Sie sind auf Seite 1von 9

This article was downloaded by: [North Carolina State University]

On: 24 September 2012, At: 15:13


Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House,
37-41 Mortimer Street, London W1T 3JH, UK

International Journal of Control


Publication details, including instructions for authors and subscription information:
http://www.tandfonline.com/loi/tcon20

Design of first- and second-order sliding mode


observers for induction motors using a stator-flux
model
a b c
Sachit Rao , Vadim Utkin & Martin Buss
a
Department of Aerospace Engineering, Indian Institute of Science, Bangalore, Karnataka,
India
b
Department of Electrical Engineering, Ohio State University, Columbus, OH, USA
c
Institute of Automatic Control Engineering (LSR), Technische Universität München, Munich,
Germany

Version of record first published: 23 Jun 2010.

To cite this article: Sachit Rao, Vadim Utkin & Martin Buss (2010): Design of first- and second-order sliding mode observers
for induction motors using a stator-flux model, International Journal of Control, 83:7, 1457-1464

To link to this article: http://dx.doi.org/10.1080/00207171003793270

PLEASE SCROLL DOWN FOR ARTICLE

Full terms and conditions of use: http://www.tandfonline.com/page/terms-and-conditions

This article may be used for research, teaching, and private study purposes. Any substantial or systematic
reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to
anyone is expressly forbidden.

The publisher does not give any warranty express or implied or make any representation that the contents
will be complete or accurate or up to date. The accuracy of any instructions, formulae, and drug doses should
be independently verified with primary sources. The publisher shall not be liable for any loss, actions, claims,
proceedings, demand, or costs or damages whatsoever or howsoever caused arising directly or indirectly in
connection with or arising out of the use of this material.
International Journal of Control
Vol. 83, No. 7, July 2010, 1457–1464

Design of first- and second-order sliding mode observers for induction


motors using a stator-flux model
Sachit Raoa*, Vadim Utkinb and Martin Bussc
a
Department of Aerospace Engineering, Indian Institute of Science, Bangalore, Karnataka,
India; bDepartment of Electrical Engineering, Ohio State University, Columbus, OH,
USA; cInstitute of Automatic Control Engineering (LSR), Technische Universität
München, Munich, Germany
(Received 4 February 2010; final version received 19 March 2010)
Downloaded by [North Carolina State University] at 15:13 24 September 2012

We extend current research in the area of ‘sensorless’ control of induction motors by presenting two observers
based on first- and second-order sliding mode control theories for the simultaneous estimation of flux and speed.
We base the observers on the stator-flux model of the motor instead of the usual rotor-flux model mainly because
of the uncertain rotor resistance that plays a significant role in the latter. By designing the observers as if they are
sliding mode controllers, we lend the properties of parameter insensitive closed-loop dynamics and finite time
convergence to the stator flux and speed estimation schemes. We also present simulation and experimental results
to validate the operation of the observers.
Keywords: induction motors; observers; sliding mode control; stator-flux model; super twisting algorithm

1. Introduction model is represented in the field-oriented domain, flux


The induction motor (IM), a device which does not and current can be independently controlled using the
need much of an introduction, has proved itself to be a input voltages. These are the two main reasons why we
reliable workhorse in industry. As this device has been choose the stator-flux model.
widely studied, we keep the background details of the Both the rotor-flux and stator-flux models have
IM at a minimum and direct the interested reader to been used to design observers. Proca and Keyhani
Krause (1986) and Leonhard (2001) for IM funda- (2007), Yan, Jin, and Utkin (2000), Yan and Utkin
mentals. As the IM is mainly used as an actuator, the (2002), Comanescu and Xu (2006), Kenne, Ahmed-Ali,
standard control problem is that of speed control – we Lamnabhi-Lagarrigue, and Arzande (2008, 2009) and
tackle this problem here. In order to do so, speed and Rao, Buss, and Utkin (2009) are a few resources for
flux should be measured either by using sensors or by sliding mode-based observer designs based on the
model-based observers. We choose the latter approach former; Kenne et al. include adaptive mechanisms to
for two basic reasons: cost (of sensors) and because the estimate RR and the motor load torque. In the case of
standard IM models reflect the actual system closely. the stator-flux model, the most common solution to
The usual IM models are in terms of rotor or stator estimate stator-fluxes is to integrate the flux dynamics
fluxes and stator currents written in the stationary – due to their simplicity – and then correct any offsets
(, ) frame of reference. Here, we adopt the stator-flux (Holtz and Quan 2002). The correcting mechanisms
model mainly because the temperature dependent, and have been designed using: asymptotic observers in
hence uncertain, rotor resistance (RR) does not play a Marchesoni, Seqarich, and Soressi (1997); adaptive
direct role in the observer design. An observer based low-pass filters in Shin, Hyun, Cho, and Choe (2000);
on the rotor-flux model will depend on RR and hence extended Kalman filters based on linearised IM models
that observer, for it to yield accurate estimates, should in Barut, Bogosyan, and Gokasan (2007); simple PI
adapt to the time-varying resistance value. But, controllers based on simplified IM models in Boussak
observers designed on the stator-flux model are free and Jarray (2006); and even sliding mode control
of this parametric uncertainty. Moreover, when this (SMC) theory in Kheloui, Aliouane, Medjaoui, and

