See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/221941309

Signal injection and averaging for position estimation of Permanent-Magnet

Synchronous Motors

Article · March 2012

DOI: 10.1109/CDC.2012.6426887 · Source: arXiv

CITATIONS READS
10 111

4 authors:

Al Kassem Jebai François Malrait

Schneider Electric 36 PUBLICATIONS   435 CITATIONS   
11 PUBLICATIONS   107 CITATIONS   
SEE PROFILE
SEE PROFILE

Philippe Martin Pierre Rouchon

MINES ParisTech MINES ParisTech
162 PUBLICATIONS   9,155 CITATIONS    360 PUBLICATIONS   12,271 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Control of partial differential equations View project

Flat systems View project

All content following this page was uploaded by Philippe Martin on 31 May 2014.

The user has requested enhancement of the downloaded file.

51st IEEE Conference on Decision and Control
December 10-13, 2012. Maui, Hawaii, USA

Signal injection and averaging for position estimation of

Permanent-Magnet Synchronous Motors
Al Kassem Jebai, Franois Malrait, Philippe Martin and Pierre Rouchon

Abstract— Sensorless control of Permanent-Magnet Syn- of the approach and the necessity of considering saturation
chronous Motors at low velocity remains a challenging task. to correctly estimate the position.
A now well-established method consists in injecting a high-
frequency signal and use the rotor saliency, both geometric
and magnetic-saturation induced. This paper proposes a clear
II. A N ENERGY- BASED MODEL FOR THE SATURATED
and original analysis based on second-order averaging of how PMSM
to recover the position information from signal injection;
this analysis blends well with a general model of magnetic A. Notations
saturation. It also experimentally demonstrates the relevance In the sequel we denote by xij := (xi , xj )T the vector
for position estimation of a simple parametric saturation model
made from the real numbers xi and xj , where ij can be dq,
recently introduced by the authors.
αβ or γδ. We also define the matrices
( ) ( )
I. I NTRODUCTION cos µ − sin µ 0 −1
Mµ := and K := ,
sin µ cos µ 1 0
Permanent-Magnet Synchronous Motors (PMSM) are
widely used in industry. In the so-called “sensorless” mode and we have the useful relation
of operation, the rotor position and velocity are not measured
dMµ
and the control law must make do with only current measure- = KMµ = Mµ K.
ments. While sensorless control at medium to high velocities dµ
is well understood, with many reported control schemes B. Energy-based model
and industrial products, sensorless control at low velocity
remains a challenging task. The reason is that observability The model of a two-axis PMSM expressed in the syn-
degenerates at zero velocity, causing a serious problem in chronous d − q frame reads
the necessary rotor position estimation. dϕdq
A now well-established method to overcome this problem = udq − Ridq − ωK(ϕdq + ϕm ) (1)
dt
is to add some persistent excitation by injecting a high- J dω 3 τL
frequency signal [1] and use the rotor saliency, whether geo- = iTdq K(ϕdq + ϕm ) − (2)
n2 dt 2 n
metric for Interior Permanent-Magnet machines [2], [3], [4], dθ
or induced by main flux saturation for Surface Permanent- = ω, (3)
dt
Magnet machines [5]. To get a good position estimation
under high-load condition it is furthermore necessary to take with ϕdq flux linkage due to the current; ϕm := (λ, 0)T
into account cross-saturation [6], [7], [8]. constant flux linkage due to the permanent magnet; udq
impressed voltage and idq stator current; ω and θ rotor
The contribution of this paper is twofold: on the one hand
(electrical) speed and position; R stator resistance; n number
it proposes a clear and original analysis based on second-
of pole pairs; J inertia moment and τL load torque. The
order averaging of how to recover the position information
physically impressed voltages are uαβ := Mθ udq while
from signal injection; this analysis blends well with a general
the physically measurable currents are iαβ := Mθ idq . The
model of magnetic saturation including cross-saturation. On
current can be expressed in function of the flux linkage
the other hand the paper experimentally demonstrates the
thanks to a suitable energy function H(ϕd , ϕq ) by
relevance for position estimation of the saturation model
introduced by the authors [9]. ( )
∂1 H(ϕd , ϕq )
The paper is organized as follows: section II extends idq = Idq (ϕdq ) := , (4)
∂2 H(ϕd , ϕq )
the saturation model introduced in [9] and studies its first-
order observability; in section III position estimation by where ∂k H denotes the partial derivative w.r.t. the k th
signal injection is studied thanks to second-order averaging; variable [10], [11]; without loss of generality H(0, 0) = 0.
finally section IV experimentally demonstrates the relevance Such a relation between flux linkage and current naturally
encompasses cross-saturation effects.
A-K. Jebai, P. Martin and P. Rouchon are with the Centre Automatique et For an unsaturated PMSM this energy function reads
Systmes, MINES ParisTech, 75006 Paris, France {al-kassem.jebai,
philippe.martin, pierre.rouchon}@mines-paristech.fr 1 2 1 2
F. Malrait is with Schneider Toshiba Inverter Europe, 27120 Pacy-sur- Hl (ϕd , ϕq ) = ϕd + ϕ
Eure, France francois.malrait@schneider-electric.com 2Ld 2Lq q

