
u
3
44a
with
m
2
c
4
_
z
4
2U
ref

U
ref
2
_
U
2
ref
2L
seq

d 2U
ref
_
U
ref

_
z
4
_ _
2L
seq
_

d U
2
ref
z
4
_ _
2a
1
x
4
a
1
x
4
L
seq

dx
6
px
3
x
7
_ _
2a
1
x
6
a
2
x
4


dx
6
a
3
px
3
x
7
_ _
2a
1
a
1
x
5
L
seq

dx
7
px
3
x
6
_ _
x
5
2ax
7
a
2
x
5
a
3
px
3
x
6


dx
7
_ _
44b
m
2
c
4
2L
seq

d
_ _
c
4
z
4
z
6
2U
ref

U
ref
2
_
U
2
ref
4L
seq

dU
ref
_
U
ref
2L
seq
_

d U
2
ref
z
4
_ _
2a
1
x
3
px
7
x
4
x
5
x
6
2a
1

d U
2
ref
z
4
_ _
2a
1
L
seq

d a
2
_ _
x
4
x
6
x
5
x
7
2a
1

2
x
2
4
x
2
5
44c
To analyze the stability of the error system, composed of equa-
tions (35a), (35b), (39) and (44a), consider the following aug-
mented Lyapunov function candidate:
V
4

1
2
z
2
3

1
2
z
2
4

1
2
z
2
5

1
2
z
2
6

1
2

J
2
J

1
2

T
2
L
J

1
2
~
f
2
J
45
Its time-derivative along the trajectory of the state vector
z
3
; z
4
; z
5
; z
6
is:
_
V
4
z
3
_
z
3
z
4
_
z
4
z
5
_
z
5
z
6
_
z
6

J
_

J
J

~
f
_
~
f
J

T
L
_

T
L
J
46
Using (36), (39) and (44a), Eq. (46) gives:
_
V
4
z
3
c
3
z
3

J
J
c
3
z
3

_
X
ref
_ _

T
L
J

~
f
J
x
3

1
J
z
5
_ _
z
4
c
4
z
4
z
6

z
5
c
3

J
^
f
_ _

J
J
c
3
z
3

_
X
ref
_ _

T
L
J

~
f
J
x
3
_ _
z
5
l
2
c
3
z
5
pa
3
x
2
x
7
u
2
x
6
u
3
z
6
m
2
2a
1
x
2
a
3
x
7
u
3
x
6
u
2
z
5
c
3
z
5

J
J

f
J
z
5
z
5
~
f
J
_ _

J
J
_

T
L
J
_

T
L

~
f
J
_
~
f
z
5
_

Jc
3
z
3

_
X
ref

_

T
L

_
~
f x
3
_ _
47a
Adding c
5
z
2
5
c
5
z
2
5
c
6
z
2
6
c
6
z
2
6
to the right side of (47a) and
rearranging terms, yields:
_
V
4
c
3
z
2
3
c
4
z
2
4
c
5
z
2
5
c
6
z
2
6

f
J
z
2
5

1
J
z
3
z
5

J
J
_

J c
3
z
3

_
X
ref
_ _
z
3
z
5
c
3

J
^
f
_ _
c
3
z
3

_
X
ref
_ _
c
3
z
2
5
_ _

T
L
J
_

T
L
z
3
z
5
c
3

J
^
f
_ _ _ _

~
f
J
_
~
f z
3
x
3
z
5
x
3
c
3

J
^
f
_ _
z
2
5
_ _
z
5
l
2
c
5
c
3
z
5
pa
3
x
2
x
7
u
2
x
6
u
3

z
5
_

J c
3
z
3

_
X
ref
_ _

T
L

_
~
f x
3
_ _
z
6
z
4
m
2
c
6
z
6
2a
1
x
2
a
3
x
7
u
3
x
6
u
2
47b
which suggests the following parameter adaptation laws:
_

T
L
k
T
L
;
_

J k
J
and
_
~
f k
f
47c
with
k
J
c
3
z
2
5
z
3
c
3
z
3

_
X
ref
_ _
z
5
c
3

J
^
f
_ _
c
3
z
3

_
X
ref
_ _
47d
k
T
L
z
3
c
3

J
^
f
_ _
z
5
47e
k
f
z
3
x
3
z
5
x
3
c
3

J
^
f
_ _
z
2
5
47f
238 A. El Fadili et al. / Electrical Power and Energy Systems 43 (2012) 230244
Substituting the parameter adaptation laws (47c) to
_

