Proceedings of the 20th World Congress

The International
Proceedings of Federation of Congress
Automatic Control
Proceedings
The of the
the
International 20th
20th World
World
Federation of Congress
Automatic Control
Toulouse,
The France,
Proceedings of the
International July
20th9-14,
World
Federation 2017
of Congress
The International
Toulouse, France,Federation of Automatic
Automatic
July 9-14, 2017 Control
Available online at www.sciencedirect.com
Control
The International
Toulouse,
Toulouse, France, Federation
France, July
July 9-14, of Automatic Control
9-14, 2017
2017
Toulouse, France, July 9-14, 2017
ScienceDirect
IFAC PapersOnLine
Robustness of 50-1 (2017) position
rotor 15403–15408 observer for
Robustness
Robustness of
of rotor
rotor position
position observer
observer for
for
permanent
Robustness
permanent of magnet synchronous
rotor synchronous
magnet position observermotors
for
motors
permanent
with
permanent
with magnet
unknown
magnet synchronous
magnet
synchronous flux motors
motors
with unknown
unknown magnet magnet flux
flux
with unknown ∗ magnet flux
Pauline Bernard ∗ Laurent Praly ∗∗
Pauline Bernard ∗ Laurent Praly ∗
Pauline
Pauline Bernard Bernard ∗∗ LaurentLaurent Praly Praly ∗∗
∗
Control and Pauline
Systems Bernard
Center, Laurent Praly
∗
∗ ControlUniversity.
and Systems Center, MINES ParisTech, PSL
MINES ParisTech,
(pauline.bernard@mines-paristech.fr, PSL Research
Research
∗
∗
ControlUniversity.
Control and Systems
and Systems Center,
Center, MINES
MINES ParisTech,
ParisTech,
(pauline.bernard@mines-paristech.fr, PSL Research
PSL Research
ControlUniversity.
and Systems
University. Center, MINES ParisTech, PSL Research
laurent.praly@mines-paristech.fr)
(pauline.bernard@mines-paristech.fr,
(pauline.bernard@mines-paristech.fr,
University. laurent.praly@mines-paristech.fr)
(pauline.bernard@mines-paristech.fr,
laurent.praly@mines-paristech.fr)
laurent.praly@mines-paristech.fr)
laurent.praly@mines-paristech.fr)
Abstract:
Abstract: We
We introduce
introduce a
a newnew sensorless
sensorless rotor
rotor position
position observer
observer for
for flux permanent
permanent magnet
magnet
synchronous
Abstract:
Abstract:
synchronous motors
We
We
motors which
introduce
introduce
which does
a
a
does new
new not
not require
sensorless
sensorless
require the
the knowledge
rotor
rotor position
position
knowledge of
of the magnet’s
observer
observer
the magnet’s for
for permanent
permanent
flux :
: only
only electrical
magnet
magnet
electrical
measurements
Abstract:
synchronous
synchronous
measurements We and
motors
motors
and (approximate)
introduce
which
which does
(approximate) a new
does not
notknowledge
sensorless
require
require the
knowledge of
the the
the resistance
ofrotor position
knowledge
knowledge
resistance of and
observer
of the inductance
andmagnet’s
the magnet’s for flux
inductance flux are
permanent needed.
:: only
are only
needed. This
magnet
electrical
electrical
This
synchronous
observer
measurements
measurements extendsmotors
and
andthe which
gradient
(approximate)
(approximate) does not
observer require
from
knowledge
knowledge the
Lee
of
of knowledge
et
the
theal. [2010]
resistance
resistance ofwith
the
and
and magnet’s
the estimation
inductance
inductance fluxare :ofonly
are the
needed.
needed. electrical
magnet’s
This
This
observer
measurements
flux and extends
makes and
itthe gradient
(approximate)
globally observer
convergent from
knowledge
provided Lee et
of the al.
theal. [2010]
resistance
rotation with the
andthe
speed estimation
inductance
remains away areof the
needed. magnet’s
This
observer
observer
flux and extends
extends
makes the
itthe gradient
gradient
globally observer
observer
convergent from Lee
from
provided Lee et
et
the al. [2010] with
[2010]
rotation with
speed the estimation
estimation
remains away offrom
of the
the
from zero.
zero. We
magnet’s
magnet’s We
study
observer
flux
flux
study and
and its
its sensitivity
extends
makes
makes it
it
sensitivity the to
globally
globally
to uncertainties
gradient observer
convergent
convergent
uncertainties on
from the
provided
provided
on the resistance
Lee et
the
the al. and
[2010]
rotation
rotation
resistance and inductance
with
speed
speed the
inductance remains
remains and
estimation
and to
away
away
to the
of
the the
from
from presence
magnet’s
zero.
zero.
presence We
Weof
of
saliency.
flux
study
study
saliency.and its Its
makes
its Its performances
it
sensitivity
sensitivity globally
performances to in open-loop
convergent
uncertainties
to uncertainties
in open-loop on are
provided
on arethe illustrated
the
resistance
the illustrated
resistance via
rotation
and
and an implementation
speed
inductance
inductance
viaobservers
an remains
implementation and away
to
and toofthe using
the from
using real
zero.
presence
presence data
real dataWeof
of
and
study
saliency.
saliency.
and compared
its
compared Its to
sensitivity other
Its performances
performances to existing
uncertainties
to other existing in magnet
in open-loop on
magnet are
open-loop flux
the independent
resistance
fluxillustrated
are illustrated
independent and
via inductance
an in
an implementation
viaobservers terms
and to
in terms ofusing
implementation thecomputational
using presence
computationalreal
real data
data of
cost
saliency.
and
and and
and robustness.
cost compared
compared Its performances
to
to other
robustness. other existing in open-loop
existing magnet are
magnet fluxillustrated
flux independent
independent viaobservers
an implementation
observers in
in terms
terms of ofusing
computational
computationalreal data
and
cost compared
cost and
and robustness. to other existing magnet flux independent observers in terms of computational
robustness.
© 2017,
cost and1. IFAC (International Federation of Automatic Control)
robustness.
INTRODUCTION the Hostingobserver
by Elsevier Ltd. All rights reserved.
1. INTRODUCTION the gradient
from gradient
the observer previously
resistance. previously mentioned mentioned independent independent
1. INTRODUCTION
1. INTRODUCTION the
the gradient
fromgradient observer previously mentioned
observer previously mentioned independent
the resistance. independent
1. INTRODUCTION the
from
from gradient
thepaper,
the observer previously mentioned independent
resistance.
resistance.
1.1 In this we
we focus
focus on on observers which
which require the
1.1 Context
Context from
In thisthepaper,
knowledge
resistance.
of the observers require the
1.1 Context
1.1 Context In
In this paper,
this
knowledge paper,
of the weresistance
we focus on
focus
resistance onand
and the
observers
observers
the inductance,
which require
which
inductance, but
require
but not
not of
the
the
of
1.1 Context the
In magnet
this
knowledge
knowledge
the magnetpaper,flux.
of
of the
the
flux. we First
focus steps
resistance
resistance
First stepson and
andin
in this
observers
the
the
this direction
which
inductance,
inductance,
direction are
are reported
require
but
but not
notthe
reported of
of
To
To minimize the
the cost
cost and and increase the
the reliability
reliability of of Per- in
manentminimize Magnet increase Per- in Henwood
knowledge
the
the magnetofflux.
magnet
Henwood et
theal.
flux.
et [2012]
al.resistance
First
First
[2012] steps
stepswith
withandin the
in thedesign
this
this
the inductance,
direction
direction
design of
of aaareLuenberger
are but not of
reported
reported
Luenberger
To
To minimize
minimize
manent Magnet the Synchronous
the cost and
cost
Synchronousand increase Motors
increase
Motors the(PMSM),
the reliabilityit
reliability
(PMSM), it ofis
of
is still
Per-
Per-
still the
in
in magnet
observer
Henwood
Henwood
observer (seeflux.
et
et al.
(seeetHenwood
First
Henwood
al. [2012]
[2012] steps
[2014]
with
with
[2014]
in thethis
for
the a direction
design
design
for design much of
of
a muchof more a
aare
more reported
detailed
Luenberger
Luenberger
detailed
important
To
manentminimize to
to make
Magnet progress
the Synchronous
cost
Synchronous on
and increase estimating
Motors the(PMSM),
(PMSM), their
reliability state is vari-
it ofis Per- in Henwood
still analysis), and al.
in [2012]
Bobtsov with et the
al. a 2016],
Luenberger
manent
important
ables,
manent in Magnet make
particular
Magnet
progress
the
Synchronous rotoronMotors
estimating
position
Motors and
(PMSM),
their it
state
speed,
it is
still
vari-
with
still
observer
observer
analysis), (see
(see
and Henwood
Henwood
in Bobtsov [2014]
[2014] et for
for
al. aa[2015a,b,
much more
much
[2015a,b, more with
detailed
detailed
2016], with
important
important
ables, in to
to make
make
particular progress