*Corresponding author. Email: sachit@aero.iisc.ernet.in

ISSN 0020–7179 print/ISSN 1366–5820 online


ß 2010 Taylor & Francis
DOI: 10.1080/00207171003793270
http://www.informaworld.com
1458 S. Rao et al.

Davat (2000) – this work treats speed as a perturbation where , are the stator fluxes; i, are the stator
that is suppressed during flux estimation. currents; ! is the rotor angular speed; u, are the input
We use results from conventional and higher order stator voltages and the positive constants
SMC theories to design observers for flux and speed
estimation. Our motivation to enforce sliding modes L2M RR RS þ ðL2M =L2R ÞRR
 ¼1 , ¼ , s ¼ P þ
for state estimation is to make the estimates converge LS LR LR LS
to their real counterparts in finite time. The observer
designs proposed in this article do not require the that are functions of the self and mutual inductances
integration of stator fluxes and hence avoid offset LS,R,M and the stator and rotor resistances RS,R. J, P
corrections. As we will show, speed and flux are and TL are the rotor moment of inertia, number of
simultaneously estimated by simply enforcing sliding pole-pairs and the load torque, respectively.
mode on some switching surfaces – whose selection By following the idea of cascaded control, we select
forms the main body of work. the motor torque Ti to control ! and then use the
This article is organised as follows: in Section 2, the motor input voltages u, to provide this torque. We
problem tackled in this article is formally defined; in choose SMC theory to design the input voltages as
Downloaded by [North Carolina State University] at 15:13 24 September 2012

Section 3, a brief overview of first- and second-order well. The reason is simple: conventional sliding modes
sliding modes as well as their contributions to the IM require rapidly switching discontinuous control inputs
observers is provided; in Section 4, the speed and flux and such inputs can be provided by power converters
control laws as well as their requirements are stated; that drive the IMs.
the observers, along with their convergence properties, We design the observers under the following
which yield the flux and speed estimates are presented assumptions:
in Section 5; and finally, in Sections 6 and 7, the (1) The time-varying rotor resistance RR does not
experimental results of the performance of the pro- affect  s greatly. This assumption is not very
posed observers are shown and concluding comments restrictive since the stator resistance RS 4 RR,
are made, respectively. usually, and the inductances LS,R are small
(to ensure fast motor dynamics);
(2) the stator currents i, and the input voltages
2. Problem statement u, can be measured;
We address the following problems: the control of (3) RS and the inductances LS,R,M are known and
motor speed using its estimate as required by the are time-invariant.
‘sensorless’ control problem and the estimation and
control of stator fluxes because the motor torque,
which we use for speed control, depends on them. As
solutions, we offer two sliding mode-based observer 3. Sliding modes
designs which estimate the unknowns. The reason we We design two observers, both of which perform the
choose SMC theory is to be able to provide finite time same task, but whose inputs are designed based on
convergence of the estimates to the actual values – this first- and second-order SMC algorithms, respectively.
is in contrast with the Luenberger-type observers In this section, we give a brief overview of these
which, formally, assure only asymptotic convergence. algorithms. The inputs to the first type of observer –
We design the controllers and the observers based on termed the FOSM observer – are designed based on the
the model first-order sliding mode algorithm (Utkin 1992), and
P u for the second – termed the SOSM observer – the
i_ ¼ s i  P!i þ ð  þ! Þ þ ð1aÞ inputs are selected to enforce the second-order sliding
LS LS
mode based on the super-twisting algorithm (Fridman
P u and Levant 2002).
i_ ¼ P!i  s i  ð!   Þ þ ð1bÞ
LS LS