T
L
;
_

J and
_
~
f
in the right side of (47b) yields:
_
V
4
c
3
z
2
3
c
4
z
2
4
c
5
z
2
5
c
6
z
2
6

f
J
z
2
5

1
J
z
3
z
5
z
5
l
2
c
5
c
3
z
5
pa
3
x
2
x
7
u
2
x
6
u
3

z
5
k
J
c
3
z
3

_
X
ref
k
T
L
k
f
x
3
_ _
z
6
m
2
z
4
c
6
z
6
2a
1
a
3
x
2
x
6
u
2
x
7
u
3
48
where c
5
and c
6
are new arbitrary positive real design parameters.
Eq. (48) suggests that the control signals u
2
; u
3
should be chosen
in order to set to zero the two quantities between curly brackets
(on the right side of (48)). Letting these quantities equal to zero
and solving the resulting second-order linear equation system with
respect to u
2
; u
3
, gives the following control law:
u
2
u
3
_ _

k
0
k
1
k
2
k
3
_ _
1
U
A
U
B
_ _
49
with:
U
A
k
J
c
3
z
3

_
X
ref
k
T
L
k
f
x
3
c
5
c
3
z
5
l
2
U
B
z
4
c
6
z
6
m
2
; k
0
pa
3
x
7
x
2
; k
1
pa
3
x
6
x
2
;
k
2
2a
1
a
3
x
6
x
2
; k
3
2a
1
a
3
x
7
x
2
50
Note that inversed matrix in (49) is nonsingular in practice
because its determinant is given by D k
0
k
3
k
1
k
2

2pa
2
1
a
3
x
2
2
x
2
6
x
2
7
and

x
2
6
x
2
7
_
U
r
never vanish in practice be-
cause of the machine nonzero remnant ux and the small varia-
tions of x
2
v
dc
with respect to its high nominal value 600 V.
Finally, supposing the unknown parameters T
L
; J; f be constant
values, one gets from (35c) and (47c)(47f), the following parame-
ter adaptive laws:
_

J c
3
z
2
5
z
3
c
3
z
3

_
X
ref
_ _
z
5
c
3

J
^
f
_ _
c
3
z
3

_
X
ref
_ _
51
_

T
L
z
3
c
3

J
^
f
_ _
z
5
52
_
^
f z
3
x
3
z
5
x
3
c
3

J
^
f
_ _
z
2
5
53
The properties of the speed/ux regulator thus designed are de-
scribed in the following proposition:
Theorem 1 (Speed/ux regulator). Consider the closed-loop system
composed of the induction machine described by the model (5),
supposing there the unknown parameters T
L
; J; f to be constant,
and the nonlinear adaptive regulator dened by the control law (49),
the parameter update laws (51)(53) and the ux generator (25).
Then, one has the following properties:
(1) The closed-loop error system undergoes the following equa-
tions, in the z
3
; z
4
; z
5
; z
6
coordinates:
_
z
3
c
3
z
3

1
J
z
5
f
3
z
3
; z
5
;

H; t
_ _
54a
_ z
4
c
4
z
4
z
6
54b
_
z
5
c
5

f
J
_ _
z
5
f
5
z
3
; z
5
;

H; t
_ _
54c
_
z
6
c
6
z
6
z
4
54d
These are given the more compact form:
_
Z
2
A
2
Z
2
v
2
Z
2
;

H; t
_ _
55
with:
Z
2
z
3
z
4
z
5
z
6

T
;

H

J

T
L
~
f
_ _
T
56a
A
2

c
3
0 1=J 0
0 c
4
0 1
0 0 c
5
f =J 0
0 1 0 c
6
_

_
_

_
; 56b
v
2
Z
2
;

H; t
_ _
f
3
Z
2

H; t
_ _
0 f
5
Z
2
;

H; t
_ _
0
_ _
T
56c
f
3
z
3
; z
5
;