progress
the rotoron
on estimating
estimating
position andtheir
their state
state
speed, vari-
vari-
with observer
the design
analysis), (see
of
and Henwood
an in observer
Bobtsov [2014]
basedet foron
al. a much
tools
[2015a,b, more
from detailed
parameter
2016], with
a minimum
important
ables, to of
in particular sensors
make
particular and
progress
theand rotorfast
on algorithms.
estimating
position andtheirTo this
state
speed, end,
vari-
with analysis),
the design and
of an in Bobtsov
observer basedet al.
on [2015a,b,
tools from 2016], with
parameter
ables,
a minimum
studies in of sensors the rotorfast position
algorithms. and Tospeed,
this end, linear
with analysis),
the
the
linear identification.
design
design and aninobserver
of an
of
identification. In
Bobtsov
observer
Insamefact,
fact,based
basedetwe
weal. will
on
will show
on [2015a,b,
tools
tools
show that
from
from
that those
2016],
parameter
parameter
those two
with
two
ables,
astudies
a minimum
minimum inhave
have of
of been
particularsensors
sensors
been made
madetheand
andfor
rotor
for aa long
fast
fast long time
position
algorithms.
algorithms.
time on
and
on the
To
To
the so-called
speed,
this
this with
end,
end,
so-called observers
the
lineardesign rely
of an on
identification. the
observer
In fact,based regression
we on
will tools
show equation
from
that but
parameter
those the
two
”sensorless”
a minimum
studies control
of
have control sensors
been made madewhich
and uses
fast
forusesa long
long no mechanical
algorithms.
time on the To variables
this
the variables end,
so-called former linear
observers identification.
rely
solves iton at theIn
each fact,
same
time we will
regression
whereas show
the that
equation
latter those
but
solvestwo
the
it
studies
”sensorless” have been which for a no time on
mechanical so-called linear
observers
observers
former identification.
rely
relyiton
solves on at the
theIn fact,
same
eachsame we will
regression
timeregression show
whereas Convergence that
equation
equation
the those
but
but
latter solves two
the
the
it
measurement,
studies
”sensorless”
”sensorless”
measurement, have only
been
control
control
only electrical
madewhich
which for
electrical a
uses
usesones.
long
ones. no
no A
time
A review
on
mechanical
mechanical
review theof
of the
so-called
variables
variables
the first
first with time
observers with
rely a
on gradient-like
the same scheme.
regression equation comes
but the
used methods
”sensorless” was
control given
which in Acarnley
uses no and
mechanicalWatson [2006]),
variables former
former
with solves
solves
time with it
ita at
at each
each
gradient-liketime
time whereas
whereas
scheme. the
the latter
latter
Convergence solves
solves
comes it
it
measurement, only electrical ones. A review of the first with an assumption of
measurement,
used amethods
then
measurement,
used Luenberger
methods
onlyobserver
was electrical
given
onlyobserver
was electrical
given was ones.
in Acarnley
ones.
in Acarnley
Acarnley
Aand
proposed
Aand
and
review
review
Watson
in of the
Poulain
Watson of the etfirst
[2006]), al.
first
[2006]),
former
with
with
for the
solves
time
time withitaa at
with
anformer,
assumption of ainvertibility
each
gradient-like
gradient-liketime scheme.
invertibility whereas
scheme. of
of the theregressor
Convergence
Convergence
the regressor matrix
latter solves
comes
comes
matrixit
used
then amethods
Luenberger was given in was proposed Watson
in Poulain [2006]),
et al. with
with
for thetime
an
an with and
assumption
assumption
former, and on
a gradient-like
of
of
on a persistent
invertibility
invertibility
persistent scheme. excitation
of
of the
the
excitation condition
Convergence
regressor
regressor
conditioncomesfor
matrix
matrix
for
[2008].
used
then
then
[2008]. More
More recently,
aamethods
Luenberger
Luenberger recently,givenaa invery
was observer
observer was simple
Acarnley
was
very proposed
proposed
simple and gradient
Watson
in
in
gradient Poulain
Poulain observer,
[2006]),
et al.
et
observer, al. the with latter.
anformer,
assumption of aainvertibility of the regressor matrix
then a Luenberger observer was proposed in Poulain al. for for the
the the former,
latter. and on
and on persistent excitation
persistent excitation condition
condition for
for
proposed
[2008].
[2008].
proposed in
More
More
in Lee
Lee et
recently,
recently,
et al.
al. [2010]
a
a
[2010] very
veryand
and analyzed
simple
simple
analyzed in
gradient
gradient
in Ortega
Ortega et
observer,
observer,
et al. for
the the former,
latter. and on a persistent excitation condition for
[2008]. More recently, a be very the
Here, latter.
we start by proposing, for the same goal, another
[2011],
proposed
proposed
[2011], has
has inbeen
in Lee
Lee
been shown
et al.
et
shown to
al. [2010]
[2010]
to be andsimple
extremely
and
extremely analyzed
analyzed
gradient
effective
in Ortega
in
effective in
Ortega
in
observer,
practice
et
practice al. the
et al. Here, latter.
we start by proposing, for the same goal, another
proposed
as inbeen
Lee shown
et estimator.
al. [2010] and analyzed in Ortega et al. observer
Here, we which
start by byis
is aaproposing,
direct
proposing, extension,
for the the same with
same estimation
goal, anotherof
as rotor
[2011],
[2011],
point rotorit
hasposition
hasis beenconditionally
position
only shown to be
to
estimator. From
From the
be extremely
extremely
convergent the theoretical
effective
effective
theoretical
but it
in practice
in
was
view
view Here,
practice
shown
observer
the
Here,
we
magnet
observer
start
which
we whichflux,
start
which byisof
direct
the
directgradient
aproposing,
for
extension,
forobserver
extension, the same
with
with
goal,
obtained
goal,
another
estimationin of
Lee
another
estimation of
[2011],
as
as
pointrotor
rotorithasis been
position
position
only shown to
estimator.
estimator.
conditionally be extremely
From
From
convergent the
theeffective
theoretical
theoretical
but it in
was practice
view
view
shown observer
the magnet flux, isofa direct
the gradientextension, observer with estimation
obtained in of
Lee
in et
theal.
observer [2010].
magnet which We
flux, claim
Weisclaim
ofathe
the its
direct global
gradientextension, convergence.
observer with estimation
obtained in Leeof
Lee
in Malaizé
as
point
point
it
rotor
can
it is
it
Malaizé be
et
et al.
isposition
only
only
made
[2012]
conditionally
al.conditionally
[2012]
globally
how,
estimator.
how, via aa very
From verythe
convergent
convergent
via
convergent
minor
minor
thanks
but
but modification,
theoretical
it
it
to
was
was view the
shown
shown
modification,
convexity
et al.magnet
the
[2010].
al.magnet
et al. [2010].
flux,
flux,
of
of theits
Wevarious
claim
gradient
its global
gradient
its global observer
observer
convergence.
convergence.
obtained in
obtained in Lee
point
in
in it
Malaizé
it Malaizé
can be madeis only
et al.conditionally
[2012]
et al. [2012]
globally how,
how, convergent
via a very
via a verythanks
convergent minorbut it was shown
modification,
minor modification,
to convexity et et
Then, [2010].
for theseWe claim global
observers, convergence.
we study the sensitivity
arguments.
in Malaizé et al. [2012] how, via a verythanks minor modification, al. [2010].
Then, for theseWevarious
claim its global convergence.
observers, we study theand sensitivity
it can be be made
made globally convergent thanks to convexity
convexity of the estimates to errors in
it
it
can
arguments.
can
arguments. be made
globally
globally
convergent
convergent thanks
to
to convexity
Then,
Then,
of the
tance,
for
for these various
these
estimates
and also
various
toto errors
the action in the
observers,
observers, the
of
weresistance
we study the
study
resistance
(ignored)
theand
saliency
induc-
sensitivity
sensitivity
induc-
and
arguments.
These observers require typically the knowledge of the Then,
of
of the
tance, for
the and these
estimates
estimates
also various
toto the
to observers,
errors
errorsaction in the
in the
of we study
resistance
(ignored) the
resistancesaliency sensitivity
and induc-
and induc-
and
arguments.
These observers require typically the knowledge of the we
of illustrate
the
tance, and
estimates
and also compare
toto errors
the their
action in performances
the
of resistance
(ignored) in open-loop
and
saliency induc-
and
resistance,
These
These
resistance, magnet
observers
observers
magnet flux
require
require
flux and
typically
typicallyinductance.
and inductance. the
the Unfortunately
knowledge
knowledge
Unfortunatelyof
of the
the tance,
we and
illustrate also
and to the
compare action
their of (ignored)
performances saliency