_  ¼ RS i þ u ð1cÞ 3.1 Forms


The general procedure for the design of first- or
_  ¼ RS i þ u ð1dÞ second-order SMCs for control-affine systems can be
outlined in only two steps. For brevity, we consider
3 LM the simple system s_ ¼ u, s, u 2 <, where u should be
J!_ ¼ PðTi  TL Þ; Ti ¼ P ð  i   i Þ, ð1eÞ
2 LR selected so that s ! 0. To enforce first-order sliding
International Journal of Control 1459

mode, the control input is selected as u ¼ M sign(s), convergence properties of the observers. These laws
M 4 0, and as are initially derived under the assumption that all
pffiffiffiffiffi states, including speed and flux, are known. We use
u ¼  jsj signðsÞ þ v the field-oriented scheme to design the controllers,

u if juj 4 M, ð2Þ and so transform the electrical equations in (1) to
v_ ¼ the model
 signðsÞ if juj  M
 
to enforce the second-order sliding mode, , , M 4 0. _iq ¼ s iq þ P!id  P! d þ RS iq id þ 1  id uq
In the case of the FOSM controller, the equality s ¼ 0 LS d LS d
will hold after a finite time interval, and in the case of ð3aÞ
the SOSM controller, s ¼ s_ ¼ 0 will be true, again in
finite time. It is the property of finite-time convergence, P 1 iq  
offered by both control algorithms, that we exploit for i_d ¼ s id  P!iq þ d þ ud  RS iq  uq
LS LS d
our observers. To our knowledge, the super-twisting
ð3bÞ
algorithm has not been previously applied for state
Downloaded by [North Carolina State University] at 15:13 24 September 2012

estimation in IMs.
_ d ¼ RS id þ ud , ð3cÞ

where the new states are given by


3.2 Contributions i   i  i  þ i 
iq ¼ , id ¼ ;
We offer the following benefits with our approach: d d
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
. Machine information: Both observers are ¼ 2 þ 2, ¼ 0; and ð4aÞ
d   q
based only on the electrical equations of the
IM, (1a)–(1d), so exact knowledge of mechan-
u   u  u  þ u 
ical parameters is not necessary. Moreover, we uq ¼ , ud ¼ : ð4bÞ
d d
only need measure currents and input
voltages. In the transformed model, the motor torque is given by
. Accuracy: As both observers will be imple-
mented in the discrete form, say with a 3 LM
Ti ¼ P d iq : ð5Þ
sampling interval D, the FOSM observer, 2 LR
with its accompanying chattering issue, will We select
provide estimates with an accuracy propor-
tional to D (Utkin and Lee 2006). On the other J
Ti ¼  ð!ref  !Þ þ TL ð6Þ
hand, the SOSM observer provides chattering- P
free estimates with an accuracy proportional as the speed control law where !ref is the desired motor
to D2. As these observers will anyway be speed, and the constant  4 0. When this law is
implemented on fast computers, a small D can substituted in the motor mechanical equation (1e),
always be chosen to improve performance. ! ! !ref at a rate decided by . To generate the desired
. Tuning: The FOSM observer requires the motor torque, say Ti , current iq and flux d should be
tuning of the discontinuous control magnitude
controlled to, say iq and d , so that they satisfy (5).
and the time-constant of a first-order low-pass
Hence the primary speed control problem has
filter. If the SOSM observer is used, then the
now become the two sub-problems of controlling d
gains ,  and M have to be tuned. Note that
and iq – for which we design the input voltages ud and
the respective theories that form the founda-
uq, respectively, as outputs of conventional SMCs. We
tions of these algorithms do provide upper-
assume in the rest of this article, for simplicity, that !ref
and/or lower-bounds of these parameters.
is a constant; the results that we obtain can be easily
modified even when !ref is time-varying. With this
assumption, iq , d also become constants.
4. Control laws Remarks: The decoupled nature of Equations (3a)
Before we proceed with the observers’ designs, and (3c) with respect to the real control inputs ud,q and
we state the sliding mode control laws for flux and the lack of direct dependence on rotor resistance RR,
speed. We state them at this stage itself as some of which are attributes offered by the stator-flux model is
the intermediate results will be used to prove the now evident.
1460 S. Rao et al.