H; t
_ _

def

J
J
c
3
z
3

_
X
ref
_ _

T
L
J

~
f
J
X
ref
z
3
56d
f
5
z
3
; z
5
;

H; t
def

J
J
c
3

J
^
f
_ _
c
3
z
3

_
X
ref
_ _
c
3
z
5
_ _
c
3

J
^
f
_ _

T
L
J

~
f
J
c
3

J
^
f
_ _
x
3
z
5
_ _
56e
(2) Let the design parameters c
3
and c
5
be sufciently large in the
sense that c
3
>
1
2J
and c
5
>
1
2J

f
J
. Then, all errors
z
3
; z
4
; z
5
; z
6
;

J;

T
L
;
~
f
_ _
remain bounded and z
3
; z
4
; z
5
; z
6

vanish asymptotically, whatever the initial conditions.

Proof. Part 1. Eq. (54a) and (54b) are immediately obtained from
(35a) and (35b). Eq. (54c) is obtained substituting the control law
(49) to u
2
; u
3
on the right side of (39). Eq. (54d) is obtained
substituting the control law (49) to u
2
; u
3
on the right side of
(44a). This proves Part 1.
Part 2. Substituting the control law (49) to u
2
; u
3
on the right
side of (48) yields:
_
V
4
c
3
z
2
3
c
4
z
2
4
c
5
z
2
5
c
6
z
2
6

f
J
z
2
5

1
J
z
3
z
5
57a
Using the inequality jz
3
z
5
j 6
1
2
z
2
3

1
2
z
2
5
, one obtains from (57a):
_
V
4
6 c
3

1
2J
_ _
z
2
3
c
4
z
2
4
c
5

f
J

1
2J
_ _
z
2
5
c
6
z
2
6
57b
As c
3

1
2J
and c
5

f
J

1
2J
are both positive, due to the assumption
in Part 2, it follows from (57b) that
_
V
4
< 0 whenever the state vec-
tor z
3
; z
4
; z
5
; z
6
is non-zero. Then, applying the Lasalles invariant
set principle, it follows that V
4
is bounded and z
3
; z
4
; z
5
; z
6
con-
verges to zero whatever the initial conditions.
Theorem 1 is established. h
Remark 3. The derivatives
_
U
ref
and

U
ref
are obtained using the
relation:
U
ref
F I
s
_ _
h
0
h
1
I
s
h
2
I
2
s
h
n
I
n
s
58a
where
I
s

x
2
4
x
2
5
_
58b
Specically, one has:
_
U
ref

dF I
s
_ _
dt

dF I
s
_ _
dI
s
dI
s
dt

dF I
s
_ _
dI
s
x
4
_ x
4
x
5
_ x
5
I
s
59a

U
ref

d
2
F I
s
_ _
dI
2
s
x
4
_ x
4
x
5
_ x
5
I
s
_ _
2

dF I
s
_ _
dI
s

_ x
4

2
x
4
x
4
_ x
5

2
x
5
x
5
I
s

dF I
s
_ _
dI
s
x
4
_ x
4
x
5
_ x
5

2
I
3
s
59b
A. El Fadili et al. / Electrical Power and Energy Systems 43 (2012) 230244 239
3.5. Achievement of PFC and DC link voltage regulation
In this section, it will be formally shown that, for a specic class
of reference signals, including periodic signals, the control objec-
tives are achieved (in the mean) with an accuracy that depends,
among others, on the network frequency x
e
. The analysis is based
on Eq. (19a) which, in view of (49) becomes:
_
Z
1
A
1
Z
1
f Z
1
; t qZ
2
; tg h_ y