in open-loopand
while
These
resistance,
resistance,
whilelong the
the latter
observers
magnet
magnet
latter may
may be
require
flux
flux
be considered
typically
and
and
considered as
theknown
inductance.
inductance.
as knownknowledge and
Unfortunately
Unfortunately
andthe of the through
constant
constant
tance,
we
we simulations
and
illustrate
illustrate
through also
and to
and
simulations on
on real
the
compare
compare action
realtheir
theirdata.
data. of (ignored) saliency
performances
performances in open-loop
in open-loopand
(as
resistance, as there
magnet is no
flux magnetic
and saturation),
inductance. Unfortunately other we illustrate
through
through and compare
simulations
simulations on
on real
realtheirdata.
data. performances in open-loop
while
while the as
the latter
latter may
may be
is be considered
noconsidered as
as known and
known andthe constant
constant 1.2
(as long
two
while
(as
(as
two do
long
long
do the vary
as
as
vary
there
significantly
latter
there
there may is be
is
significantly
magnetic
with
noconsidered
no magnetic
magnetic
with the
the
saturation),
temperature
as known andand
saturation),
saturation),
temperature
other
these
constant
the
the
and other
other
these 1.2 System
through
System model
model and
simulations andonproblem
real data.
problem statement
statement
variations
(as
two long
do should
as
vary there be
significantlyistaken
no into
magnetic
with account
the in the
saturation),
temperature observer.
the
and For
other
these 1.2
1.2 System
System model
model and
and problem
problem statement
statement
two do vary
variations significantly
should be taken with the temperature
into account in the the and
observer. these
For Using Joule’s and Faraday’s laws, aa simpler
example,
two do vary
variations
variations
example, for
should
should
for aa given
significantly
be taken
be
given injected
taken
injected intocurrent,
with
into account
account
current, when
the temperature
in the
in
when the the magnet’s
and
observer.
observer. these
magnet’s
1.2
Using
For expressed
For
System
Joule’s
in
model
aand
fixed
and problem
Faraday’s
αβ-frame laws, statement
reads simpler PMSM PMSM model model
temperature
variations
example, should
for aincreases,
be
given taken its
injected magnetic
into account
current, flux
in
when the decreases,
observer.
the magnet’s and
For Using
Using
expressed Joule’s
Joule’s
in aand
and
fixed Faraday’s
Faraday’s
αβ-frame laws,
laws,
reads aa simpler
simpler PMSM
PMSM model
model
example,
temperature for aincreases,
given injected current, flux
its magnetic whendecreases,
the an magnet’s and Using expressed Joule’s
in aaandfixed Faraday’s
αβ-frame laws,
reads a simpler PMSM model
u−
the
the produced torque
given becomes smaller.
current,Therefore, online Ψ̇
example,
temperature
temperature
produced
estimation
for aincreases,
of
increases,
torque
the
injected
magnet’s
its magnetic
its
becomes magnetic
fluxsmaller.
enables
whendecreases,
flux
flux
Therefore,
to :
the an
decreases,magnet’s and expressed
and
online expressed
in
in a
fixed
fixed Ψ̇ =
αβ-frame
αβ-frame =u R
R ii
−reads
reads
(1)
(1)
temperature
the
the produced
produced
estimation increases,
torque
of torque
the magnet’s its
becomes
becomes magnetic
fluxsmaller.
smaller.
enables flux
Therefore,
Therefore,
to : decreases,an and
online where Ψ is
an online where Ψ is the total flux the total flux
Ψ̇ =generated
u
Ψ̇ =generated − R i by the
u − R i by the windings and windings and
(1)
(1)
the produced torque becomes smaller. Therefore, an online the permanent the magnet, Ψ̇(u, u − Rthe
=generated
i) i by (1)
estimation
estimation
•• adapt
estimation
adapt
of
of
the
of
the
the
the
the
magnet’s
magnet’s
control
magnet’s
control law
law
flux
flux
in
flux
in
enables
enables
real time
enables
real time
toand
to
toand
:: thus
: thus ensure ensure
where
where
thethe
of
Ψ is
Ψ is the
permanent
current in
total flux
total
magnet,
the
flux
fixed i) are
generated
(u,frame are andthe by voltage
R the
voltage
the
and
and intensity
the windings
windings
stator
and
and
intensity
winding
where
the Ψ is
permanent the total
magnet, flux (u, generated
i) are the by the
voltage windings
and and
intensity
•• aadapt
a torque
adapt torque thecontrol
the control which
control
control law in
law
which inis robust
real
isreal timeto
time
robust and
toand the
thethusmachine’s
thus ensure
machine’s thethe
of
ensure resistance.
thethe
of the
permanent
current
permanent
currentThe inmagnet,
the fixed
quantities
inmagnet,
the fixed
fixed
(u,frame
i) iare
u,
(u,frame
i) iare
frame and the
and
the
and Ψ voltage
Rarethetwo
voltage
R
and
the stator
and
intensity
winding
dimensional
stator intensity
winding
• temperature
adapt
atemperature
a torque
torque thecontrol ;;
control
control law inis
which
which isreal timeto
robust
robust toand thethus
the ensure of
machine’s
machine’s current
resistance. The in the
quantities u, andand Ψ Rarethe
two stator winding
dimensional
•• have
ahave an
torque
temperature
temperaturean estimation
control
;;
estimation of
ofofthe
which the isrotor’s to the machine’s vectors,
robusttemperature
rotor’s temperature
of
resistance.
resistance.
vectors, and,
the current
and, The
Thefor
for the
the case
the fixedof
inquantities
quantities
case u,aa iinon-salient
offrame
u, and
andandΨΨRare
non-salient are PMSM,
the
two
two stator
PMSM, the total
winding
dimensional
dimensional
the total
• have
temperature an estimation
; the magnet’s magnetization flux may
resistance.
vectors,
vectors,
flux may be
and,
and,
be expressed
Theforquantities
for the case
the
expressed case as
as ofu,aa inon-salient
of and Ψ are PMSM,
non-salient two dimensional
PMSM, the total
the total
• have
• have an
an
an estimation
estimation
estimation of
ofofthe
the the rotor’s
rotor’s
magnet’stemperature
temperaturemagnetization vectors, and, for the case of a 
non-salient
 cos θ   PMSM, the total
flux may
may be be expressed
expressed as as
•• degradation
have
have an
degradationan estimation
an with
with time.
estimation
estimation ofof
time.ofthe therotor’s
the magnet’s
magnet’stemperaturemagnetization flux
magnetization flux may be expressed Ψ
Ψ= =as Li
Li + +Φ Φ  cos
sin θθθθ 
 (2)
(2)
• degradation
have an estimation
degradation with
with time.
time.of the magnet’s magnetization  cos
cos
sin 
That
That is why
degradation
is why efforts
efforts with have
have been
time.
been made
made to
to look
look for
for observers
observers Ψ
Ψ =
= Li
Li +
+ Φ
Φ cos
sin θ (2)
(2)
which where
where L L is the
the inductance,
Ψ = Li +Φ ΦΦthe magnet’s
sin θ flux,
flux, and
and θθ the
(2)
That
That
which isdo
is why
dowhy not rely
efforts
notefforts
rely on the
have
onhave
theet knowledge
been
been
knowledge made to
made of
to
of those
look for
look
those parameters.
for observers electrical
observers
parameters. isphase. inductance,
This relation the magnet’s
sin
implies θ the
For
That
which instance,
is why
do not in Romero
efforts
not in rely onhave beenal. [2016],
made the
to authors
look for propose
observers where
where
electricalL
L is
is the
the
phase. inductance,
inductance,
This relation Φ
Φ the
the
impliesmagnet’s
magnet’s flux,
flux, and
and θ
θ the
the
which do
For instance, rely
Romeroon the
theetknowledge
knowledge
al. [2016], of
thethose
of those
authors parameters.
parameters.
propose where
electricalL isphase.
the inductance,
phase. This|Ψ relation
− Φ2 the 2magnet’s flux, and θ the
implies
and
which
For
For studydo not
instance,
and instance,
study via
viaininsimulations
rely
Romero
Romeroon theet
simulations an
etknowledgeadaptive
al. [2016],
al.
an [2016],
adaptive of
the observer
thethoseauthors
authors
observer to
parameters.make
make electrical
propose
propose
to electrical phase.
This
This − Li|Li|22 −
|Ψ relation
relation
Φ
Φ22 =
−implies
implies = 00 (3)
(3)
For
and instance,
study via in Romero
simulations et al.
an
and study via simulations an adaptive observer to make [2016],
adaptive the authors
observer propose
to make |Ψ − Li| 2 − Φ2 = 0
|Ψ − Li| − Φ = 0 (3)
(3)
and study via simulations an adaptive observer to make |Ψ − Li|2 − Φ2 = 0 (3)
Copyright © 2017 IFAC 15973
Copyright
2405-8963 ©© 2017
2017, IFAC 15973Hosting by Elsevier Ltd. All rights reserved.
IFAC (International Federation of Automatic Control)
Copyright
Copyright
Peer review© 2017
2017 IFAC
© under IFAC 15973
15973Control.
responsibility of International Federation of Automatic
Copyright © 2017 IFAC
10.1016/j.ifacol.2017.08.1866 15973
Proceedings of the 20th IFAC World Congress
15404
Toulouse, France, July 9-14, 2017 Pauline Bernard et al. / IFAC PapersOnLine 50-1 (2017) 15403–15408

