6

Sensorless First- and Second-Order Sliding-Mode

Control of a Wind Turbine-Driven Doubly-Fed

Induction Generator
Ana Susperregui, Gerardo Tapia and M. Itsaso Martinez
University of the Basque Country (UPV/EHU)
Spain

1. Introduction
The doubly-fed induction generator (DFIG) is a wound-rotor electric machine on which
about 75% of the wind turbines installed nowadays are based. As sketched in Fig. 1,
when generating power, its stator is directly connected to the grid, while a back-to-back
double-bridge converter comprising both the rotor- (RSC) and grid-side (GSC) converters
interfaces its rotor with the grid, hence allowing the ow of slip power both from the grid
to the rotor at subsynchronous speeds and vice-versa at supersynchronous speeds
within a certain speed range.
Given that only the slip power has to be managed by the bidirectional rotor converter, it is
sufcient to size it so that it typically supports between 25% and 30% of the DFIG rated power
(Ekanayake et al., 2003; Pea et al., 1996). This is more than probably the main reason for the
success of the DFIG in the eld of variable-speed wind generation systems.
.

.G r i d
.
.
Q

P
s

r,

D C b u s

G e a rb o x

D F IG

R S C

G S C

Fig. 1. Structure of a DFIG-based wind turbine

Standard eld oriented control (FOC) schemes devised to command wind turbine-driven
DFIGs comprise proportional-integral (PI)-controlled cascaded current and power loops,
which require the use of an incremental encoder (Tapia et al., 2003). Although stator-side
active and reactive powers can be independently governed by adopting those control
schemes, the system transient performance degrades as the actual values of the DFIG
resistances and inductances deviate from those based on which the control system tuning
was carried out during commissioning (Xu & Cartwright, 2006). In addition, the optimum

[Link]

110

Sliding Mode Control

power curve tracking achievable using PI-based control schemes shows a considerable room
for improvement. Even if feedforward decoupling control terms are traditionally incorporated
to enhance the closed-loop DFIG dynamic response, they are extremely dependent on DFIG
parameters (Tapia et al., 2006; Xu & Cartwright, 2006).
In this framework, alternative high dynamic performance power control schemes for DFIGs
are being proposed, among of which a strong research line focuses on the so-called direct
power control (DPC) (Xu & Cartwright, 2006; Zhi & Xu, 2007). Several others explore the
alternative of applying sliding-mode control (SMC), both standard rst-order (Beltran
et al., 2008; Susperregui et al., 2010), and higher-order (Beltran, Ahmed-Ali & Benbouzid,
2009; Beltran, Benbouzid & Ahmed-Ali, 2009; Ben Elghali et al., 2008).
Moreover, since, as already mentioned, the back-to-back rotor converter is sized to manage a
slip power up to 25% or 30% of the wind generator rated power, DFIGs are kept connected
to the grid provided that their rotational speed remains within a certain range. Accordingly,
connection of DFIGs to the grid is only accomplished if the wind is strong enough to extract
energy from it protably. In particular, the four-pole 660-kW DFIG considered in this chapter
is not connected to the grid until its rotational speed exceeds the threshold value of 1270 rpm.
Yet, connecting the DFIG stator to the grid is not straightforward. In fact, although
wind-turbine-driven DFIGs are asynchronous machines, owing to the double-bridge rotor
converter managing the slip power, they behave as real synchronous generators. Accordingly,
prior to connecting the stator of a DFIG to the grid, the voltage induced at its stator terminals
must necessarily be synchronized to that of the grid.
However, even though control of wind turbine-driven DFIGs is a topic extensively covered
in the literature, not many contributions outline or describe in some detail possible strategies
for smooth connection of DFIGs to the grid. So far, the synchronization problem has been
approached from different viewpoints, hence giving rise to alternative methods, as open-loop
stator voltage control (Pea et al., 2008), closed-loop regulation of rotor current (Peresada et al.,
2004; Tapia et al., 2009), and phase-locked loop (PLL) (Abo-Khalil et al., 2006; Blaabjerg et al.,
2006) or even direct torque control (DTC) of the voltage induced at the open stator (Arnaltes
& Rodrguez, 2002).
Considering those precedents, together with the robustness and tracking ability naturally
conferred by SMC, both a rst-order and a higher-order sensorless SMC algorithms, conceived
to command the RSC feeding the rotor of a DFIG, are described and evaluated in this chapter.
Those two algorithms are not only aimed at governing active and reactive power exchange
between the DFIG stator and the grid, but also at ensuring the synchronization required for
smooth connection of the DFIG stator to the grid.
The chapter is organized as follows. Given that the DFIG exhibits different dynamics
depending on whether its stator is connected to the grid or not (Tapia et al., 2009), the
mathematical model corresponding to each of those two operating conditions is rst briey
presented. Conditions to reach synchronization are also provided. After selection of
the switching functions associated, respectively, to the power control and synchronization
objectives, a global rst-order sliding-mode control (1-SMC) algorithm, based on Utkins
research work on various other types of electric machines (Utkin et al., 1999; Utkin, 1993;
Yan et al., 2000), is described in detail. Stability analyses are also provided for both the power
control and synchronization operation regimes. An overall second-order sliding-mode control
(2-SMC) algorithm, alternative to the previous one, is next presented. Special attention is paid
to the derivation of effective tuning equations for all its gains and constants. The practical
issue related to bumpless transition between the controllers in charge of synchronization

[Link]

Sensorless First- and Second-Order Sliding-Mode

Control of a Wind Turbine-Driven Doubly-Fed Induction Generator

111

and power control, at the instant of connecting the DFIG stator to the grid, is then tackled.
Adaptation of the model reference adaptive system (MRAS) observer put forward in (Pea
et al., 2008), so that it remains valid for sensorless control during synchronization, is also
dealt with. Sensorless versions of the two SMC algorithms proposed are evaluated via
real-time hardware-in-the-loop (HIL) emulation over a virtual 660-kW DFIG prototype. The
chapter nishes with a conclusion section, devoted to analyze the results arising from the HIL
emulation tests carried out.

2. Review of DFIG model and grid synchronization

Focused on a 660-kW DFIG, the main objective of the two alternative versions of the
control system presented along this chapter consists in succeeding in the achievement of the
maximum active power the machine is able to generate at each rotational speed; i.e., to track
the DFIG optimum power curve. As a secondary goal, but still essential from the point of
view of the electricity supply quality, the reactive power the machine generates or absorbs
from the grid is also managed.
Before raising the modelling of the machine, and, to get ride of misunderstandings due to the
diverse nomenclature used to identify the reference frames taking part in FOC, Fig. 2 presents
the terminology that is going to be adopted hereafter. It can be observed that the stator direct
and quadrature axes are represented as sD and sQ, respectively, and that the rotor reference
frame, which forms the r turning angle with respect to sD axis, is denominated r-r. Since
the machine is going to be rotor-side controlled, the magnitudes will be referred to a frame,
labeled as x-y, whose direct axis is aligned with the stator ux, 
s and, therefore, with the
stator magnetizing current, ms . The latter reference frame is turned s with respect to the
sD-sQ plane.
y

r b

s Q

w
y

H
im
s

x
s

r
s

m s

w
r

r a
r

s D

Fig. 2. Stator-ux-oriented reference frame

When connected to the grid, the rotor-side-voltage generator model regarding the
stator-ux-oriented reference frame x-y may be expressed as (Tapia et al., 2009; Vas, 1998)
s |
L m d| 
dirx
+
sl Lr iry
dt
L s dt
diry
Lm
= Rr iry + Lr
| s | + sl Lr irx ,
+ sl
dt
Ls

vrx = Rr irx + Lr

(1)

vry

(2)

where vrx and vry are the components of the rotor voltage, irx and iry represent the rotor
currents, and sl = ms r stands for the slip frequency between the r and x axes. Rr , L s