ref
60
with:
qZ
2
; t
k
J
c
3
z
3

_
X
ref
k
T
L
k
f
X
ref
z
3

_ _
z
5
l
1

a
1
p
2
U
2
r

c
5
c
3
z
5
l
2
a
1
p
2
U
2
r
z
5
l
1

z
4
c
6
z

6
m
2
m
1
z
6

4a
3
1
U
2
r
61
Theorem 2 (PFC and DC-link regulation). Consider the system
including an AC/DC power rectier, a DC/AC inverter and an induction
motor, connected as shown in Fig. 3. For control design purpose, the
system is represented by its average model (6a)(6e). Consider the
controller dened by the control laws (10), (18) and (49), the
parameter update laws (51)(53) and the optimal ux generator
(25). Let the reference signals v
dcref
; X
ref
and U
ref
be selected such that
v
dcref
> 0; X
ref
P0 and U
ref
P0 and v
dcref
and X
ref
, be periodic with
period Np=x
e
(for some positive integer N), and suppose them to be
time derivable (up to second order for X
ref
) with bounded derivatives.
Finally, suppose the design parameters c
1
; c
2
; d to be positive. Then,
there exists a positive real e

such that, if 0 < e < e

, with e 1=x
e
,
then:
(1) The tracking error z
2
y y
ref
and the tuning parameter k are
harmonic signals continuously depending on e.
(2) Furthermore, one has:
i lim
e!0
z
2
t; e 0 and iilim
e!0
kt; e
1
C
q0
4

b
0
62
where q0
4
denotes the mean value of the periodic time function
q0
4
; t and 0
4
2 IR
4
denotes the null vector
Proof. As v
dcref
and X
ref
as well as their derivatives are constant or
periodic (with period Np=x
e
, it follows that the system (60) is
periodically time-varying. Therefore, the averaging theory turns
out to be a suitable framework to analyze its stability (see e.g.
[6]). To this end, introduce the time-scale change s x
e
t and the
following signal changes:
W
1
s Z
1
t; y

ref
t y
ref
Nt
2x
e
_ _
; X

ref
t X
ref
Nt
2x
e
_ _
63a
This readily implies that y

ref
and X

ref
are in turn periodic, with
period 2p, and:
y
ref
t y

ref
2s
N
_ _
; X
ref
t X

ref
2s
N
_ _
: 63b
Also, it is easily seen that
_
W
1
s dW
1
s=ds edZ
1
t=dt
e
_
Z
1
t with e 1=x
e
. Then, using (63b) it follows from (60) that
the state vector W
1
undergoes the following state equation:
_
W
1
eA
1
W
1
ef
1
W
1
; s; e egq
1
W
2
; s; e h_ y

ref
2s
N
_ _
eAW
1
ef
1
W
1
; s; e egq
1
0; s; e h_ y

ref
2s
N
_ _
eg~ q
1
W
2
; s; e 64a
with:
f
1
W
1
; s; e f W
1
; es; W
2
s Z
2
t 64b
q
1
W
2
; s; e qW
2
; es ~ q
1
W
2
; s; e q
1
W
2
; s; e
q
1
0; s; e 64c
It readily follows from (64b) that:
f
1
W
1
; s; e 0 b
0
wcos2s b
2
w
1
coss 0
T
64d
where the following notations are adopted in coherence with (19b):
W
1
w
1
w
2
w
T
65
By Theorem 1 and (64c) one has W
2
s ! 0 as s ! 1. Since
q
1
W
2
; s; e is continuous in W
2
, it follows that ~ q
1
W
2
; s; e ! 0
as s ! 1. Then, according to the averaging theory, one gets stabil-
ity results regarding the system of interest (64a) and (64b) by ana-
lyzing the averaged system dened by:
_
W
1
eAW
1
e

f
1
W
1
_ _
eg q
1
0
4
66a
where

f
1
W
_ _

def
lim
e!0
1
2pN
_
2pN
0
f
1
W; s; e
_ _
ds 66b
q
1
W
2
_ _

def
lim
e!0
1
2pN
_
2pN
0
q
1
0; s; e ds 66c
Note that the fourth term on the right side of (64a) has not been
accounted for in (66a) because its average value is null, due to the
periodicity (with period 2p of y

ref
. From (64c), one has:

f
1
W
_ _
0 0 0
T
67
In view of (67) the average system (66a) simplies to:
_
W
1
eA
1
W
1
eg q
1
0
4
68
Now, let us check that A
1
is in turn Hurwitz. Its characteristic
polynomial is:
detkI A
1
k
3
c
1
dk
2
dc
1
c
2
k dc
1
c
2
69
Applying for instance the well known Rouths algebraic criteria,
it follows that all zeros of the polynomial (69) have negative real
parts if the coefcients c
1
; c
2
and d are positive which actually is
presently the case. Hence, the matrix A
1
is Hurwitz implying that
the autonomous part of the linear system (68) is globally exponen-
tially stable. Then, the solution of the system (68) satises:
lim
t!1
W
1
t q
1
0
4
A
1
1
g; exponentially and whatever W
1
0
70
The exponential feature of the convergence is due to the linear-
ity of (68). From (70), it follows that the state vector
W