and the electrical phase θ is nothing but the argument of studying the invariant sets. Unfortunately it is too long
Ψ − Li. Therefore, in the case where L and i are known, θ and too technical to be given here.
can be recovered simply through an estimate of the total
flux Ψ. This theorem tells us that unlike for observer (4), no
convexification is needed to achieve global convergence of
Our interest in this work is about observers of Ψ using the gradient observer (6). Hence, even when the parameter
measurements of u and i, (approximate) knowledge of Φ is known, we may prefer to use observer (6) instead
R and L but not of Φ. In particular, we look at the of observer (4). In this way, although the observer state
computational cost and study how the estimate they
give depend on uncertainties on R, L and saliency. To is augmented with Φ̂, we get global convergence and
guarantee observability, we assume, all along, the electric independence with respect to Φ.
rotation speed ω = θ̇ remains away from 0. 3. ALTERNATIVE PATH
Notations : The rotation matrix of angle θ is denoted R(θ),
i-e   The observer presented in the previous section is based on
cos θ − sin θ the system 
R(θ) = sin θ cos θ .
 Ψ̇ = u − Ri
Φ̇ = 0 (7)
2. GRADIENT OBSERVER  2 2
y = |Ψ − Li| − Φ
In Lee et al. [2010], the authors proposed the gradient with inputs (u, i), state (Ψ, Φ) and measurement y which
observer : is constantly zero. This system is nonlinear because of its
 2  output function. Fortunately, this function is quadratic
˙  
Ψ̂ = u − R̂i − 2q (Ψ̂ − L̂i) Ψ̂ − L̂i − Φ2 . (4) in (Ψ, Φ), and (Ψ̇, Φ̇) does not depend on (Ψ, Φ). Hence
linearity can be obtained by time derivation. Namely, we
where R̂ and L̂ are estimates of R and L the model have
T
˙
parameters. This observer turned out to be quite efficient ẏ = 2 (Ψ − Li) (u − Ri − Li)
in practice but it was proved in Ortega et al. [2011] which is linear in Ψ and independent from Φ. The new
that it was only conditionally convergent. In particular problem we face now is the presence of the time derivative
it may admit several equilibrium points depending on ˙
the rotation speed ω. In fact, later in Malaizé et al. Li. A well known fix to this, is to use a strictly causal
[2012], it was shown that taking rather the following filter. Namely, let
gradient
 observer, basedon the “convexified” expression η̇ = −λ(η + y) , yf = η + y (8)
2
 