[Link]

112

Sliding Mode Control

and L m denote the rotor resistance, and the stator and magnetizing inductances, respectively.
Finally, Lr = Lr symbolizes the transient inductance of the rotor, where = 1 L2m /( L s Lr )
is the total leakage factor.
Taking into account that power generation is not protable at low speeds less than 1270
rpm in this particular case, the generator will not be connected to the grid until this
threshold value is exceeded. Therefore, a new "grid-non-connected" state appears where the
machine dynamic behaviour differs from that in which its stator is connected to the grid, and,
consequently, the model changes. Moreover, the transition between the disconnected and
connected states is not trivial, since the grid voltage and that induced at the open stator of the
DFIG may present magnitude and/or phase differences. At this point, aiming at removing the
risk of short circuit, it can be taken advantage of a properly controlled "grid-non-connected"
state, turning it into a synchronization stage.
Let a new x -y reference frame be dened when the stator is disconnected from the grid,
where, as shown in Fig. 3, its y quadrature axis and the grid voltage space-phasor are
collinear. Moreover, assuming steady-state regime, and, if rotor current is stable, it can be
demonstrated (Tapia et al., 2009) that the stator ux and voltage space-vectors are collinear to
s vs .
x and y axes, respectively; i.e., 
s Q
y '
y

H
s

g r id

w
s

r
s

x '

r s
s D

Fig. 3. New reference frame for synchronization

Bearing in mind that stator current is null when disconnected from the grid, the new
open-stator model, expressed according to the x -y reference frame, arises (Tapia et al., 2009):
v
Rr
di
dirx
sl Lr iry rx = rx
i + sl iry
dt
dt
Lr
Lr rx
diry
vry
diry
Rr
= Rr iry + Lr
+ sl Lr irx
=
iry sl irx .
dt
dt
Lr
Lr

vrx = Rr irx + Lr

(3)

vry

(4)

As evidenced in Fig. 3, synchronization may be achieved if x-y and x -y reference frames are
aligned. However, for a complete match-up, the grid and stator voltage space-vectors must
be not only collinear but also identical in magnitude. The two conditions are satised if the
following rotor current values are achieved (Tapia et al., 2009):




v grid 
; i
= 0,
(5)
irx re f =
s L m ry re f

and, consequently, synchronization is ensured; i.e.:





v grid 
irx = irx =
; iry = iry = 0; s = s .
s Lm

[Link]

(6)

Sensorless First- and Second-Order Sliding-Mode

Control of a Wind Turbine-Driven Doubly-Fed Induction Generator

113

Furthermore, if the current values presented in (6) are substituted into the stator-side reactive
and active power expressions given next (Vas, 1998):
Qs =

3 Lm
3 |vs |
(| 
s | L m irx ); Ps =
|vs |iry ,
2 Ls
2 Ls

(7)

it follows that, at the instant of the connection, zero power-exchange is achieved.

For the sake of a proper performance of the whole system, each state must be commanded
with its own controller. Moreover, the transition between the two states must carefully be
followed up, depending on the control strategy being applied, in order to achieve a bumpless
connection. This aspect will be thoroughly described in a later section.

3. Sensorless sliding-mode control arrangement for the DFIG

Aiming to track the optimum power curve of the DFIG, sliding-mode control theory has
been adopted, which provides the system with superior tracking ability and high robustness
despite uncertainties or parameter variations. The basis of SMC is the judicious election of a
switching variable, which usually depends on a linear combination of the error of the variable
to be commanded and its subsequent time derivatives. Here, the proposed switching variables
for optimum power control stator connected to the grid are
s Qs = e Qs + c Q
s Ps = e Ps + c P

e Qs dt

(8)

e Ps dt,

(9)

where e Qs = Qs re f Qs and e Ps = Ps re f Ps represent de errors in reactive and active powers,

respectively, and the integral terms, weighted by c Q and c P positive constants, are added for
steady-state response improvement (Utkin et al., 1999).
Examining the set-points proposed in (5), it can be derived that rotor current regulation must
be carried out if synchronization is required. Therefore, when the stator is disconnected from
the grid, and, similarly to the previous case, the following switching functions are suggested:
sirx = eirx + c x
siry = eiry + cy

eirx dt

(10)

eiry dt,

(11)

where eirx = irx re f irx and eiry = iry re f iry represent the errors in irx and iry ,
respectively, and crx and cry are positive constants.
The switching variable denes the relative degree of a system, and, as a result, the order of the
applicable SMC (Levant, 1993). As the system is of rst-order relative degree in both states,
connected and disconnected from the grid, it may be commanded applying 1-SMC or 2-SMC
(Bartolini et al., 1999). The design of the two controllers is detailed in subsequent sections.
3.1 First-order sliding-mode control

In this section, a 1-SMC scheme is proposed. Due to the different dynamic behaviours
presented by the DFIG when disconnected or connected to the grid, a different DFIG model is
considered to conceive the control of each of those two cases, and a rst-order sliding-mode
controller is accordingly synthesized for each of them.

[Link]

114

Sliding Mode Control

In particular, the 1-SMC applied is that based on V. I. Utkins research work (Utkin et al., 1999;
Utkin, 1993; Yan et al., 2000), which sets out the following: most of the electrical systems must
modulate the control signals in order to command the transistors gates of their converters; so,
why not directly generate those gating signals thus eluding the use of pulse-width modulation
(PWM) or space-vector modulation (SVM) techniques? (Yan et al., 2008) This theory ts
perfectly the present case, in which controllers for the RSC of the back-to-back conguration
are designed for the two possible connection states of the DFIG.

u
0

w 1

a ,v

w 3

w 2

ira

ra

b ,v
N

irb

rb

c ,v
s
- u

w 4

w 5

irc

rc

w 6

Fig. 4. Rotor-side converter scheme

Depending on whether or not the DFIG stator is connected to the grid, its model and
controllers do vary, but the RSC to be commanded, displayed in Fig. 4, remains obviously
the same. Analyzing this scheme (Utkin et al., 1999), it is possible to nd a link between:
the signals generated by controllers based on a synchronous frame, vrx and vry , and those
between the midpoints of the converter legs and the DC link, vraN , vrbN and vrcN :
V abc

V xy
D
  

   

 vraN
2
2
vrx
cos cos( 3 ) cos( + 3 )
vrbN .
=
2
vry
sin sin( 2
3 ) sin( + 3 )
vrcN

(12)

If the opposite relation is needed, the inverse of D matrix must exist. But, as it is not square,
Moore-Penrose pseudo-inverse concept (Utkin et al., 1999) may be used to calculate its
inverse, D+ = D T (DD T )1 , resulting the previous matrix expression in:
D+





sin
cos
vraN
vrbN = cos( 2 ) sin( 2 ) vrx ,
3
3
vry
2
vrcN
cos( + 2
3 ) sin( + 3 )

where = s r .

(13)

the voltages vraN , vrbN and vrcN , and the transistors gating signals, sw1 , sw2 , sw3 , sw4 , sw5
and sw6 :
sw1 = 0.5(1 + vraN /u0 ) sw4 = 1 sw1
sw2 = 0.5(1 + vrbN /u0 ) sw5 = 1 sw2 .
(14)
sw3 = 0.5(1 + vrcN /u0 ) sw6 = 1 sw3
The following sections describe the design of the control scheme for the cases mentioned
above: DFIG connected to and disconnected from the grid.