978-1-4673-2066-5/12/$31.00
978-1-4673-2064-1/12/$31.00 ©2012
©2012 IEEE
IEEE 7608
where Ld and Lq are the motor self-inductances, and we than performing the exact inversion, we can take advantage
recover the usual linear relations of the fact the coefficients αi,j are experimentally small. At
ϕd first order w.r.t. the αi,j we have ϕd = Ld id + O(|αi,j |) and
id = ∂1 Hl (ϕd , ϕq ) = ϕq = Lq iq + O(|αi,j |); plugging these expressions into (7)-
Ld 2
ϕq (8) and neglecting O(|αi,j | ) terms, we easily find
iq = ∂2 Hl (ϕd , ϕq ) = . (
Lq
ϕd = Ld id − 3α3,0 L2d i2d − α1,2 L2q i2q
Notice the expression for H should respect the symmetry )
− 4α4,0 L3d i3d − 2α2,2 Ld L2q id i2q (9)
of the PMSM w.r.t the direct axis, i.e. (
ϕq = Lq iq − 2α1,2 Ld Lq id iq −
H(ϕd , −ϕq ) = H(ϕd , ϕq ), (5) )
2α2,2 L2d Lq i2d iq − 4α0,4 L3q i3q . (10)
which is obviously the case for Hl . Indeed (1)–(3) is left
unchanged by the transformation Notice the matrix
( )
Gdd (idq ) Gdq (idq ) ( −1 )
(ud , uq , ϕd , ϕq , id , iq , ω, θ, τL ) → := DIdq Idq (idq ) , (11)
Gdq (idq ) Gqq (idq
(ud , −uq , ϕd , −ϕq , id , −iq , −ω, −θ, −τL ).
with coefficients easily found to be
C. Parametric description of magnetic saturation
1
Magnetic saturation can be accounted for by considering Gdd (idq ) = + 6α3,0 Ld id + 12α4,0 L2d i2d + 2α2,2 L2q i2q
a more complicated magnetic energy function H, having Ld
Hl for quadratic part but including also higher-order terms. Gdq (idq ) = 2α1,2 Lq iq + 4α2,2 Ld id Lq iq
From experiments saturation effects are well captured by 1
Gqq (idq ) = + 2α1,2 Ld id + 2α2,2 L2d i2d + 12α0,4 L2q i2q ,
considering only third- and fourth-order terms, hence Lq

H(ϕd , ϕq ) = Hl (ϕd , ϕq ) is by construction symmetric; indeed

( )
∑
3 ∑
4
∂11 H(ϕd , ϕq ) ∂21 H(ϕd , ϕq )
+ α3−i,i ϕ3−i i
α4−i,i ϕ4−i i DIdq (ϕdq ) =
d ϕq + d ϕq . ∂12 H(ϕd , ϕq ) ∂22 H(ϕd , ϕq )
i=0 i=0