max Ψ̂ − L̂i − Φ2 , 0 , with λ any complex number with strictly positive real part.
  It is easy to check that the evaluation of yf +(c+2Li)T Ψ−
2
˙   (z + L2 |i|2 ), along any solution, decreases as exp(−λt)
Ψ̂ = u − R̂i − 2q (Ψ̂ − L̂i) max Ψ̂ − L̂i − Φ2 , 0 (5) when c andz are solutions of
ċ = −λc − 2λLi − 2(u − Ri)
enables to achieve global asymptotic stability when R̂ = R (9)
ż = −λz + cT (u − Ri) − λL2 |i|2 .
and L̂ = L.
So, instead of the design model (7), we can use :
In this paper, we propose the following observer for Ψ and 
Φ  Ψ̇ = u − Ri
 2  (10)
 ˙   yf = −(c + 2Li)T Ψ + (z + L2 |i|2 )

 Ψ̂ = u − R̂i − 2γ (Ψ̂ − L̂i) Ψ̂ − L̂i − Φ̂ 2
with inputs (u, i, c, z), state Ψ and measurement yf . Also
 2  (6)
 ˙   because of (8), we pick yf constantly zero as we did above

 Φ̂ = γ Φ̂ Ψ̂ − L̂i − Φ̂ 2
with y. The system (10) can be seen as a linear time
varying system and therefore any observer design for such
where γ is an arbitrary strictly positive real number. It is systems apply. It can be a Kalman filter or more simply
a straightforward extension of the gradient observer (4). the following gradient observer :

Let Ψ(ψ, t) be the solution at time t of model (1)  ċ = −λc − 2λL̂i − 2(u − R̂i)

satisfying (3) initialized at ψ at time 0. Similarly, let ż = −λz + cT (u− R̂i) − λL̂2 |i|2 
(Ψ̂(ψ̂, φ̂, t), Φ̂(ψ̂, φ̂, t)) be the solution at time t of observer 
Ψ̂ ˙
= u − R̂i + γ c + 2L̂i −(c + 2L̂i)T Ψ̂ + z + L̂2 |i|2 .
(6) initialized at (ψ̂, φ̂) at time 0.
(11)
Theorem 1. Assume that there exists a strictly positive where γ is an arbitrary strictly positive real number. In
number w such that ω is lower-bounded by w, and that Bobtsov et al. [2015a], the authors propose the following
the state variables of the PMSM are bounded. Assume non minimal version of this observer :
also R̂ = R and L̂ = L. Then, for any strictly positive real 

 Ψ̂ = ξ14 + ξ89
number γ, for any (ψ, ψ̂, φ) in R4 × (0, +∞), we have 
 ξ˙14 = u − R̂i


lim |Ψ̂(ψ̂, φ̂, t) − Ψ(ψ, t)| + |Φ̂(ψ, φ, t) − Φ| = 0 .  ˙ 2
ξ5 = −λ(ξ  5 − |ξT14 −L̂i| )
t→∞ (12)

 ξ˙89 = γΩ y − Ω ξ89


Taking θ̂ as the argument of Ψ̂ − L̂i, we also obtain 
according to (2)  Ω = −λ(c + 2L̂i) 2

y = −λ|ξ14 − L̂i| − λξ5
lim θ̂ − θ = 0 . where c verifies the dynamics (11) and we have the relation
t→∞
T
The proof of Theorem 1 goes with changing coordinates, z = ξ14 (c + ξ14 ) + ξ5 .
building an appropriate weak Lyapunov function and with z satisfying (11).