4.1 Control of wd For simplicity, we analyse the linearised dynamics


To control flux, we apply the results of conventional of x in the vicinity of its equilibrium point, x*, which,
sliding mode theory (Utkin 1992). Hence, first define for constant d , iq , is given by
the switching surface pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1  1  KC
sd ¼  
ð7Þ x ¼ , KC ¼ 4KðCLS Þ2 :
d d 2CLS
and evaluate s_d ¼ RS id þ ud . Next, select ud ¼ Md  As id, are both real, we assume that
d
sign(sd) with Md 4 jRSidj 4 0 so that sliding mode
occurs on sd ¼ 0 and as a result d ¼ d in finite time. ð1  KC Þ 4 0 ) x 4 0: ð14Þ
We also calculate the equivalent control udeq as the
solution to s_d ¼ 0: Thus, the linearised
 dynamics,
 which can be shown
to be x_ ¼  1  xK2 x, is asymptotically stable
udeq ¼ RS id : ð8Þ
about x*, and so iq – hence speed – can indeed be
controlled. But, to do so, we would need to know the
Downloaded by [North Carolina State University] at 15:13 24 September 2012

stator-fluxes and of course the speed itself. In the


next section, we present the observers which will
estimate them.
4.2 Control of iq
We follow the same procedure to control current.
Define the switching surface
5. Observer designs
sq ¼ iq  iq ð9Þ We base the observers on the electrical equations in (1).
They are of the form
and evaluate
P! iq id _^ P u
s_q ¼ s iq þ P!id  d þ RS i ¼ s i^  P!^ i^ þ ð ^  þ !^ ^  Þ þ þ K1 
LS d
LS LS
  ð15aÞ
1 id
þ xuq , where x ¼  : ð10Þ
LS d
_^ P u
Select uq ¼ Mq sign(sq) with i ¼ P!^ i^  s i^  ð!^ ^    ^  Þ þ þ K2 
  LS LS

1 P! iq id  ð15bÞ
Mq 4  s iq þ P!id  d þ RS ,
x LS d

so that sq ¼ 0 and hence iq ¼ iq ; also calculate the _^ ð15cÞ


 ¼ RS i þ u þ K3 
equivalent control
 
1 P! iq id _^ ð15dÞ
s iq  P!id þ  ¼ RS i þ u þ K4 ,

uqeq ¼ d  RS  : ð11Þ
x LS d
where ^ , , i^, are the estimates of the stator fluxes
But, iq can be controlled only if x 6¼ 0 in (10). To and stator currents, respectively, and !, ^  are the
show that this condition is true, we study the dynamics observer inputs. The terms K1, . . . ,4 are additional
of x using: the fact that d is a constant; udeq from (8) design parameters which have to be selected.
because of constant flux control and the result A design step that is common to both the FOSM

  and SOSM observers is the selection of the switching
 RS iq
RS iq  uqeq ¼  s   P!: ð12Þ surfaces. The difference between the two is in the form
LS x
of the inputs, in this case !^ and , which will enforce
With these conditions, we obtain sliding mode on them. We select as switching surfaces,
   the functions
RS K 1
x_ ¼  s  xþ   f ðxÞ ð13Þ
LS x CLS s1 ¼ i~ ^   i~ ^  , s2 ¼ i~ ^  þ i~ ^  , ð16Þ
  2
iq L2R ðs LS  RS Þ
where K ¼ , C ¼ where i~, ¼ i^,  i, and ~ , ¼ ^ ,  , are the

d L2M RR
  errors in the estimation of currents and fluxes
RS respectively. Next, the inputs !^ and  are designed to
and s  4 0:
LS enforce sliding mode on s1 and s2, respectively.
International Journal of Control 1461