[Link]

Sensorless First- and Second-Order Sliding-Mode

Control of a Wind Turbine-Driven Doubly-Fed Induction Generator

115

3.1.1 DFIG connected to the grid Optimum power generation

Once synchronization is completed and the DFIG is connected to the grid, it is going to be
commanded applying the following multivariable control law, in order to achieve optimum
power generation:
(15)
V abc = u0 sgn(S ),

T
where S = s1 s2 s3
contains the switching variable expressions represented in a-b-c
three-phase reference frame. Note that, as the system to be controlled presents negative gain,
that of the control law must also be negative if stability is pursued.
Aiming to ease the design of the controllers and, subsequently, to demonstrate the stability
T

of the closed-loop system, the model can be transferred to subspace S QP = s Qs s Ps , if the
time derivatives of (8) and (9) are taken, and making use of (1)-(2):
S QP

V xy

FQP

   
  
  


F1
v
s Qs
=
+ a rx
F2
vry
s Ps

(16)

where F1 = f ( Q s re f , |vs |, | 
s |, Qs re f , irx , wsl , iry ), F2 = f ( P s re f , |vs |, | 
s |, Ps re f , irx , wsl , iry ),

and a = 23 LLs mL |vs |.

r
It is possible to relate the new model in (16) to the voltage signals between the midpoints of
the converter legs and the DC link, if D transformation matrix in (12) is applied
Da


S QP = F QP + aD V abc .

(17)

It can be noticed that control signals are transformed from a-b-c to the stator-ux-oriented
reference frame by means of D a matrix. Now, it seems logical to derive the S in (15) by
T

arranging (8) and (9) in matrix format, S QP = s Qs s Ps , and then transforming S QP by
means of D+
a :
S = D+
(18)
a S QP .
This allows obtaining the three-phase control signals as:

sgn(s Ps sin s Qs cos s )

V abc = u0 sgn(s Ps sin( 2
3 ) s Q s cos(
sgn(s Ps sin( + 2
3 ) s Q s cos( +

3 )) ,
2 ))
3

(19)

where 1/a constant should appear multiplying the terms inside every sgn function. However,
as its value is always positive, it does not affect the nal result, and this is the reason why it
has been removed from (19). To conclude, the transistor gating signals are achieved just by
replacing (19) in (14).
Due to the discontinuous nature of the generated command signals which are in fact the
transistors gating signals, a bumpless transition between synchronization and optimum
generation states takes place spontaneously, without requiring the use of further control
techniques, as that proposed in (Tapia et al., 2009).

[Link]

116

Sliding Mode Control

[Link] Stability proof

In order to conrm that the designed control signals assure the zero-convergence of the
switching variables, the following positive-denite Lyapunov function candidate is proposed:
V=

1 T
S S ,
2 QP QP

(20)

and, as it is well-known, its time derivative must be negative-denite:


1 T
V =
S QP S QP + S TQP S QP = S TQP S QP < 0.
2

(21)

Considering (17) and (18), the Lyapunov function time derivative can be rewritten as:

T
sgn(s1 ) 0.5sgn(s2 ) 0.5sgn(s3 )
s1
4
V = S T F a2 u0 s2 sgn(s2 ) 0.5sgn(s3 ) 0.5sgn(s1 ) ,
9
sgn(s3 ) 0.5sgn(s1 ) 0.5sgn(s2 )
s3

(22)

where F = D Ta F QP = [ F1 F2 F3 ] T .
Taking into account that the elements of V abc will never coincide in sign at every moment, nor
will S components, as it can be inferred from (15). Therefore, sgn(sl ) = sgn(sm ) = sgn(sn ),
where l = m = n, for l, m, n {1, 2, 3}. Let l = 1, m = 2 and n = 3; moreover, suppose that
sgn(s1 ) = +1 = sgn(s2 ) = sgn(s3 ), then (22) could be transformed into:
4
V = s1 F1 + s2 F2 + s3 F3 a2 u0 (2| s1 | + | s2 | + | s3 |) .


 9


p

(23)

If V < 0 must be guaranteed, it can be stated that | q | > | p|. Furthermore, if the most restrictive
case is considered, the following condition must be derived:
4 2
a u0 (2| s1 | + | s2 | + | s3 |) > | s1 || F1 | + | s2 || F2 | + | s3 || F3 |.
9

(24)

Comparing each accompanying term of | s1 |, | s2 | and | s3 |, u0 can be xed by guaranteing that



| F1 |
9
u0 > 2 max
(25)
, | F2 |, | F3 |
2
4a
is satised. Nevertheless, bearing in mind the remaining signs combinations between
switching functions s1 , s2 and s3 , and taking into account the most demanding case, the above
proposed condition turns out to be:
u0 >

9
max (| F1 |, | F2 |, | F3 |) .
4a2

(26)

Provided that the controller supplies the convenient voltage, derived from (26), the system is
robust even in the presence of disturbances, guarantying thus the asymptotic convergence of
s Qs and s Ps to zero. (26) presents a very conservative condition, but, in practice, a lower value
of u0 is usually enough to assure the stability of the whole system.

[Link]

Sensorless First- and Second-Order Sliding-Mode

Control of a Wind Turbine-Driven Doubly-Fed Induction Generator

117

3.1.2 DFIG disconnected from the grid Synchronization stage

When rotor speed-threshold is achieved, the control system activates the synchronization
stage. As mentioned before, in order to avoid short circuit, the goal is to match the stator
and grid voltages in magnitude and phase by requesting the reference values presented in (5).
Let the same multivariable control law structure exposed in (15) be employed, considering,
of course, the new subspace where it must be applied. Combining (3) and (4) with the
time derivatives of (10) and (11), the model can be transferred to the above-mentioned new

T
subspace S x y = sirx siry :
S x y

x y
x y
    
  



s irx
v
M1
=
+ b rx ,
s iry
vry
M2

where M1 =
b = 1/Lr .

f (irx re f , irx re f , irx , wsl , iry ), M2

(27)

f (iry re f , iry re f , iry , wsl , irx ), and

Following a similar procedure to that presented in 3.1.1, the switching variables referred to
a-b-c reference frame will be obtained as:

S = D+
b Sx y ,

(28)

where D+
b is the Moore-Penrose pseudo-inverse of D b = bD. Substituting (28) in the proposed
control law (15), the three-phase command signals to be generated turn out to be:

sgn(sirx cos siry sin )

2
2
V abc = u0sgn(sirx cos( 3 ) siry sin( 3 )) ,
(29)
sgn(sirx cos( +

2
3 ) s i ry

sin( +

2
3 ))

where = s r . The gating signals should easily be achieved by replacing (29) in (14).
[Link] Stability proof

Analogous to the case in [Link], the asymptotic zero-convergence of switching functions is

assured if the following condition is accomplished:
u0 >

9
max (| M1 |, | M2 |, | M3 |) ,
4b2

(30)

where M = DbT M x y = [ M1 M2 M3 ] T , and the positive-denite Lyapunov function candidate

is selected as:
1
(31)
V = S Tx y S x y .
2
3.2 Higher-order sliding-mode controller

The proposed structure based on 1-SMC leads to a variable switching frequency of the RSC
transistors (Susperregui et al., 2010), which may inject broadband harmonics into the grid,
complicating the design of the back-to-back converter itself, as well as that of the grid-side
AC lter (Zhi & Xu, 2007). As an alternative to the 1-SMC, higher-order sliding-mode control
(HOSMC) could be adopted. In particular, and owing to the relative order the system presents,

[Link]

118

Sliding Mode Control

a 2-SMC realization, known as the super-twisting algorithm (STA), may be employed (Bartolini
et al., 1999; Levant, 1993). The control signal comprises two terms; one guaranteing that
switching surface s = 0 is reached in nite time, and another related to the integral of the
switching variable sign. Namely,
u = | s| sgn(s) w

sgn(s)dt,

(32)

where = 0.5 assures a real second-order sliding-mode. This technique gives rise to a
continuous control signal, which not only alleviates or completely removes the "chatter" from
the system, but must also be modulated. To this effect, SVM may be applied, therefore
obtaining a xed switching frequency which results in elimination of the above-mentioned
drawback.
As previously remarked, two controllers must be designed in order to command the
performance of the DFIG when connected and disconnected from the grid.
3.2.1 DFIG connected to the grid Optimum power generation