1

def
q
1
0
4
A
1
1
g 2 IR
3
71
is a globally exponentially stable equilibrium of the average system
(68). Now, invoking averaging theory, e.g. Theorem 10.4 in [6], we
conclude that there exists a positive real constants e

such that,
for all 0 < e < e

, the differential equation (64a) has a 2p-periodic

solution W
1
s W
1
s; e, that continuously depends on e and that:
lim
e!0
W
1
s; e W

1
; exponentially 72a
The same result applies to the differential equation (60) using
the relation Z
1
t W
1
s with s x
e
t. That is, Z
1
t Z
1
t; e is
2p=x
e
-periodic, continuously depends on e and satises:
lim
e!0
Z
1
t; e W

1
72b
240 A. El Fadili et al. / Electrical Power and Energy Systems 43 (2012) 230244
This establishes Part 1 of Theorem 2.
To prove Part 2, note that from (71) one has:
W

1
q
1
0
4
A
1
1
g 73a
Also, it is readily checked using (19d) and (19e) that:
W

1
q
1
0
4
A
1
1
g q
1
0
4
0
dCb
0
b
3
Cb
0
b
1
1
Cb
0
_ _
T
73b
Furthermore, it is obviously seen from (19b) that:
d Cb
0
b
3
Cb
0
a
1
0 73c
Consequently, one gets from (72b), (73b) and (73c) that:
lim
e!0
z
2
t; e w

2
0 74a
lim
e!0
kt; e w

q
1
0
4

Cb
0
74b
Finally, notice that:
q
1
0
4
lim
e!0
1
2pN
_
2pN
0
q
1
0
4
; s; e ds using 66c
lim
e!0
1
2pN
_
2pN
0
q0
4
; es ds using 64b 75a
Introducing the variable change s x
e
t, (75a) becomes:
q
1
0
4
lim
e!0
x
e
2pN
_
2pN=xe
0
q0
4
; tdt q0
4
75b
This, together with (74a) and (74b), establishes Part 2-b and
completes the proof of Theorem 2. h
Remark 4
(a) The motor speed and the rotor ux norm both converge to
their respective references because the errors z
3
; z
5
con-
verge to zero, a result of Theorem 1.
(b) Using Proposition 1 and the fact that the tuning parameter k
and its time derivative are available, we gets that the error
z
1
x
1
x

1
i
e
kv
e
converges exponentially fast to zero.
The importance of Theorem 2 (part 2) lies (partly) in the fact
that the (time varying) parameter k does converge to a xed
value (up to a harmonic error that depends on x
e
. This
demonstrates that the PFC requirement is actually fullled
with an accuracy that depends on x
e
. The larger x
e
is, the
more accurate the PFC quality. It will be seen in the next
simulation study that the usual value x
e
100p rd=s leads
to a quite acceptable quality.
(c) Theorem 2 (Part 2) also demonstrates that the tracking
objective is achieved (in the mean) for the DC-link squared
voltage y x
2
2
v
2
dc
with an accuracy that depends on the
voltage network frequency x
e
. The class of admissible refer-
ences v
dcref
and X
ref
includes periodic signals with period
Np=x
e
. That is, these signals must oscillate less rapidly than
the network voltage.
(d) The fact that the tracking error z
2
v
2
dc
v
2
dcref
is harmonic
proves the existence of output ripples. Theorem 2 (Part 2)
ensures that the effect of these ripples is insignicant if x
e
is sufciently large. It will be observed through simulations
that the value x
e
100p rd=s leads to sufciently small
ripples.
4. Simulation
The experimental setup has been simulated within the Matlab/
Simulink environment. The simulated system has the following
characteristics:
Supply network: v
e
t