This is a perturbative model where the higher-order terms and ∂12 H = ∂21 H. Therefore the inductance matrix
appear as corrections of the dominant term Hl . The nine ( ) ( )−1
Ldd (idq ) Ldq (idq ) Gdd (idq ) Gdq (idq )
coefficients αij together with Ld , Lq are motor dependent. := .
Ldq (idq ) Lqq (idq ) Gdq (idq ) Gqq (idq )
But (5) implies α2,1 = α0,3 = α3,1 = α1,3 = 0, so that the
energy function eventually reads is also symmetric, though this is not always acknowledged
in saturation models encountered in the literature.
H(ϕd , ϕq ) = Hl (ϕd , ϕq ) + α3,0 ϕ3d + α1,2 ϕd ϕ2q
+ α4,0 ϕ4d + α2,2 ϕ2d ϕ2q + α0,4 ϕ4q . (6) E. First-order observability
From (4) and (6) the currents are then explicitly given by To emphasize position estimation is not easy at low speed,
ϕd we study the observability of (1)–(3), augmented by
id = + 3α3,0 ϕ2d + α1,2 ϕ2q + 4α4,0 ϕ3d + 2α2,2 ϕd ϕ2q (7)
Ld dτL
ϕq = 0, (12)
iq = + 2α1,2 ϕd ϕq + 2α2,2 ϕ2d ϕq + 4α0,4 ϕ3q , (8) dt
Lq
around a permanent trajectory defined by
which are the so-called flux-current magnetization curves.
To conclude, the model of the saturated PMSM is given 0 = udq − Ridq − ωK(ϕdq + ϕm )
by (1)–(3) and (7)-(8), with ϕd , ϕq , ω, θ as state variables. 3 T τL
0 = idq K(ϕdq + ϕm ) −
The magnetic saturation effects are represented by the five 2 n
parameters α3,0 , α1,2 , α4,0 , α2,2 , α0,4 . dθ
= ω.
D. Model with id , iq as state variables dt
The model of the PMSM is usually expressed with cur- Notice such a permanent trajectory is not a steady state point
rents as state variables. This can be achieved here by time unless ω := 0 since θ hence uαβ and iαβ are time-varying
differentiating idq = Idq (ϕdq ), (ϕdq , idq , udq , ω, τ L are on the other hand constant). Also
the rationale for augmenting the model with (12) is that it is
didq dϕdq
= DIdq (ϕdq ) , usually also desired to estimate an unknown load torque.
dt dt From iαβ = Mθ idq we then get
dϕ
with dtdq given by (1). Fluxes are then expressed as ϕdq =
−1 δiαβ = δMθ idq + Mθ δidq = Mθ (Kidq δθ + δidq ), (13)
Idq (idq ) by inverting the nonlinear relations (7)-(8); rather