Considering the time derivatives of (8) and (9) together with expressions (3), (4) and (7), it
turns out that




Rr
3 1
s | +
|vs | c Q | 
c Q L m irx sl L m iry + c Q Qs re f +
s Qs = Q s re f

2 Ls
Lr
3 Lm
+
|vs | vrx
(33)
2 L s Lr



3 Lm
Rr
Lm
|vs | c P iry sl irx sl
| s | + c P Ps re f +
s Ps = Ps re f +
2 Ls
Lr
L s Lr
3 Lm
|vs | vry ,
(34)
+
2 L s Lr
 |
d|

where dts has been neglected due to the fact that the DFIG is grid connected. Aiming to track
the optimum power curve, the voltage to be applied to the rotor may be derived according to
control law
(35)
vrx = vrxST + vrxeq ; vry = vryST + vryeq ,
where the terms with subscript ST are computed, through application of the STA, as:



2 L s Lr
Q | s Qs |0.5 sgn(s Qs ) w Q sgn(s Qs )dt
(36)
vrxST =
3 |vs | L m



2 L s Lr
(37)
P | s Ps |0.5 sgn(s Ps ) w P sgn(s Ps )dt
vryST =
3 |vs | L m

with Q , w Q , P and w P being positive parameters to be tuned. The gain premultiplying

the algorithms the inverse of that affecting control signal in (33) and (34) is exclusively
applied for assisting in the process of tuning the foregoing parameters. The addends with
subscript eq in (35), which correspond to equivalent control terms, are derived by letting
s Ps = s Qs = 0 (Utkin et al., 1999). As a result,

2 L s Lr 
Qs re f + c Q ( Qs re f Qs ) + Rr irx Lr sl iry
vrxeq =
(38)
3 |vs | L m

2 L s Lr 
Lm
vryeq =
|
s | + Lr sl irx .
(39)
Ps re f + c P ( Ps re f Ps ) + Rr iry +
3 |vs | L m
L s sl

[Link]

Sensorless First- and Second-Order Sliding-Mode

Control of a Wind Turbine-Driven Doubly-Fed Induction Generator

119

It should be noted that sliding regime in manifold s Ps = s Qs = s Ps = s Qs = 0 can also be

attained by applying only the control terms in (35) corresponding to the STA. Accordingly,
the equivalent control terms in (38) and (39) are not strictly necessary, but, once included, the
more accurately they are computed, the lower is the control effort let to be done by the STA.
Equivalent control terms are hence incorporated not only to improve the system transient
response (Rashed et al., 2005), but also to ease selection of constants c P and c Q , as well as
tuning of the STA Q,P and w Q,P gains.
In effect, substituting control law (35) into (33) and (34) produces:

 0.5
 
 
s Qs = Q s Qs  sgn s Qs w Q sgn s Qs dt

(40)

s Ps = P |s Ps |0.5 sgn(s Ps ) w P sgn(s Ps ) dt.

(41)

Now, given that sgn(s) = s/ |s|, taking the time derivatives of (40) and (41) leads to:
 1
 0.5
s Qs w Q s Qs  s Qs
sQs = 0.5 Q s Qs 
sPs = 0.5 P |s Ps |

0.5

s Ps w P | s Ps |

(42)

s Ps .

(43)

Let us assume that, thanks to the rst addend in the STA, the reaching
 is satisfactorily
 phase
completed and the sliding regime is entered. From that moment on, sQs ,Ps  Q,P , with Q,P
close to zero. Considering the most unfavorable case, in which s Qs ,Ps  = Q,P , and using the
denition of s Qs ,Ps in (8) and (9), the following expressions can respectively be worked out
from (42) and (43):
a2 (c Q , Q )

a1 (c Q , Q , w Q )

a (c , w )

0 Q
Q



   


1
0.5
0.5
1
w Q c Q e Q dt = 0
Q + c Q e Q + 0.5Q
Q c Q + Q
w Q e Q + Q
eQ + 0.5Q




b2 ( c P , P )

b1 ( c P , P , w P )

(44)

b (c , w )

0 P
P



   



0.5

0.5

1
eP + 0.5P P + c P e P + 0.5P P c P + P w P e P + P w P c P e P dt = 0.




(45)

Taking the time derivatives of (44) and (45), the following differential equations reecting the
e Q and e P error dynamics while in sliding regime are obtained:
...
e Q + a2 eQ + a1 e Q + a0 e Q = 0
...
e P + b2 eP + b1 e P + b0 e P = 0.

(46)
(47)

Hence, once Q,P is xed, adequate selection of c Q,P , Q,P and w Q,P allows attaining certain
target error dynamics established through the third-order characteristic equation given next:




p2 + 2 n p + n2 ( p + n ) = p3 + (2 + ) n p2 + 1 + 2 2 n2 p + n3 = 0 (48)
  






d2

d1

d0

which, provided that is selected high enough typically 10, gives rise to a pair
of dominant poles with respect to a third one placed at p = n . As a result, it can
be considered that target error dynamics are entirely dened via damping coefcient
and n natural frequency. Those designer-dened error dynamics would theoretically be

[Link]

120

Sliding Mode Control

achieved just by tuning c Q,P , Q,P and w Q,P so that a2 = d2Q ( Q , nQ ), a1 = d1Q ( Q , nQ ),
a0 = d0Q ( Q , nQ ), b2 = d2P ( P , nP ), b1 = d1P ( P , nP ) and b0 = d0P ( P , nP ) are
simultaneously fullled. Note that both and n could in general take different values if
different dynamic behaviours for reactive and active power errors were required.
Considering the expressions for a2 , a1 , a0 , b2 , b1 and b0 provided in (44) and (45), as well
as those for d2 , d1 and d0 reected in (48), the latter conditions lead to the following tuning
equations:
c3Q d2Q c2Q + d1Q c Q d0Q = 0; c3P d2P c2P + d1P c P d0P = 0

 0.5
Q = 2 d2Q c Q Q
; P = 2 (d2P c P ) P0.5


w Q = d1Q c Q (d2Q c Q ) Q ; w P = [d1P c P (d2P c P )] P .

(49)
(50)
(51)

It is important to note that the coefcients in (49) coincide with those of target characteristic
equation (48), except for the signs of the squared and independent terms, which are negative.
It therefore turns out that the three possible values for c Q,P are equal to the roots poles of
target characteristic equation (48), although their real parts have opposite signs. Since the real
parts of the desired poles must necessarily be negative to ensure stability, the latter implies
that the real parts of the three possible values for c Q,P will always be positive. As a result,
given that expressions in (49) are third-order equations, it is guaranteed that at least one of the
three solutions for c Q,P will be both real and positive, as required.
Specically, depending on the value chosen for Q,P , one of the following three cases arises:
1. If 0 < Q,P < 1, only one of the three solutions for c Q,P is both real and positive,
c Q,P = Q,P nQ,P .
2. If Q,P = 1, two different acceptable solutions for c Q,P are obtained, c1Q,P = nQ,P and
c2Q,P = nQ,P .
3. If

the three


= nQ,P Q,P 2Q,P 1 ,
= Q,P nQ,P .