15974
Proceedings of the 20th IFAC World Congress
Toulouse, France, July 9-14, 2017 Pauline Bernard et al. / IFAC PapersOnLine 50-1 (2017) 15403–15408 15405

Convergence of these observers (11) or (12) is guaranteed
as long as Ω satisfies a persistent excitation condition
which, as proved in Bobtsov et al. [2015a], holds when
the rotation speed is sufficiently rich.
Inspired from nonlinear Luenberger observers, another
observer is proposed in Henwood et al. [2012]. It consists
in using m filters of the type (9), with poles λk , with k in
{1, . . . , m}, to obtain m equations in Ψ̂
This means that Ψ̂ converges to Ψ and Φ̂ to the ”equivalent
flux” Φs . But this time, it is not sufficient to compute
the argument of Ψ̂ − Ls i to obtain an estimate of θ,
since according to (15), it converges either to θ or θ + π
depending on the sign of Φ + 2L1 id . In fact, defining θ0 as
the argument of Ψ − Ls i and idq,0 as
 
i
idq,0 = id,0 = R(−θ0 ) i
q,0

(ck + 2L̂i)T Ψ̂ − (zk + L̂2 |i|2 ) = 0 (13) we have :

which are solved in a least square sense. It is proved in - if Φ + 2L1 id > 0, then Φs = Φ + 2L1 id , θ0 = θ,
Henwood [2014] that the matrix of the ck + L̂i is full id,0 = id and Φs − 2L1 id,0 = Φ > 0
column rank when ω stay away from 0, m ≥ 3 and the - if Φ + 2L1 id < 0, then Φs = −Φ − 2L1 id , θ0 = θ + π,
λk are chosen in a generic way. id,0 = −id and Φs − 2L1 id,0 = −Φ < 0 .
Actually, observer (11), observer (12) of Bobtsov et al. Therefore, computing the argument θ̂0 of Ψ̂−Ls i, and îdq,0
[2015a], or the one in Henwood et al. [2012], are identical defined by
except in their way of solving in Ψ̂ equations (13). The  
îd,0
former two solve (13) with only one λ (m = 1) but îdq,0 = = R(−θ̂0 ) i ,
dynamically along time. The later solve them at each time, îq,0
with at least two λ (m ≥ 2). and taking
In the remainder of the paper, we intend to compare the θ̂ = θ̂0 if Φ̂ − 2L1 îd,0 ≥ 0
performances of observer (6) introduced in the previous θ̂ = θ̂0 + π otherwise ,
section with those of this other family of observers, in
particular observer (11). we obtain convergence of θ̂ to θ. This convergence is a clear
argument in favor of observer (6) with respect to observer
4. PERFORMANCES (5). Indeed, the flexibility provided by the estimation of
Φ enables to apply the same observer to salient motors
4.1 Computational cost without losing convergence of θ. The same conclusions
hold for the observers presented in Section 3. Not to be
We already see that the small dimension of observer (6) forgotten, all this holds when id is constant.
and its great simplicity of implementation provides a
significant advantage. Indeed, in our matlab simulations,
CPU time was found to be twice smaller than for the other 4.3 Sensitivity to errors on R and L when (id , iq , ω) is
observers presented in Section 3. This numerical efficiency constant
constitutes an important feature since those observers are
intended to run online where processing power is often In Theorem 1, we claimed convergence for observer (6)
limited. assuming perfect knowledge of the resistance and the
inductance and the absence of saliency. Then, in the latter
4.2 Sensitivity to the presence of saliency when id is subsection, we extended this result to salient models as
constant long as the current in the dq frame id is constant. We
study here the possible consequences of having R̂ and L̂
According to Bodson and Chiasson [1998], the simplest different from R and L. For this we restrict our attention
way to take saliency into account in the model of a PMSM to the case where R(−θ) i = idq and ω are constant.
is to keep (1) but to replace the expression (2) of the total This configuration is often considered in practice, since it
flux by corresponds to a constant rotation speed with a constant
   
cos 2θ sin 2θ cos θ load torque. In this case the model with saliency made of
Ψ = L0 i + L1 sin 2θ − cos 2θ i + Φ sin θ (14) (1) and (14) has an asymptotic behavior given by
where L1 is a second order inductance. Thanks to the u = R(θ)udq , i = R(θ)idq , Ψ = R(θ)Ψdq
identity where udq , idq and Ψdq are constant satisfying
    
cos 2θ + 1 sin 2θ cos θ − sin θ cos θ 0  
sin 2θ − cos 2θ + 1 sin θ cos θ = 2 sin θ 0 Φ
ωJΨdq = udq − Ridq , Ψdq − Ls idq = 0s
the above expression of Ψ can be rewritten as
  where  
cos θ 0 −1
Ψ − (L0 − L1 )i = (Φ + 2L1 id ) sin θ (15) J = 1 0 .
with the notation   Let Ψeq be defined as
i
idq = id = R(−θ) i . (16) 1
q Ψeq = J −1 R(θ)(udq − R̂idq )
This shows that, when id is constant, we recover exactly ω
the design model (7) provided we replace L and Φ by It satisfies
Ls = L0 − L1 , Φs = |Φ + 2L1 id | . a) Ψ̇eq = u − R̂i
Hence Theorem 1 holds in the case with saliency at least
when the signals obtained from the motor are such that i-e the same dynamics as Ψ but with R̂ instead of R.
 
id is constant. Specifically, by implementing observer (6) 1 −1  
with L̂ = Ls , we directly obtain : b) Ψeq − L̂i = R(θ) J udq − R̂idq − L̂idq . (17)
ω
lim |Ψ̂(ψ, φ, t) − Ψ(t)| + |Φ̂(ψ, φ, t) − Φs | = 0   
t→∞ constant

15975
Proceedings of the 20th IFAC World Congress
15406
Toulouse, France, July 9-14, 2017 Pauline Bernard et al. / IFAC PapersOnLine 50-1 (2017) 15403–15408