To do so, evaluate Following the flux control design process of Section



4.1, the control voltage ud is now designed as an SMC
s_1 ¼ Pð!^  !Þ i^ ^  þ i^ ^  þ P!s2  s s1 so that ^ d ¼ d . Hence, once sd ¼ ^ d  d ¼ 0, we can
Pð!^  !Þ ^ 2

^ 2  P! ^  ~  þ ^  ~ 

calculate the  equivalent control as udeq ¼ RS i^d 
  þ  ðK3 ^  þ K4 ^   . We now choose
LS LS d

P ~ ^

þ ~ ^
     þ i 
~ _^ K3 ¼ ^  , K4 ¼ ^  ð21Þ
LS

so that
_
 i~ ^  þ K2 ^   K1 ^   ð17aÞ

udeq ¼ RS i^d  
d , ð22Þ
s_2 ¼ Pð!^  !Þ i^ ^   i^ ^   P!s1  s s2
which, after using the transformations (19), leads to the
P ~ ^ ~ ^

P!
~ ^ ~ ^

expressions
þ   þ   þ     
LS

LS _ _ s2

_ _ i~ ^   i~ ^  ¼   RS i^q þ uq


þ K1 ^  þ K2 ^   þ i~ ^  þ i~ ^  : ð17bÞ d
Downloaded by [North Carolina State University] at 15:13 24 September 2012

_ _ s1

Observe that both !^ and  appear in the dynamics i~ ^  þ i~ ^  ¼  RS i^q þ uq : ð23Þ
d
of s1 and s2 and to design these inputs based on the
super-twisting algorithm, they will have to be Thus, we finally have the decoupled dynamic systems
decoupled. We select the constants K1,2, for this  2 P d  
purpose, in the next section. s_1 ¼ Pð!^  !Þx^  d þ  ~q  ! ~d
LS
s2 ^

þ  uq  RS iq þ P!s2  s s1 ð24aÞ
d
5.1 Decoupling
To simplify the decoupling process, we concentrate on P d  ~ 
s_2 ¼ Pð!^  !Þ ^
d iq  P!s1 þ ! q þ  ~d
removing the influence  in (17a). Now, by inspecting LS
these dynamics, it can be seen that, if K1,2 are
  2
s1 ^
selected as  s s2 þ  uq  RS iq þ d , ð24bÞ
d LS
1 ^ 1 ^ ^ 
K1 ¼ , K2 ¼ , ð18Þ where x^   ð L1 S  i^d 4 0 by following Section 4.2 –
LS LS d
note that this result is similar to (14). Also, (22) will
then, the last term of (17a) becomes zero. But, as  also hold even when flux is not controlled.
appears in the term We are now ready to design the inputs !^ and  so
_ _   that s1, s2 ¼ 0.
i~ ^   i~ ^  ¼ RS i i~  i i~ þ u i~  u i~
 
þ K3 i~  K4 i~ ,
we select K3,4 so that its influence can be completely 5.2 Observer inputs
removed. To select them, we require a few transfor- If the FOSM observer is to be used, then the inputs are
mations and a result of flux control from Section 4.1. chosen as
We first introduce the transformations !^ ¼ !0 signðs1 Þ and  ¼ 0 signðs2 Þ ð25Þ
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ^ ^ ^ ^
^ d ¼ ^ 2 þ ^ 2 , i^q ¼ i   i  , i^d ¼ i  þ i  with sufficiently high !0, 0 4 0 so that s1,2 ¼ 0 in finite
  ^d ^d time. Proof of the existence of sliding mode and method
ð19Þ of selection of !0 and 0 can be found in Utkin (1992).
If the SOSM observer is to be used for state
and the new variables estimation, then
^ ~  ^ ~ ^ ~ þ ^ ~ pffiffiffiffiffiffiffi
~q ¼ , ~d ¼ : ð20Þ !^ ¼ 1 js1 j signðs1 Þ þ !1
^d ^d
!_ 1 ¼ 1 signðs1 Þ and ð26aÞ
Next, we observe that to control flux its estimate ^ d pffiffiffiffiffiffiffi
will have to be used and whose dynamics is of the form  ¼ 2 js2 j signðs2 Þ þ 1