Q,P

c1Q,P
c3Q,P

>


1,

solutions
c2Q,P

for

are real and positive,




nQ,P Q,P + 2Q,P 1
and

c Q,P

For cases 2 and 3, two or three possible sets of values for c Q,P , Q,P and w Q,P are respectively
obtained. The set of parameters leading to the best performance may, for example, be
identied through simulation.
3.2.2 DFIG disconnected from the grid Synchronization stage

The design and tuning process of the current controllers, which synchronizes the voltage
induced at the open stator to that of the grid, is analogous to that presented in the preceding
section 3.2.1. Let the control law with respect to x -y reference frame be
+ vrx ; vry = vry + vry .
vrx = vrxST
eq
eq
ST

(52)

Taking the time derivatives of (10) and (11), the expressions given next arise if (3) and (4) are
considered:


1
Rr
crx irx sl iry vrx
(53)
s irx = irx re f + crx irx re f +
Lr
Lr


1
Rr
cry iry + sl irx vry .
(54)
s iry = iry re f + cry iry re f +
Lr
Lr

[Link]

Sensorless First- and Second-Order Sliding-Mode

Control of a Wind Turbine-Driven Doubly-Fed Induction Generator

121

The terms corresponding to the STA may be obtained as




vrx ST = Lr x | sirx |0.5 sgn(sirx ) + w x sgn(sirx )dt



0.5
vry ST = Lr y | siry | sgn(siry ) + wy sgn(siry )dt ,

(55)
(56)

where x , w x , y and wy are positive parameters to be tuned. As far as equivalent control

terms are concerned, they are derived by zeroing (53) and (54), thus yielding


Rr
(57)
irx sl iry + crx (irx re f irx )
vrx eq = Lr irx re f +
Lr


Rr
vry eq = Lr iry re f +
(58)
i + sl irx + cry (iry re f iry ) .
Lr ry

Again, note that all the control terms are premultiplied by a Lr gain in this case, which is the
inverse of that affecting control signals in (53) and (54). As mentioned before, its only purpose
is to facilitate the tuning of the parameters involved in the commanding algorithm.
Substitution of control law (52) into (53) and (54) leads to
s irx = x | sirx |0.5 sgn(sirx ) w x
s iry = y | siry |0.5 sgn(siry ) wy

sgn(sirx )dt

(59)

sgn(siry )dt,

(60)

expressions which turn out to be identical to those presented in (40) and (41) provided that
irx and x subscripts are respectively replaced by Qs and Q, and, likewise, iry and y
subscripts are interchanged with Ps and P. Therefore, the same reasoning detailed in section
3.2.1 can be followed in order to achieve the tuning equations of x , w x , c x , y , wy and cy
parameters. As a result,
c3x d2x c2x + d1x c x d0x = 0; c3y d2y c2y + d1y cy d0y = 0


y0.5
x = 2 (d2x c x ) x0.5
; y = 2 d2y c y



w x = [d1x c x (d2x c x )] x ; wy = d1y cy (d2y cy ) y

(61)
(62)
(63)

3.2.3 Bumpless connection

Considering that the DFIG presents different dynamics when disconnected or connected to the
grid, two STA-based controllers have been designed for generating a continuous command
signal. So far, the performance for each state has only been considered, but undesirable
phenomena may appear if the switch between the two controllers is not properly carried out.
If a direct transition is accomplished, a discontinuity arises in the command signal at the
instant of connection, due to the magnitude mismatch between the rotor voltages generated
by the two controllers. This effect produces high stator current values, leading the machine
to an excessive power exchange with the grid. Aiming to avoid this "bump", it is possible to
apply the same value of the control signal previous to and just after the transition k 1 and
k instants respectively; i.e.,

[Link]

vrx = vrx

(64)

vry = vry .

(65)

122

Sliding Mode Control

However, this way the "bump" is only delayed one sample time and it actually takes effect in
the next sampling instant k + 1.
A bumpless transition may take place if the above proposed solution is slightly modied
(strm & Hgglund, 1995). Setting the focus on the rotor voltage components when the
DFIG is connected to the grid, appropriate combination of (36), (38) and (37), (39) produces



2 L s Lr
Q | s Qs |0.5 sgn(s Qs ) + w Q sgn(s Qs )dt + Q s re f + c Q ( Qs re f Qs ) +
vrx =
3 |vs | L m

+ Rr irx Lr sl iry
(66)



2 L s Lr
vry =
P | s Ps |0.5 sgn(s Ps ) + w P sgn(s Ps )dt + Ps re f + c P ( Ps re f Ps ) +
3 |vs | L m
Lm
 s | + Lr sl irx .
+ Rr iry +
|
(67)
L s sl


Two integral terms, Isgn( s Qs ) = sgn(s Qs )dt and Isgn( s Ps ) = sgn(s Ps )dt, can be observed.
Their initial values, which are set to zero when connection occurs, are the source of the
mentioned "bump". Aiming at lessening or even eliminating this effect, it can be taken
advantage of (64) and (65) to calculate those initial values at connection time. Substituting
(66) and (67) into (64) and (65), respectively, leads to

3 |vs | L m 
v Rr irx + Lr sl iry
Isgn( s Qs ) =
0
2 L s Lr w Q rx
Q | s Qs |0.5 sgn(s Qs ) + Q s re f + c Q ( Qs re f Qs )

wQ


3 |vs | L m
Lm

v
Isgn( s Ps ) =

R
i

s
r ry
ry
r sl rx
0
2 L s Lr w P
L s sl
P | s Ps |0.5 sgn(s Ps ) + Ps re f + c P ( Ps re f Ps )
.

wP

(68)

(69)

3.3 Sensorless scheme Adaptation for synchronization

Both the 1-SMC and 2-SMC designs are combined with the MRAS observer proposed by (Pea
et al., 2008) in order to build two alternative sensorless control schemes. As a result, the
controller is provided with the estimated rotor electrical speed and position, thus avoiding
both the use of mechanical components encoders and the initial rotor positioning required
for the synchronization of the stator and grid voltages (Tapia et al., 2009). Morover, observers
may be used for chattering phenomenon alleviation (Utkin et al., 1999; Utkin, 1993).
It is worth pointing out that the MRAS observer must be adapted for the case of being
disconnected from the grid. On the one hand, since stator currents are null in this state,
calculation of stator ux in the stationary s D -s Q frame must be slightly modied
sQ =
sD =

(vsQ Rs isQ )dt sQ =

(vsD Rs isD )dt sD =

vsQ dt

(70)

vsD dt.

(71)

On the other hand, the stator voltages are those induced by the rotor currents for
synchronization, and present a considerable noise. For a proper control, the affected signals

[Link]

Sensorless First- and Second-Order Sliding-Mode

Control of a Wind Turbine-Driven Doubly-Fed Induction Generator

123

are ltered out by means of a 500-Hz bandwidth second-order Butterworth lter, which in
turn produces a phase lag of = 8.1297 at 50Hz. This lag must be compensated in order
to estimate the components of the actual stator voltage phasor. Considering Fig. 5, it can be
stated that

|vs | = |vs f iltered| = v2sD f iltered + v2sQ f iltered
(72)
= arctan

vsQ

f iltered

vsD

f iltered

(73)

Furthermore, given that the phase lag is known deriving the argument of vs from Fig. 5 as
arg(vs ) = + , the components of the actual stator voltage space-phasor given next arise:
vsD = |vs | cos( + )

(74)

vsQ = |vs | sin( + ).