R + 1%R L + 1%L Parameter Motor 1 Motor 2

Regime variable : Figure 1 constant : 2000 rpm
ω Obs θ̃ (rad) Φ̃/Φ θ̃ (rad) Φ̃/Φ Ld 0.72 mH 0.142 mH
(6) 0.015 2.6 % 5.4 10−3 0.3 % Lq 0.78 mH 0.62 mH
500 rpm Φ 8.94 mWb 18.5 mWb
(11) 0.015 2.6 % 5.2 10−3 0.3 %
R 0.151 Ω 0.023 Ω
(6) 3.8 10−3 0.7 % 5.4 10−3 0.3 % Pairs of poles (np ) 10 2
2000 rpm
(11) 3.3 10−3 0.6 % 4.9 10−3 0.3 %
Table 2. Parameters for Motor 1 and 2.
Table 1. Sensitivity of observers (6) and (11) we applied in open-loop (and offline) the observers (6) and
with respect to R and L at two different
electrical rotation speeds with the notation (11) to real data obtained from two PMSM used in test
beds at IFPEN : Motor 1 and Motor 2. The available data
θ̃ = |θ̂ − θ| and Φ̃/Φ = |Φ̂−Φ|
Φ . are the measurements of voltages um and currents im in
the αβ fixed frame, the measurement of the rotor position
Thus, with Φeq the constant real number defined as θm , the physical parameters given in Table 2.
 
 1 −1    The norms of um and im for each motor are given in

Φeq =  J udq − R̂idq − L̂idq  Figures 2 and 5. Note that unlike Motor 2, Motor 1 is
ω     submitted consecutively to four regimes : around 150 rpm,
 Φs J −1  450 rpm, 1000 rpm and finally 1500 rpm (see Figure 1).

=  0 + [R − R̂] + Ls − L̂ idq 
ω The motors differ in terms of saliency. According to
we have Bodson and Chiasson [1998], L0 and L1 in (14) are given
|Ψeq − L̂i|2 − Φ2eq = 0 . by
It follows that Ψeq is solution of the model (1)-(3) if we Ld + L q L d − Lq
L0 = , L1 = .
replace (R, L, Φ) by (R̂, L̂, Φeq ). So, according to Theorem 2 2
and therefore
1, the observer (6), implemented with R̂ and L̂, gives L s = L0 − L 1 = Lq .
lim |Ψ̂(ψ, φ, t) − Ψeq (t)| + |Φ̂(ψ, φ, t) − Φeq | = 0 . We conclude that saliency is weak for Motor 1 ( LL0 ≈ 4%),
1
t→∞
     L1
 Φ −1  but significant for Motor 2 ( L0 ≈ 80%).
Hence Φ̂ converges to  0s + [R − R̂] Jω + Ls − L̂ idq .
And with θ̂ computed as the argument of Ψ̂ − L̂i, we have We have implemented the observers using the measured
asymptotically values um and im as u and i, and an explicit Euler scheme
with the sample time (dt1 = 10−4 s, dt2 = 2 10−5 s).
  
We chose the parameters of the observers to ensure the
  cos(θ̂ − θ)
Ψ̂ − L̂i responses have all approximately the same time constant
sin(θ̂ − θ)   (γ(6) = 20000, γ(11) = 50000, λ = 50) and so that
= R(−θ) Ψeq − L̂i convergence is obtained in less than two rotations of the
1   motor. The results are presented in Figures 3-4 for Motor
= J −1 udq − R̂idq − L̂idq 1 and in Figures 6-7 for Motor 2. The performances are
ω   J −1
 globally better for Motor 1 than Motor 2, but it is mainly
Φ due to the fact that the data were noisier for the latter.
= 0s + [R − R̂] + Ls − L̂ idq , (18)
ω For θ (Figures 3 and 6), both observers provide similar
results, with a final oscillatory error of amplitude smaller
where we have used (17). In
 other
  words the−1error θ̂−θ con-
 than 0.05 rad for Motor 1 (0.09 rad for the last regime)
Φ
verges to the argument of 0s + [R − R̂] Jω + Ls − L̂ idq . and 0.12 rad for Motor 2. But (the mean value of ) the
Up to the first order, this is exactly the same result as estimation θ̂ does not converge to the measurement θm .
the one obtained in Henwood [2014] for the Luenberger There are static errors. They are likely due, in part at least,
observer presented in Henwood et al. [2012]. Of course we to an offset in the sensor for θm . But there is more since,
recover the fact that without any errors on R and L, the according to Figure 3, these biases depend on the regime.
One explanation comes from (18) where the regime ω
asymptotic value of Φ̂ is Φs and θ̂ converges to θ. appears explicitly. Another possible explanation has been
We illustrate formula (18) via simulations with ideal data proposed and studied in Henwood [2014]. It is the effects
obtained for L = 0.65 mH, R = 0.167 Ω, Φ = 7.3 mWb, of the dynamics of the sensors providing the measurements
id = −3.46 A, iq = 6 A, for two different regimes. The um and im . When they are modelled simply by
results are given in Table 1 for observers (6) and (11). i̇m = −τi (im − i) , u̇m = −τu (um − u)
Both observers were implemented with an Euler scheme the phase shift of these first order systems (depending on
with dt = 1.2 10−4 s and give similar results. The reader
may check that the absolute error on θ and the relative the regime) is directly translated in a static error on θ̂ and
error on Φ correspond exactly to the expected theoretical consequently on Φ̂. We refer the reader to Henwood [2014]
errors. for more details.
5. TESTS WITH REAL DATA Concerning Φ (Figures 4 and 7), although both observers
provide again the same mean for the final errors, the
To illustrate the results above about the sensitivity with transient of observer (11) seems to be more oscillatory.
respect to the parameters, to saliency, but also to noise, This difference could be explained by the fact that Φ̂ is

15976
Proceedings of the 20th IFAC World Congress
Toulouse, France, July 9-14, 2017 Pauline Bernard et al. / IFAC PapersOnLine 50-1 (2017) 15403–15408 15407

directly estimated by observer (6) while it is reconstructed

from the norm of Ψ̂−Lq i for observer (11). Here again (the 1500

mean value of) Φ̂ does not tend to Φ. Let us concentrate on

the data from Motor 2 and from the first regime of Motor
1, where the norm of the current is constant. Assuming
that the offset θ̂ − θm mentioned above is only due to
1000
the position of the sensor and therefore that θ̂ is actually
the correct rotor position, we compute id as the first
component of R(−θ̂)i and find
Motor 1: id,1 = −4.2 A
Motor 2: id,2 = −201 A . 500

If the values of R, Ld , Lq and Φ in Table 2 are correct, we

can expect Φ̂ to tend to Φs = Φ + 2L1 id , i-e
Motor 1: Φs,1 = 9.2 mWb
Motor 2: Φs,2 = 115 mWb . 0
0 5 10 15 20 25 30
This is verified for both motors on Figures 4 (first regime)
and 7. We could conclude that the values of R and L used
in the observers are correct. Unfortunately we cannot go Fig. 1. Motor 1 : Regime (rpm).
further in the analysis since, for the other regimes in Figure
4, the steady state is not reached.