_^ ^ ^ ^ :
d ¼ RS id þ ud þ K3  þ K4  _ 1 ¼
 if jj 4 M,
ð26bÞ
^d 2 signðs2 Þ if jj  M
1462 S. Rao et al.

so that s1,2 ¼ 0 and s_1,2 ¼ 0 again, in finite time. uqeq – which can be expressed in a form similar to (11),
Conditions for the existence of the constants 1,2, 1,2 but with estimated quantities. We also use the result
and M and the rules for choosing them can be found in (12), again with estimated quantities in place. With
Fridman and Levant (2002). An upper bound cannot these results, we have
be assigned for !^ in (26a) because, formally, the real      
speed is unknown. P iq !iq
eq ¼  ! ~q  þ  ~d
Once sliding mode occurs on the surfaces s1,2, the ^ d x^  d x^  d
currents estimates’ errors i~, ¼ 0. In the following
sections, we show that the estimates of the fluxes and and the closed-loop dynamics of ~ d,q given by the
motor speed also match their real counterparts by linear system
calculating the equivalent controls !^ eq and eq – by " #
_~ ~d
doing so, we also exhibit the proof of convergence of d 1 2
_~ ¼ ~q ð30Þ
the proposed observers. 3 0
q

as the terms
Downloaded by [North Carolina State University] at 15:13 24 September 2012

5.3 Speed estimation     


!iq RS P iq
We propose to use !^ eq , given by  1 ¼ P   þ  ,  2 ¼ s  þ  ,
x^ d LS d x^
  
! ~d  ~q RS iq
!^ eq ¼ !  þ , ð27Þ  3 ¼  s  þ P!
LS x^  ^ d LS x^  ^ d LS x^ 

which is the solution to s_1 ¼ 0, as the speed are constants owing to iq , d also being constants.
estimate. Thus, if we show that the transformed flux Hence, ~ d,q ! 0 if the eigenvalues
estimates’ errors ~ q,d ! 0, then !^ eq is indeed the qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
speed estimate. 1  21 þ 42 3
!^ eq is obtained, if the FOSM observer is imple- 1,2 ¼
2
mented, as the output of a low-pass filter (LPF) whose
input is the discontinuous !, ^ i.e. in the form of the matrix in (30) have negative real parts. This will
be true if 1 4 0 and the product (23) 5 0. Note
1 that these conditions would remain the same even
!^ eq ðsÞ ¼ ^
!ðsÞ: ð28Þ
f s þ 1 if iq , d are time-variant but bounded.
We first consider 1. Since, for any iq , the
There could be a cause for concern, in terms of phase
term x^  4 0; also, d 4 0; and finally, if the motor
lag between the estimated and real speeds, if f is too
operates in the  standard
 motoring condition, then
large. But if it is chosen to be smaller than the
the product !iq 4 0 (Strangas, Khalil, Al-Oliwi,
dynamics of the mechanical component of the motor,
Laubinger, and Miller 1999). Hence, in this case,
while also being larger than the step size used for
1 4 0 and the product (23) 5 0 as well.
discrete implementation, then the phase lag will be
Of course, while the  motor
 is in the braking
nearly zero. If the SOSM observer is used, then we
condition, the product !iq 5 0 and the observers
propose the use of !1 as the speed estimate, and as it is
naturally continuous, it can be directly used. might become unstable. In this situation, we have only
one, however inelegant, recourse – switch off the
estimation process. Once the motor has returned to the
normal motoring mode, the estimation, and hence
5.4 Flux estimation
control, processes can begin. We believe that this
To show that ~ d,q ! 0, we first calculate the equiva- shortcoming is not very severe as during braking,
lent control emphasis will be shifted away from speed and flux
  i^q ! ~ q  ~ d control.
eq
¼ LS !^ eq  !   , ð29Þ We validated these theoretical results experimen-
P ^d ^d ^d
tally; we present experimental and simulation results in
as the solution to s_2 ¼ 0; note that, if the SOSM the next section.
observer is used, then 1 is the naturally continuous
equivalent control.
For further analysis, we let i^q ¼ iq and ^ d ¼ d as 6. Performance
current and flux are controlled. Moreover, with the The IM on which the experiments were conducted was
current being controlled, we use the equivalent control rated at 1.1 kW, 208 V at 50 Hz with a wye connection.
International Journal of Control 1463