(75)

s Q

s Q

H
s

s Q filte r e d

g
v

H
s filte r e d

s D

s D
s D filte r e d

Fig. 5. Actual and ltered stator voltage space-phasors

4. Hardware-in-the-loop results
The presented sensorless 1-SMC and 2-SMC algorithms are evaluated, through real-time HIL
emulation, over a full-detail virtual DFIG prototype running on eMEGAsim OP4500 F11-13
simulator by OPAL-RT. The electric parameters of the 660-kW DFIG under consideration are
collected in Table 1.
Aiming at showing some of the most illustrative results of the two alternative control
algorithms put forward, the test whose main events are reected in Table 2 is conducted.
It should be pointed out that what causes the DFIG control system to generate the order of
connection to the grid, taking place at second 0.474, is the DFIG rotational speed exceeding
the already mentioned threshold of 1270 rpm.
The 1-SMC algorithm itself is implemented on a Virtex-II Pro series FPGA by Xilinx, which
allows reaching the 40-kHz sampling rate required to avoid causing excessive chatter. Direct
measurement of the grid voltage allows accurately computing angle s , and, as a result,
identifying the exact position of the x -y reference frame. On the other hand, the s angle,
which provides the location of the stator-ux-oriented x-y reference frame, is derived from
the direct (sD ) and quadrature (sQ ) stationary-frame components of the stator ux. These
are in turn estimated by integration of the stator voltage minus the resistive drop. A digital
bandpass lter is used as a modied integrator to avoid drift (Pea et al., 2008). Regarding

[Link]

124

Sliding Mode Control

PARAMETER

VALUE

Rated r.m.s. stator voltage

Rated peak rotor voltage
Rated peak rotor current
Stator resistance per phase, Rs
Stator inductance per phase, L s
Magnetizing inductance, L m
Rotor resistance per phase, Rr
Rotor inductance per phase, Lr
General turns ratio, n
Number of pole pairs, P

398/690 V
380 V
400 A
6.7 m
7.5 mH
19.4 mH
39.9 m
52 mH
0.3806
2

Table 1. DFIG electric parameters

E VENT
Order of connection to the grid;
start of synchronization process;
initial convergence of the observer
End of synchronization process;
connection to the grid at zero power
Start of power generation
Sudden increase of wind speed
Sudden decrease of wind speed

T IME INSTANT ( S )
0.474
1.474
1.974
7
12

Table 2. Main events of the designed test

switching variables, in this particular case, c Q and c P are set to 10, while c x and cy are made
equal to 0. In addition, the integral terms in s Ps and s Qs are discretized by applying Tustins
trapezoidal method (Kuo, 1992).
Fig. 6(a) displays a general portrait of the synchronization stage. In addition, details at both
its beginning and its end are reected in Figs. 6(b) and 6(c), respectively. The former evidences
the rapid dynamic response of vsA , vsB and vsC voltages, induced at the terminals of the DFIG
open stator, when synchronizing with v grid A , v grid B and v grid C grid voltages. As expected,
no active and reactive powers are exchanged between the DFIG stator and the grid during
synchronization, as corroborated by Figs. 7(a) and 7(b).
The DFIG is then connected to the grid at zero power, until power generation according to
the optimum power curve is launched 0.5 s later. Fig. 7(a) shows the excellent performance
of the stator-side active power, Ps , when, as a result of the two sudden wind speed changes
occurring at seconds 7 and 12, a great part of the optimum power curve is tracked both up
and downwards. The also superior tracking of the target stator-side reactive power, Qs re f ,
is evidenced in Fig. 7(b). The instantaneous reference value for Qs is xed so that the DFIG
operates with a 0.95 leading capacitive power factor all through the test. Chatter in Ps
and Qs represents only 3% of the rated power.
Given that, as indicated above, control signals are updated at a 40-kHz sample rate, gating
signals swk ; k = 1, 2 . . . 6, are able to toggle every 25 s, if required. This results in a maximum
switching frequency of 20 kHz with 50% duty cycle. However, switching frequency of the RSC
insulated gate bipolar transistors (IGBTs) is variable, as dictated by switching functions s1 , s2
and s3 . This is clearly observable in Fig. 8(a), where the frequency spectrum corresponding to

[Link]

Sensorless First- and Second-Order Sliding-Mode

Control of a Wind Turbine-Driven Doubly-Fed Induction Generator

125

(a) Grid and stator voltages at the synchronization stage

0.5
0

Grid and stator voltages (kV)

0.5
0

vs a
vs b
vs c
vgrid a
vgrid b
vgrid c
0.5
1
1.5
(b) Detail at the synchronization stage

0.55
0.65
(c) Grid connection detail

0.75

1.45

1.55

0.5
0
0.5
0.45

0.5
0
0.5
1.4

1.5
Time (s)

Fig. 6. Stator and grid voltages of the 1-SMC controller-driven DFIG at the synchronization
stage
gating signal sw1 is displayed. The three-phase rotor current resulting from the gating signals
applied to the RSC IGBTs is that displayed in Fig. 8(b). As expected, it turns out to be variable
in magnitude, frequency and phase.
Fig. 9 reects the performance of the digital MRAS observer, which operates at a 1-kHz sample
rate. In this particular case, it is incorporated not only for sensorless control, but also as
a supporting tool for chatter attenuation (Utkin et al., 1999; Utkin, 1993). The observer is
launched once the order of connection to the grid is automatically generated. As evidenced
in Fig. 9(a), the estimated rotor mechanical speed converges rapidly to its actual value at
the earlier part of the synchronization stage, and, from that point onwards, keeps track of it
satisfactorily in spite of the transition from the disconnected state to the connected one taking
place at second 1.474. Moreover, a detail illustrating the fast convergence of the estimated
rotor electrical position to its actual value is displayed in Fig. 9(b).
As far as the sensorless 2-SMC algorithm is concerned, it is programmed in C language on
a DSP-based board. Control signals vrx and vry are demodulated to derive the vr and vr
voltage components, expressed in the rotor natural reference frame, which are then supplied
as inputs to the SVM algorithm generating the gating signals of RSC IGBTs. Angles s and
s , required both to estimate equivalent control terms and to demodulate vrx and vry control
signals, are derived in the same manner as for the 1-SMC algorithm. Both the 2-SMC and
SVM algorithms operate at a 5-kHz sample rate, while the MRAS observer runs, as in the
preceding case, at 1 kHz. The integral terms included in both the switching functions and
the STA algorithm itself are digitally implemented based on Tustins trapezoidal method. Yet,
aiming to elude the risk of causing derivative ringing (strm & Hgglund, 1995), Eulers
rectangular method is applied to discretize the time derivatives of Ps re f and Qs re f appearing
in equivalent control terms.

[Link]

126

Sliding Mode Control

(a)
Active power (kW)

200
0
200
400
600

Ps
Ps ref

800
0

10

12

14

16

8
10
Time (s)

12

14

16

(b)
Reactive power (kVAr)

200
0
200
400
600

Qs
Qs ref

800
0

Fig. 7. Active and reactive powers of the DFIG commanded by the 1-SMC controller
(a)
1

0.6

w1

|s (j)|

0.8

0.4
0.2
0

8
10
12
Frequency (kHz)
(b)

14

16

18

20

Rotor currents (A)

400
200
0
ir a
ir b
ir c

200
400
0

8
10
Time (s)

12

14

16

Fig. 8. Frequency spectrum of sw1 gating signal, and resulting three-phase rotor current
Selecting a x ,y = 0.01 A, control parameters for synchronization are adjusted seeking to
reach closed-loop rotor current error dynamics exhibiting a unit damping coefcient and a
nx ,y = 55.2381 rad/s natural frequency while in sliding regime. As a result, if different

[Link]

Sensorless First- and Second-Order Sliding-Mode

Control of a Wind Turbine-Driven Doubly-Fed Induction Generator

127

Mechanical speed (rad/s)

(a)
200
150
100
50
0
0

r m

r m
2

10

12

14

16

Electrical position (rad)

(b) Detail at the beginning of the position estimation

6

r
r

4
2
0
0.45

0.5

0.55

0.6

0.65

0.7

Time (s)