6. CONCLUSION 14

We have introduced a new rotor position observer for sen- U

sorless permanent magnet synchronous motors (PMSM). 12
I
It is designed from a non salient model and uses measure-
ments of voltages and current, and estimations of resis- 10
tance and inductance. But it does not need the knowledge
of the magnet flux. We have claimed its convergence in an
ideal context and for a rotating motor. 8

We have compared it with the equivalent observers pro-

posed in Henwood et al. [2012], Henwood [2014] and 6
Bobtsov et al. [2015b]. The main difference is that this
new observer is less demanding in terms of computations.
4
On the other hand it gives qualitatively the same kind of
performance, in terms of speed of convergence, sensitivity
to errors in the resistance or the inductance and also in 2
presence of saliency. 0 5 10 15 20 25 30

At least three important issues remain to be addressed:

a) Sensitivity to measurement noise or more interestingly Fig. 2. Motor 1 : Norm of the voltage um (V) and current
the definition of a tuning policy in presence of such im (A).
disturbances. This kind of study has been made in
Henwood [2014] for the Luenberger observer proposed
in Henwood et al. [2012] . The same kind of tools should
be useful in our context. 3
b) Use of the observer in closed loop. Tests via simulations Obs (11)
or test beds for the observers in Henwood et al. [2012] Obs (6)
and Bobtsov et al. [2015b] are reported in those papers. 2

But as far as we know no theoretical results are yet

available. 1
c) Extension to non salient models. We are unaware of
any observer for this case.
0
ACKNOWLEDGEMENTS
-1
We are extremely grateful to Jérémy Malaizé and Wissam
Dib for suggesting the gradient scheme (6). Our deep
thanks go also to Nicolas Henwood and Wissam Dib -2
for making available the data on Motor 1 and Motor 2
respectively and for their valuable help in exploiting them.
-3
0 5 10 15 20 25 30
REFERENCES
P.P Acarnley and J.F Watson. Review of position-
sensorless operation of brushless permanent-magnet ma- Fig. 3. Motor 1: Error θ̂ − θm (rad) given by observers (6)
chines. IEEE Transactions on Industrial Electronics, 53 and (11) , where θm is a measurement of θ
(2):352–362, 2006.

15977
Proceedings of the 20th IFAC World Congress
15408
Toulouse, France, July 9-14, 2017 Pauline Bernard et al. / IFAC PapersOnLine 50-1 (2017) 15403–15408

0.14
0.02
Obs (11)
0.018 0.12
Obs (6)
)
0.016
0.1

0.014

0.08 Obs (11)

0.012
Obs (6)
)
0.01
0.06

0.008
0.04
0.006

0.004 0.02

0.002
0
0 0.1 0.2 0.3 0.4 0.5 0.6
0
0 5 10 15 20 25 30

Fig. 7. Motor 2: Φ̂ given by observers (6) and (11)

Fig. 4. Motor 1: Φ̂ given by observers (6) and (11) compared to Φ
compared to Φ
A. Bobtsov, A. Pyrkin, and R. Ortega. A new approach for
estimation of electrical parameters and flux observation
of permanent magnet synchronous motors. Int. J.
300 Adapt. Control Signal Process., 30:1434–1448, 2015a.
A. Bobtsov, A. Pyrkin, R. Ortega, S. Vukosavic,
A. Stankovic, and E. Panteley. A robust globally con-
250
vergent position observer for the permanent magnet
synchronous motor. Automatica, 61:47–54, 2015b.
200
A. Bobtsov, D. Bazylev, A. Pyrkin, S. Aranovsky, and
R. Ortega. A robust nonlinear position observer for
U synchronous motors with relaxed excitation conditions.
150 I International Journal of Control, 2016.
M. Bodson and J. Chiasson. Differential-geometric meth-
ods for control of electric motors. Int. J. Robust. Non-
100
linear Control, 8:923–254, 1998.
N. Henwood. Estimation en ligne de paramtres de
50
machines lectriques pour vhicule en vue d’un suivi
de la temprature de ses composants. PhD thesis,
Control and System Center, MINES ParisTech,
0 https://pastel.archives-ouvertes.fr/pastel-00958055,
0 0.1 0.2 0.3 0.4 0.5 0.6
2014.
N. Henwood, J. Malaizé, and L. Praly. A robust nonlinear
Fig. 5. Motor 2 : Norm of the voltage um (V) and current luenberger observer for the sensorless control of sm-
pmsm : Rotor position and magnets flux estimation.
im (A). IECON Conference on IEEE Industrial Electronics So-
ciety, 2012.
J. Lee, J. Hong, K. Nam, R. Ortega, L. Praly, and A. As-
tolfi. Sensorless control of surface-mount permanent-
4
magnet synchronous motors based on a nonlinear ob-
Obs (11)
Obs (6)
server. IEEE Transactions on Power Electronics, 25(2):
3
290–297, 2010.
J. Malaizé, L. Praly, and N. Henwood. Globally convergent
2
nonlinear observer for the sensorless control of surface-
mount permanent magnet synchronous machines. IEEE
1
Conference on Decision and Control, 2012.
R. Ortega, L. Praly, A. Astolfi, J. Lee, and K. Nam.
0
Estimation of rotor position and speed of permanent
magnet synchronous motors with guaranteed stability.
-1
IEEE Transactions on Control Systems Technology, 19
(3):601–614, 2011.
-2
F. Poulain, L. Praly, and R. Ortega. An observer for
permanent magnet synchronous motors with currents
-3
and voltages as only measurements. IEEE Conference
-4
on Decision and Control, 2008.
0 0.1 0.2 0.3 0.4 0.5 0.6 J-G. Romero, R. Ortega., Z. Han, T. Devos, and F. Mal-
rait. An adaptive flux observer for the permanent mag-
net synchronous motor. Int. J. Adapt. Control Signal
Fig. 6. Motor 2: Error θ̂ − θm (rad) given by observers (6) Process., 30:473–487, 2016.
and (11) , where θm is a measurement of θ

15978