We conducted the IEEE recommended tests to using a saturation function or by increasing the power
get: LS ¼ LR ¼ 0.1093H, LM ¼ 0.1H; RS ¼ 2.7 and converter switching frequency.
RR ¼ 0.5 at the time of measurement; also P ¼ 3. We As can be expected, the errors in both control and
conducted the experiments and simulations under the estimation are lower in the simulation results than in
no-load condition, i.e. we set TL ¼ 0; we also assumed a the experimental ones. One reason for errors could be
rotor moment of inertia of J ¼ 0.005. As we could only because of approximating the mechanical parameters
measure the motor shaft position – using a position of the IM, as they were not critical to the development
encoder – the experimental results of speed that are of the observer. Moreover, the step-size that we were
presented are actually a numerically differentiated and forced to adopt for discrete implementation, 0.3 ms,
smoothened version of the position signal. To compute also contributes to a loss of accuracy, especially when
the input voltages uq,d in terms of the power the FOSM observer is implemented.
converter input Vm, we used the formulations given
in Yan (2007).
The experimental parameters used for the FOSM
Downloaded by [North Carolina State University] at 15:13 24 September 2012

and SOSM observers are tabulated in Table 1; these


values are very close to what we also used for the
simulations.
The experimental and simulation results for speed
control are shown in Figures 1 and 2, respectively.
We implemented the control and observer algorithms
for both simulation and experimental studies using the
simple forward-Euler scheme with a step size of 0.3 ms.
We supplied the motor, through the power converter,
with an input voltage of 70 V DC.
In both the figures, starting from the top left and
moving clockwise: the first sub-figure is that of closed-
loop motor speed when the FOSM and SOSM
observers are used for its estimation; the second is
that of closed-loop stator-flux ^ d ; the third presents
the errors in FOSM control sq,d, i.e. deviation from
ideal first-order sliding mode and the fourth shows the
errors in estimation s1,2 when the SOSM observer is
used. Figure 1. Experimental results of speed and flux control.
As can be seen in Figure 1, the closed-loop
speed error and deviation of ^ d from d are greater
when they are estimated using the FOSM observer; this
also acts as evidence that the super-twisting algorithm
offers greater accuracy – possible only after carefully
tuning the parameters (, )1,2. As ^ d is controlled
using an FOSM controller, there is the presence of
chatter in the output which can be attenuated, say by

Table 1. Observers’ parameters.

FOSM observer SOSM observer

!0 ¼ 1.2!ref 1 ¼ 50
1 ¼ 100 in (26a)
0 ¼ 0.5 2 ¼ 0.01
2 ¼ 30 in (26b)
f ¼ 0.15 in (28) M ¼ 0.5 in (26b)
!ref  28 rad/s; d ¼ 0:03 Wb
 ¼ 5 in (6)
Figure 2. Simulation results of speed and flux control.
1464 S. Rao et al.

7. Conclusions Kenne, G., Ahmed-Ali, T., Lamnabhi-Lagarrigue, F., and


Arzande, A. (2009), ‘Real-time Speed and Flux Adaptive
As a concluding comment, we try to answer a
Control of Induction Motors using Unknown
difficult but immediate question: which is the better Time-varying Rotor Resistance and Load Torque’, IEEE
observer? On the whole, the SOSM observer yields Transactions on Energy Conversion, 24, 375–387.
better results – in terms of accuracy – than the FOSM Kheloui, A., Aliouane, K., Medjaoui, M., and Davat, B.
observer, as it should. But, this increased accuracy can (2000), ‘Design of a Stator Flux Sliding Mode Observer for
be had only by careful tuning of the SOSM observer Direct Torque Control of Sensorless Induction Machine’,
parameters. So, the answer to the question is depen- in Industry Applications Conference, 2000. Conference
dent more on what the user is familiar with, rather than Record of the 2000 IEEE, Vol. 3, pp. 1388–1393.
the actual performance measures of each observer, Krause, P. (1986), Analysis of Electric Machinery, New York:
which can always be adjusted to yield similar results. McGraw-Hill.
Leonhard, W. (2001), Control of Electric Drives, Berlin:
Springer-Verlag.
Marchesoni, M., Segarich, P., and Soressi, E. (1997),
Acknowledgements ‘A Simple Approach to Flux and Speed Observation in
Downloaded by [North Carolina State University] at 15:13 24 September 2012