2
p
E cosx
e
t a single phase
220 V n 50 Hz.
AC/DC/AC converters: L
1
15 mH; C 1:5 mF; modulation fre-
quency: 10 KHz.
The induction machine characteristics are summarized in
Table 1.
The experimental protocol is described by Figs. 911. Accord-
ingly, the inertia moment J and viscous friction coefcient f deviate
from their nominal values on some time intervals (Figs. 9 and 10).
The applied load torque T
L
(Fig. 11) and rotor speed reference X
ref
(Fig. 18) are chosen so that the induction machine works in two
zones of its magnetic characteristic and operates both as motor
and generator. Indeed, the induction machine operates in motor
mode in the interval 0; 16s and in generator mode in the interval
16s; 20s. The DC-link voltage reference is constant, i.e.
v
dcref
600 V.
Recall that the nonlinear adaptive controller to be illustrated is
described by: (i) the control laws (10), (18) and (49); (ii) the
parameter update laws (51)(53); and (iii) the optimal ux refer-
ence generator (25). The following values of the controller design
parameters proved to be suitable:
c
1
1000; c
2
30; c
3
100; c
4
400; c
5
500;
c
6
1000; d 100
The above adaptive controller will be compared with its simpli-
ed (constant ux reference) version obtained keeping the ux ref-
erence constant equal to its nominal value, i.e. 0:56 Wb. To avoid
confusion, the two controllers will be referred to Optimal Flux Ref-
erence (OFR) adaptive controller and Constant Flux Reference (CFR)
adaptive controller.
The performances of both controllers are illustrated by
Figs. 1220. It is seen from Fig. 12 that the input current i
e
is
higher with the CFR adaptive controller than with the OFR.
Fig. 13 shows the reference and measured input current i
e
response of the OFR controller: it is observed that the current
amplitude changes whenever the speed reference or the load
torque vary. However, the current frequency is insensitive to
these changes. Specically, the current remains most time either
in phase (motor mode) or opposed phase (generator mode) with
the supply net voltage, complying with the PFC requirement.
This is further demonstrated by Figs. 1416 which show that
the ratio k takes a constant value after short transient periods
0 5 10 15 20
0
0.5
1
1.5
2
2.5
x 10
-3
F
r
i
i
c
t
i
o
n

c
o
e
f
f
i
c
i
e
n
t

N
m
.
s
/
r
d
Time (s)
Fig. 9. Friction coefcient variation.
A. El Fadili et al. / Electrical Power and Energy Systems 43 (2012) 230244 241
following changes in the references signal and load torque. Note
that the ratio k is negative when the network gets back energy.
Fig. 17 shows that the DC-link voltage x
2
v
dc
is well regulated
and quickly settles down after each change in the speed refer-
ence or the load torque. As stipulated by Theorem 2, and com-
mented on in Remark 4c, the DC link voltage v
dc
is subject to
small amplitude ripples oscillating at the supply net frequency
x
e
. Fig. 18 shows that both controllers ensure a perfect asymp-
totic speed reference tracking despite the uncertainty and
changes of the mechanical parameters (load torque T
L
, rotor
inertia J and viscous friction coefcient f). Fig. 19 shows the
resulting (state-dependent) optimal ux reference (for the OFR
adaptive controller) and the constant ux reference (for the sim-
plied CFR adaptive controller). It is clearly seen that the state-
dependent ux reference varies signicantly in function of the
load torque and the speed. Fig. 20 shows the absorbed stator
currents for both controllers, in various operation conditions. It
is seen that the OFR controller requires a smaller current than
the constant ux controller. This difference is more signicant
in presence of low load torques, because in this situation the
optimal ux reference is most different from the constant ux
reference.
0 5 10 15 20
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
R
o
t
o
r

i
n
e
r
t
i
a

K
g

m

Time (s)
Fig. 10. Rotor inertia variation.
0 5 10 15 20
-30
-20
-10
0
10
20
30
40
50
60
L
o
a
d

t
o
r
q
u
e

T
L

(
N
m
)
Time (s)
Fig. 11. Applied load torque.
0 0.05 0.1 0.15 0.2
-150
-100
-50
0
50
100
150
i
e