Fig. 9. Actual and estimated mechanical speed and electrical position of the DFIG rotor
from zero, errors in rotor current irx and iry components would vanish, according to the
2% criterion (Ogata, 2001), in 105 ms, showing no overshoots. Similarly, forcing P,Q to be
equal to 0.1 kW, and specifying P,Q = 1 and nP,Q = 82.8571 rad/s, respectively, as target
damping coefcient and natural frequency for closed-loop power error dynamics, possible
errors arising in active and reactive powers would theoretically decay to zero in 70 ms, with no
overshoots. If is made equal to 10, the values resulting for the 2-SMC algorithm parameters
are those collected in Table 3.
PARAMETER
c x , cy
x , y
w x , wy
cP , cQ
P , Q
wP, wQ

VALUE
55.2381
121.5238
305.1247
82.8571
1.8229 104
6.8653 106

Table 3. Values for the parameters belonging to the 2-SMC algorithm

The most signicant results corresponding to the applied sensorless SVM-based 2-SMC
are illustrated by Figs. 10, 11 and 12. Given that those gures are very similar to
their corresponding 1-SMC counterparts Figs. 6, 7 and 8, respectively, only the main
differences between them will accordingly be commented on.
Comparison of Figs. 6 and 7 with Figs. 10 and 11, respectively, reveals that the resulting chatter
is somewhat lower for the 2-SMC case. In addition, synchronization of the voltage induced in
the DFIG open stator to that of the grid is achieved faster when applying the 1-SMC algorithm,

[Link]

128

Sliding Mode Control

(a) Grid and stator voltages at the synchronization stage
0.5
0

Grid and stator voltages (kV)

0.5
0

vs a
vs b
vs c
vgrid a
vgrid b
vgrid c
0.5
1
1.5
(b) Detail at the synchronization stage

0.55
0.65
(c) Grid connection detail

0.75

1.45

1.55

0.5
0
0.5
0.45

0.5
0
0.5
1.4

1.5
Time (s)

Fig. 10. Stator and grid voltages of the 2-SMC controller-driven DFIG at the synchronization
stage
as evidenced by Figs. 6 and 10. Furthermore, even though the 2-SMC algorithm is provided
with bumpless transfer from the disconnected state to the connected one, Figs. 7 and 11 prove
that power exchange with the grid at the instant of connection is considerably lower for the
case of the 1-SMC.
On the other hand, since the SVM-based 2-SMC leads to a 5-kHz constant switching frequency
of the RSC IGBTs, in Fig. 12(a) the frequency spectrum of Fig. 8(a) has been replaced with the
vr and vr voltage components supplied as inputs to the SVM algorithm. The smoothness
of vr and vr in Fig. 12(a) indicates that, like in classical PI controller-based FOC schemes,
chatter in Ps and Qs observable in Fig. 11 is just attributable to SVM, not to the 2-SMC
algorithm itself.
To conclude, as it turns out that the MRAS observer performance is extremely similar to that
resulting in the case of the sensorless 1-SMC, it is not included here to avoid reiteration.

5. Conclusion
Real-time HIL emulation results obtained by running sensorless versions of the 1-SMC and
2-SMC arrangements presented in this chapter reveal that excellent tracking of a predened
rotor speed-dependent optimum power curve is reached in both cases. In addition, prior
to connecting the DFIG stator to the grid, they are also capable of achieving satisfactory
synchronization of the voltage induced at the open stator terminals to that of the grid.
In any case, it may be of interest to contrast both SMC algorithms, so as to identify the
strengths and weaknesses associated to each of them. This section will hence focus on that
comparison.
As far as the complexity of the algorithm itself is concerned, the 1-SMC version turns out
to be considerably simpler than the 2-SMC one. Given that the control signals generated

[Link]

Sensorless First- and Second-Order Sliding-Mode

Control of a Wind Turbine-Driven Doubly-Fed Induction Generator

129

(a)
Active power (kW)

200
0
200
400
Ps
Ps ref

600
800
0

10

12

14

16

Reactive power (kVAr)

(b)
200
0
200
400
Qs
Qs ref

600
800
0

8
10
Time (s)

12

14

16

Fig. 11. Active and reactive powers of the DFIG commanded by the 2-SMC controller
(a)
Rotor voltage (kV)

0.4
0.2
0
0.2
0.4
0

vr
vr
2

10

12

14

16

(b)
Rotor currents (A)

400
200
0
ir a
ir b
ir c

200
400
0

8
10
Time (s)

12

14

16

Fig. 12. Rotor voltage components fed into the SVM algorithm, and resulting three-phase
rotor current
by the 1-SMC correspond to the gating signals of the RSC IGBTs, no additional modulation
techniques such as pulse-width modulation (PWM) or SVM are required. In contrast, the

[Link]

130

Sliding Mode Control

control signals produced by the proposed 2-SMC correspond to continuous voltage direct and
quadrature components to be applied to the rotor by means of the RSC, which implies the
use of intermediate SVM modulation. Furthermore, providing an additional procedure for
bumpless transition between the algorithms devoted to synchronization and power control is
indispensable for the case of the 2-SMC, but it is not required for the 1-SMC scheme.
Regarding parameter tuning, only the c constants included in the four switching functions
considered need to be tuned for the case of the 1-SMC. It therefore turns out that satisfactory
parameter adjustment is easily achieved by mere trial and error. However, in addition to those
c constants, the and w gains present in the STAs must also be tuned for the 2-SMC variant.
Even though, as stated in (Bartolini et al., 1999), it is actually the most common practice,
trial and error tuning is not particularly effective in this latter case, as it may become highly
time-consuming. Therefore, it is believed that there exists a strong need for development of
alternative methods for STA-based 2-SMC tuning.
Concerning the switching frequency of the RSC IGBTs, it is xed at 5 kHz in the case of the
2-SMC. On the contrary, it turns out to be variable, within the range from 0 to 20 kHz, for
the 1-SMC algorithm. This feature complicates the design of both the back-to-back converter
feeding the DFIG rotor and the grid-side AC lter, since broadband harmonics may be injected
into the grid. As a result of the 25-s sample time selected for the 1-SMC scheme, which
leads to the aforementioned maximum switching frequency of 20 kHz, chatter observable
in stator-side active and reactive powers is somewhat lower than 3% of the DFIG 660-kW
rated power. Even a lower level of chatter arises from application of the SVM-based 2-SMC
algorithm put forward. Furthermore, that chatter, or at least great part of it, is caused by the
SVM, not by the 2-SMC algorithm itself.
Apart from the superior optimum power curve tracking achieved with both alternative SMC
designs, the dynamic performance resulting from realization of the proposed 1-SMC scheme
is noticeably better than that to which application of its 2-SMC counterpart leads. In effect,
focusing on the state in which the DFIG stator is disconnected from the grid, HIL emulation
results demonstrate that synchronization is reached faster by employing the 1-SMC algorithm.
On the other hand, the power exchange between the DFIG and the grid taking place at the
initial instants after connection is signicantly lower when adopting the 1-SMC algorithm put
forward, hence evidencing that its dynamic performance is also better for the stage during
which power control is dealt with. The excellent dynamic performance reachable by means of
its application supports the 1-SMC approach as a potential candidate for DFIG control under
grid faults, where rapidity of response becomes crucial.
The main conclusions drawn from the comparison conducted in this section are summarized
in Table 4.
1-SMC ALGORITHM
A LGORITHM COMPLEXITY
Relatively simple
PWM/SVM
Not required
B UMPLESS PROCEDURE
Not required
Straightforward
PARAMETER TUNING
S WITCHING FREQUENCY Variable from 0 to 20 kHz
3% of the rated power
C HATTER LEVEL
D YNAMIC PERFORMANCE
Excellent