7609
and similarly for uαβ , where δ stands for the differential. III. P OSITION ESTIMATION BY HIGH FREQUENCY
The linearization of (1)–(3) and (12) around a permanent VOLTAGE INJECTION
trajectory is then
A. Signal injection and averaging
dδϕdq
= MθT (δuαβ − Rδiαβ ) − K(ϕdq + ϕm )δω
dt ( ) A general sensorless control law can be expressed as
+ ω (ϕdq + ϕm )δθ − Kδϕdq
J dδω 3 uαβ = Mθc uγδ (15)
= δiTαβ Mθ K(ϕdq + ϕm ) dθc
n2 dt 2 = ωc (16)
3 T( ) δτL dt
− idq (ϕdq + ϕm )δθ − Kδϕdq − dη ( )
2 n = a Mθc iγδ , θc , η, t (17)
dδθ dt ( )
= δω
dt ωc = Ωc Mθc iγδ , θc , η, t (18)
dδτL ( )
= 0, uγδ = Uγδ Mθc iγδ , θc , η, t , (19)
dt
where we have used udq − Ridq = ωK(ϕdq + ϕm ). where the measured currents iαβ = Mθc iγδ are used to
On the other hand time differentiating δiαβ yields compute the feedback law uγδ , ωc and the evolution of the
( internal states η, θc of the controller.
dδiαβ dMθ dδθ dδidq )
= (Kidq δθ + δidq ) + Mθ Kidq + It will be convenient to write the system equations (1)–(3)
dt dt dt dt in the γ −δ frame defined by xγδ := Mθ−θc xdq , which gives
( dδϕdq )
= ωKδiαβ + Mθ Kidq δω + DIdq (ϕdq )
dt dϕγδ
= uγδ − Riγδ − ωc Kϕγδ − ωKMθ−θc ϕm (20)
where we have used δidq = DIdq (ϕdq )δϕdq . Therefore dt
J dω 3 τL
dδiαβ = iTγδ K(ϕγδ + Mθ−θc ϕm ) − (21)
[DIdq (ϕdq )]−1 MθT 2
n dt 2 n
dt dθ
dδϕdq = ω, (22)
= [DIdq (ϕdq )]−1 Kidq δω + + LC(δiαβ , δuαβ ) dt
dt
= KΦ(ϕdq )δω + Φ(ϕdq )ωδθ + LC(δiαβ , δuαβ ) where from (7)-(8) currents and fluxes are related by

where iγδ = Mθ−θc Idq (Mθ−θ

T
ϕ ). (23)
c γδ
−1
Φ(ϕdq ) := ϕm + ϕdq + K[DIdq (ϕdq )] KIdq (ϕdq )
To estimate the position we will superimpose on some
and LC(δiαβ , δuαβ ) is some matrix linear combination of desirable control law (19) a fast-varying pulsating voltage,
δiαβ and δuαβ . Similarly
( )
J dδω 3 δτL uγδ = Uγδ Mθc iγδ , θc , η, t + u
eγδ f (Ωt), (24)
= − Φ(ϕdq )δθ − + LC(δiαβ , δuαβ ). (14)
n2 dt 2 n
where f is a 2π-periodic function with zero mean and
Now Φ(ϕdq ) is non-zero in any reasonable situation, hence eγδ , Uγδ may depend on Mθc iγδ , θc , η, t. The constant pul-
u
Φ(ϕdq ) and KΦ(ϕdq ) are independent vectors. If ω ̸= 0 it sation Ω is chosen “large”, so that f (Ωt) can be seen as
is thus clear that δθ and δω can be expressed in function a “fast” oscillation; typically Ω := 2π × 500 rad/s in the
dδi
of iαβ , dtαβ and uαβ ; as a consequence δτL is a function experiments in section IV.
dδi d2 δi dδu
of iαβ , dtαβ , dt2αβ and uαβ , dtαβ by (14); finally δϕdq If we apply this modified control law to (20)–(22), then it
is by (13) a function of δiαβ . In other words the linearized can be shown the solution of the closed loop system is
system is observable by δiαβ .
If ω = 0 only δω can be recovered from iαβ , dtαβ
dδi eγδ
u 1
ϕγδ = ϕγδ + F (Ωt) + O( 2 ) (25)
δτL
and uαβ ; as a consequence only Φ(ϕdq )δθ + n can be Ω Ω
1
recovered from (14), and nothing new is gained by further ω = ω + O( 2 ) (26)
differentiation. In other words the linearized system is not Ω
1
observable, as pointed out in [10]. θ = θ + O( 2 ) (27)
Ω
Estimating the rotor position at low speed is thus inher- 1
ently difficult. Yet it is doable when the motor exhibits some θc = θc + O( 2 ) (28)
Ω
saliency, geometric or saturation-induced, with the help of 1
permanent excitation such as signal injection. Indeed a more η = η + O( 2 ), (29)
Ω
thorough analysis along the lines of [12], [13] would reveal
the system is in that case observable in the nonlinear sense where F is the primitive of f with zero mean (F clearly
provide the motor exhibits some saliency. has the same period as f ); (ϕγδ , ω, θ, θc , η) is the “slowly-