(
A
)
OFR
CFR
Time (s)
Fig. 12. Zoom on the input current ie response (A) over the interval 0 0:2s (solid:
OFR controller, dotted: CFR controller).
4 4.1 4.2 4.3 4.4 4.5 4.6
-30
-20
-10
0
10
20
30
i
e
kv
e
Time (s)
i
e
(
A
)

&

k
v
e

(
V
)
Fig. 13. Zoom on the reference and measured input current ie(A) over the interval
4 4:6s when using the OFR controller.
13.9 14 14.05 14.1
-50
-40
-30
-20
-10
0
10
20
30
40
50
i
e
0.04v
e
13.95
Time (s)
i
e
(
A
)

&

0
.
0
4
v
e

(
V
)
Fig. 14. Unitary power factor checking with the OFR controller when the induction
machine absorbs energy (motor mode operation).
242 A. El Fadili et al. / Electrical Power and Energy Systems 43 (2012) 230244
19.4 19.42 19.44 19.46 19.48 19.5
-15
-10
-5
0
5
10
15
Time (s)
i
e
(
A
)

&

0
.
0
4
v
e

(
V
)
i
e
0.04v
e
Fig. 15. Unitary power factor checking with the OFR controller when the machine
restores energy (generator mode operation).
0 5 10 15 20
-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
Time (s)
k
OFR
CFR
Fig. 16. Tuning parameter k. Solid: OFR controller; dotted: CFR controller.
0
200
400
600
D
C
-

l
i
n
k

v
o
l
t
a
g
e

V
d
c
0 5 10 15 20
-2
-1
0
1
2
Time (s)
V
d
c

e
r
r
o
r

(
V
)
0 5 10 15 20
Time (s)
Fig. 17. DC-link voltage v
dc
(V) response. Upper: reference and measured DC-link
voltage; lower: DC-link voltage error control.
0 5 10 15 20
0
20
40
60
80
100
120
M
o
t
o
r

s
p
e
e
d

(
r
d
/
s
)
Time (s)
Fig. 18. Rotor speed X (rd/s) response: the identical speed responses obtained by
the OFR controller and the constant ux controller.
0 5 10 15 20
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Time (s)
r
o
t
o
r

f
l
u
x

n
o
r
m

(
W
b
)
Fig. 19. Rotor ux norm reference (Wb) (solid: state-dependent optimized ux,
dotted: constant ux).
2 4 6 8 10 12 14 16 18 20
0
5
10
15
s
t
a
t
o
r