2-SMC ALGORITHM
More complex
Required
Required
Complex
Fixed at 5 kHz
Lower
Very good

Table 4. Comparison between the two SMC algorithms put forward

[Link]

Sensorless First- and Second-Order Sliding-Mode

Control of a Wind Turbine-Driven Doubly-Fed Induction Generator

131

6. References
Abo-Khalil, A. G., Lee, D.-C. & Lee, S.-H. (2006). Grid connection of doubly fed induction
generators in wind energy conversion system, Proceedings of the CES/IEEE 5th
International Power Electronics and Motion Control Conference (IPEMC 2006), Shanghai,
China, vol. 3, pp. 15.
Arnaltes, S. & Rodrguez, J. L. (2002). Grid synchronisation of doubly fed induction generators
using direct torque control, Proceedings of the IEEE 28th Annual Conference of the
Industrial Electronics Society (IECON 2002), Seville, Spain, pp. 33383343.
strm, K. J. & Hgglund, T. (1995). PID Controllers: Theory, Design and Tuning, Instrument
Society America, USA.
Bartolini, G., Ferrara, A., Levant, A. & Usai, E. (1999). On second order sliding mode
uner
Ozg
controllers, in K. Young & U.

(eds), Variable Structure Systems, Sliding Mode

and Nonlinear Control, Springer Verlag, London, UK, pp. 329350.
Beltran, B., Ahmed-Ali, T. & Benbouzid, M. E. H. (2008). Sliding mode power control
of variable-speed wind energy conversion systems, IEEE Transactions on Energy
Conversion 23(2): 551558.
Beltran, B., Ahmed-Ali, T. & Benbouzid, M. E. H. (2009). High-order sliding-mode
control of variable-speed wind turbines, IEEE Transactions on Industrial Electronics
56(9): 33143321.
Beltran, B., Benbouzid, M. E. H. & Ahmed-Ali, T. (2009). High-order sliding mode control
of a DFIG-based wind turbine for power maximization and grid fault tolerance,
Proceedings of the IEEE International Electric Machines and Drives Conference (IEMDC
2009), Miami, USA, pp. 183189.
Ben Elghali, S. E., Benbouzid, M. E. H., Ahmed-Ali, T., Charpentier, J. F. & Mekri, F. (2008).
High-order sliding mode control of DFIG-based marine current turbine, Proceedings
of the IEEE 34th Annual Conference of the Industrial Electronics Society (IECON 2008),
Orlando, USA, pp. 12281233.
Blaabjerg, F., Teodorescu, R., Liserre, M. & Timbus, A. V. (2006). Overview of control and
grid synchronization for distributed power generation systems, IEEE Transactions on
Industrial Electronics 53(5): 13981409.
Ekanayake, J. B., Holdsworth, L., Wu, X. & Jenkins, N. (2003). Dynamic modeling of
doubly fed induction generator wind turbines, IEEE Transactions on Power Systems
18(2): 803809.
Kuo, B. C. (1992). Digital Control Systems, Oxford University Press, New York, USA.
Levant, A. (1993). Sliding order and sliding accuracy in sliding mode control, International
Journal of Control 58(6): 12471263.
Ogata, K. (2001). Modern Control Engineering, Prentice Hall, Englewood Cliffs (New Jersey),
USA.
Pea, R., Crdenas, R., Proboste, J., Asher, G. & Clare, J. (2008). Sensorless control of
doubly-fed induction generators using a rotor-current-based MRAS observer, IEEE
Transactions on Industrial Electronics 55(1): 330339.
Pea, R., Clare, J. C. & Asher, G. M. (1996). Doubly fed induction generator using back-to-back
PWM converters and its application to variable-speed wind-energy generation, IEE
Proceedings - Electric Power Applications 143(3): 231241.
Peresada, S., Tilli, A. & Tonielli, A. (2004). Power control of a doubly fed induction machine
via output feedback, Control Engineering Practice 12(1): 4157.

[Link]

132

Sliding Mode Control

Rashed, M., Dunnigan, M. W., MacConell, P. F. A., Stronach, A. F. & Williams, B. W.

(2005).
Sensorless second-order sliding-mode speed control of a voltage-fed
induction-motor drive using nonlinear state feedback, IEE Proceedings-Electric Power
Applications 152(5): 11271136.
Susperregui, A., Tapia, G., Zubia, I. & Ostolaza, J. X. (2010). Sliding-mode control of
doubly-fed generator for optimum power curve tracking, IET Electronics Letters
46(2): 126127.
Tapia, A., Tapia, G., Ostolaza, J. X. & Senz, J. R. (2003). Modeling and control of a
wind turbine driven doubly fed induction generator, IEEE Transactions on Energy
Conversion 12(2): 194204.
Tapia, G., Santamara, G., Telleria, M. & Susperregui, A. (2009). Methodology for smooth
connection of doubly fed induction generators to the grid, IEEE Transactions on Energy
Conversion 24(4): 959971.
Tapia, G., Tapia, A. & Ostolaza, J. X. (2006). Two alternative modeling approaches for the
evaluation of wind farm active and reactive power performances, IEEE Transactions
on Energy Conversion 21(4): 901920.
Utkin, V., Guldner, J. & Shi, J. (1999). Siliding Mode Control in Electromechanical Systems, Taylor
& Francis, London, UK.
Utkin, V. I. (1993). Sliding mode control design principles and applications to electric drives,
IEEE Transactions on Industrial Electronics 40(1): 2336.
Vas, P. (1998). Sensorless Vector and Direct Torque Control of AC Machines, Oxford University
Press, New York, USA.
Xu, L. & Cartwright, P. (2006). Direct active and reactive power control of DFIG for wind
energy generation, IEEE Transactions on Energy Conversion 21(3): 750758.
Yan, W., Hu, J., Utkin, V. & Xu, L. (2008). Sliding mode pulsewidth modulation, IEEE
Transactions on Power Electronics 23(2): 619626.
Yan, Z., Jin, C. & Utkin, V. I. (2000). Sensorless sliding-mode control of induction motors, IEEE
Transactions on Industrial Electronics 47(6): 12861297.
Zhi, D. & Xu, L. (2007). Direct power control of DFIG with constant switching frequency
and improved transient performance, IEEE Transactions on Energy Conversion
22(1): 110118.

[Link]

Sliding Mode Control

Edited by Prof. Andrzej Bartoszewicz

ISBN 978-953-307-162-6
Hard cover, 544 pages
Publisher InTech

Published online 11, April, 2011

Published in print edition April, 2011

The main objective of this monograph is to present a broad range of well worked out, recent application
studies as well as theoretical contributions in the field of sliding mode control system analysis and design. The
contributions presented here include new theoretical developments as well as successful applications of
variable structure controllers primarily in the field of power electronics, electric drives and motion steering
systems. They enrich the current state of the art, and motivate and encourage new ideas and solutions in the
sliding mode control area.

How to reference

In order to correctly reference this scholarly work, feel free to copy and paste the following:
Ana Susperregui, Gerardo Tapia and M. Itsaso Martinez (2011). Sensorless First- and Second-Order SlidingMode Control of a Wind Turbine-Driven Doubly-Fed Induction Generator, Sliding Mode Control, Prof. Andrzej
Bartoszewicz (Ed.), ISBN: 978-953-307-162-6, InTech, Available from:
[Link]

InTech Europe

University Campus STeP Ri

Slavka Krautzeka 83/A
51000 Rijeka, Croatia
Phone: +385 (51) 770 447
Fax: +385 (51) 686 166
[Link]

InTech China

Unit 405, Office Block, Hotel Equatorial Shanghai

No.65, Yan An Road (West), Shanghai, 200040, China
Phone: +86-21-62489820
Fax: +86-21-62489821