This research was supported in part by the Deutsche Induction Motor Drives’, IEEE Transactions on Industrial
Forschungsgemeinschaft (German Research Foundation), Electronics, 44, 528–535.
for which the authors are grateful. Our thanks also go to Proca, A., and Keyhani, A. (2007), ‘Sliding-mode Flux
Mr Michael Fall who helped in conducting the IM tests to Observer with Online Rotor Parameter Estimation for
determine its electrical parameters. Induction Motors’, IEEE Transactions on Industrial
Electronics, 54, 716–723.
Rao, S., Buss, M., and Utkin, V. (2009), ‘Simultaneous State
References and Parameter Estimation in Induction Motors using First
and Second Order Sliding Modes’, IEEE Transactions on
Barut, M., Bogosyan, S., and Gokasan, M. (2007), ‘Speed- Industrial Electronics, 56, 3369–3376.
sensorless Estimation for Induction Motors using Shin, M.H., Hyun, D.S., Cho, S.B., and Choe, S.Y. (2000),
Extended Kalman Filters’, IEEE Transactions on ‘An Improved Stator Flux Estimation for Speed
Industrial Electronics, 54, 272–280. Sensorless Stator Flux Orientation Control of Induction
Boussak, M., and Jarray, K. (2006), ‘A High-performance Motors’, IEEE Transactions on Power Electronics, 15,
Sensorless Indirect Stator Flux Orientation Control of 312–318.
Induction Motor Drive’, IEEE Transactions on Industrial Strangas, E.G., Khalil, H.K., Al-Oliwi, B., Laubinger, L.,
Electronics, 53, 41–49. and Miller, J.M. (1999), ‘A Robust Torque Controller for
Comanescu, M., and Xu, L. (2006), ‘Sliding-mode MRAS Induction Motors Without Rotor Position Sensor:
Speed Estimators for Sensorless Vector Control of Analysis and Experimental Results’, IEEE Transactions
Induction Machine’, IEEE Transactions on Industrial on Energy Conversion, 14, 1448–1458.
Electronics, 53, 146–153. Utkin, V.I. (1992), Sliding Modes in Control and
Fridman, L., and Levant, A. (2002), Higher Order Sliding Optimization, Berlin: Springer-Verlag.
Modes in: Sliding Mode Control in Engineering, New York: Utkin, V.I., and Lee, H. (2006), ‘The Chattering Analysis’,
Marcel Dekker. in EPE-PEMC 2006 12th International Power Electronics
Holtz, J., and Quan, J. (2002), ‘Sensorless Vector Control and Motion Control Conference, pp. 2014–2019.
of Induction Motors at Very Low Speed using a Yan, W. (2007 ), ‘Multilevel Sliding Mode Control in Hybrid
Nonlinear Inverter Model and Parameter Power Systems’, thesis, Ohio State University.
Identification’, IEEE Transactions on Industrial Yan, Z., Jin, C., and Utkin, V. (2000), ‘Sensorless
Applications, 38, 1087–1095. Sliding-mode Control of Induction Motors’, IEEE
Kenne, G., Ahmed-Ali, T., Lamnabhi-Lagarrigue, F., Transactions on Industrial Electronics, 47, 1286–1297.
and Arzande, A. (2008), ‘Nonlinear Systems Time- Yan, Z., and Utkin, V. (2002), ‘Sliding Mode Observers for
varying Parameter Estimation: Application to Electric Machines – An Overview’, in IECON 02, 28th
Induction Motors’, Electric Power Systems Research, Annual Conference of the Industrial Electronics Society,
78, 1881–1888. Vol. 3, pp. 1842–1847.

Das könnte Ihnen auch gefallen