c
u
r
r
e
n
t

(
A
)
2 4 6 8 10 12 14 16 18 20
0
5
10
15
Time (s)
Time (s)
s
t
a
t
o
r

c
u
r
r
e
n
t

(
A
)
Fig. 20. Absorbed stator current (A) (Upper: constant ux controller. Lower: OFR
controller).
A. El Fadili et al. / Electrical Power and Energy Systems 43 (2012) 230244 243
5. Conclusion
The problem of controlling associations including an AC/DC rec-
tier, a DC/AC inverter and induction motor has been addressed
with consideration of magnetic characteristic saturation. The sys-
tem dynamics have been described by the averaged 7th order non-
linear state-space model (6a)(6g). Based on such a model, a
multiloop nonlinear adaptive controller has been designed and
analyzed in several steps. The controller consists of: (i) the control
laws (10), (18) and (49); (ii) the parameter update laws (51)(53);
and (iii) the optimal ux reference generator (25). Their perfor-
mances have been analyzed using tools from the Lyapunov stabil-
ity and averaging theory. It has been formally established that the
controller actually meets the objectives it has been designed to, i.e.
(i) almost unitary power factor; (ii) well regulated DC-link voltage
v
dc
; (iii) satisfactory rotor speed reference tracking over a wide
range of mechanicals parameters variation; and (iv) tight optimiza-
tion and regulation rotor ux norm for minimizing the stator cur-
rent absorbed. These results have been conrmed by a simulation
study. To the authors knowledge, it is the rst time that a so com-
plete formal design and analysis framework has been developed
for inductions motors in presence of magnetic saturation control
design.
References
[1] Abdelaziz M, Ghedjati K. Control by feedback linearization of the torque and
the ux of the induction motor. In: 7th WSEAS international conference on
systems theory and scientic computation, Athens, Greece, vol. 7; 2007. p. 84
90.
[2] De Wit Canudas C, Ramierz J. Optimization, discretization, observers and
control of inductions motors. Paris: Hermes Science Europe Ltd.; 2000.
[3] Mahmuda MA, Potaa HR, Hossainb MJ. Full-order nonlinear observer-based
excitation controller design for interconnected power systems via exact
linearization approach. Int J Electr Power Energy Syst 2012;41:5462.
[4] Jasinski M, Cichowlas M, Kazmierkowski MP. Direct control for AC/DC/AC
converter-fed induction motor with active ltering function. Int J Comput
Math Electrical Electronic Eng 2006;25:23542.
[5] Kazmierkowski MP. Control strategies for PWM rectier/inverter-fed
induction motors. In: IEEE international symposium on industrial electronics
(ISIE 2000), vol. 1; 2000. p. 1521.
[6] Khalil H. Nonlinear systems. NJ, USA: Prentice Hall; 2003.
[7] Krstic M, Kanellakopoulos I, Kokotovic P. Nonlinear and adaptive control
design. John Willey & Sons, Inc.; 1995.
[8] Leonard W. Control of electrical drives. NY: Springer; 2001.
[9] Zaveri T, Bhalja BR, Zaveri Naimish. A novel approach of reference current
generation for power quality improvement in three-phase, three-wire
distribution system using DSTATCOM. Int J Electrical Power Energy Syst
2011;33(10):170210.
[10] Michael J, Ryan D, Rik W. Modeling of sinewave inverters: a geometric
approach. In: IEEE industrial electronics conference (IECON 1998), vol. 1; 1998.
p. 396401.
[11] Montanan M, Peresada S, Tilli A. A speed-sensorless indirect eld-oriented
control for induction motors based on high gain speed estimation. Automatica
2006;42(10):163750.
[12] Moreno-Eguilaz JM, Peracaula J. Efciency optimization for induction motor
drives: past, present and future. In: Electrimacs 99: modelling and simulation
of electric machines converters and systems. IMACS No. 6, Lisboa, Potugal;
1999 p. 18791.
[13] Ouadi H, Giri F, De Leon-Morales J, Dugard L. Accounting for magnetic
characteristic nonlinearity in designing induction motor observer. In: IFAC
symposium on nonlinear control systems, NOLCOS, Stuttgart, Germany; 2004.
p. 140914.
[14] Ouadi H, Giri F. Induction motor robust adaptive control. In: Mediterranean
conference on control and automation MED 2002, Lisbon Portugal; 2002.
[15] Novotnak RT, Chiasson J, Bodson M. High performance motion control of an
induction motor with magnetic saturation. IEEE Trans Contr Syst Technol
1999;7:31527.
[16] Singh B, Bhuvaneswari G, Garg V. Improved power quality ACDC converter for
electric multiple units inelectric traction. In: Power India conference; 2006. p. 6.
[17] Sira H, Silva R. Control design techniques in power electronics
devices. Springer; 2006.
[18] Hansen S, Malinowski M, Blaabjerg F, Kazmierkowski MP. Sensorless control
strategies for PWM rectier. In: Applied power electronics conference and
exposition, 2000. APEC 2000, vol. 2; 2000. p. 8328 [Fifteenth annual IEEE].
[19] Traor D, Plestan F, Glumineau A, de Leon J. Sensorless induction motor: high
order sliding mode controller and adaptive interconnected observer. IEEE
Trans Ind Electron 2008;55:381827.
[20] Saha S, Aldeen M, Tana CP. Fault detection in transmission networks of power
systems. Int J Electrical Power Energy Syst 2011;33(4):887900.
[21] Singh B, Jayaprakash P, Kothari DP. New control approach for capacitor
supported DSTATCOM in three-phase four wire distribution system under
non-ideal supply voltage conditions based on synchronous reference frame
theory. Int J Electrical Power Energy Syst 2011;33(5):110917.
[22] Ouadi H, Giri F, Dugard L. Accounting for magnetic saturation in induction
machines modeling. Int J Model Identif Control 2011;14:2736.
244 A. El Fadili et al. / Electrical Power and Energy Systems 43 (2012) 230244