7610
varying” component of (ϕγδ , ω, θ, θc , η), i.e. satisfies K2 (y, εσ, σ)
:= ∂1 f1 (y,εσ, σ)u1 (y, εσ, σ) − ∂1 u1 (y, εσ, σ)g1 (y, εσ)
dϕγδ
= uγδ − Riγδ − ω c Kϕγδ − ωKMθ−θc ϕm 1 dF 2 (σ)
dt = [f 1 , fe1 ](y, εσ)F (σ) + ∂1 fe1 (y, εσ)fe1 (y, εσ)
J dω 3 T τL 2 dσ
= iγδ K(ϕγδ + Mθ−θc ϕm ) − ∫
n2 dt 2 n 1 T
g2 (y, εσ) := K2 (y, εσ, s)ds = 0.
dθ T 0
=ω
dt We have set
dθc
= ωc [f 1 , fe1 ](y, εσ) := ∂1 f 1 (y, εσ)fe1 (y, εσ)−∂1 fe1 (y, εσ)f 1 (y, εσ)
dt
dη ( ) ∫σ ∫T ∫σ
= a Mθc iγδ , θc , η, t , and F (σ) := 0 f (s)ds − T1 0 0 f (s)dsdσ, i.e. F is the
dt (of course T -periodic) primitive of f with zero mean.
where Translating back to the original variables eventually yields
the desired result (25)–(29).
iγδ = Mθ−θc Idq (Mθ−θT
ϕγδ ) (30)
( c
) B. Position estimation
ω c = Ωc Mθc iγδ , θc , η, t
( ) We now express the effect of signal injection on the
uγδ = Uγδ Mθc iγδ , θc , η, t . currents: plugging (25) into (23) we have
Notice this slowly-varying system is exactly the same iγδ = Mθ−θc +O( 1 )
as (20)–(22) acted upon by the unmodified control law (15)– Ω2
( ( eγδ
u 1 ))
(19). In other words adding signal injection: Idq Mθ−θ +O(
T
ϕγδ +
1 F (Ωt) + O( 2 )
c ) Ω Ω
• has a very small effect of order O( Ω2 ) on the mechan-
1 Ω2
1
ical variables θ, ω and the controller variables θc , η = iγδ + eiγδ F (Ωt) + O( 2 ), (31)
Ω
• has a small effect of order O( Ω ) on the flux ϕγδ ; this
1

effect will be used in the next section to extract the where we have used (30) and performed a first-order expan-
position information from the measured currents. sion to get
( T ) T u eγδ
The proof relies on a direct application of second-order eiγδ := M
θ−θ c DIdq Mθ−θ c ϕγδ Mθ−θ c
averaging of differential equations, see [14] section 2.9.1 Ω
( ( T )) T u eγδ
and for the slow-time dependance section 3.3. Indeed setting −1
= Mθ−θc DIdq Idq Mθ−θ iγδ Mθ−θ . (32)
x := (ϕγδ , ω, θ, θc , η), ε := Ω1 and σ := εt (hence ẋ = ε dσ
dx
), c c Ω

(20)–(22) acted upon by the modified control law (15)–(18) We will see in the next section how to recover eiγδ and iγδ
and (24) is in the so-called standard form for averaging (with from the measured currents iγδ . Therefore (32) gives two
slow-time dependance) (redundant) relations relating the unknown angle θ to the
dx ( ) known variables θc , eidq , iγδ , u
edq , provided the matrix
= εf1 (x, εσ, σ) := ε f 1 (x, εσ) + fe1 (x, εσ)f (σ) , ( ( T )) T
dσ −1
S(µ, iγδ ) := Mµ DIdq Idq Mµ iγδ Mµ
with f1 T -periodic w.r.t. its third variable and ε as a small
parameter. Therefore its solution can be approximated as effectively depends on its first argument µ. This “saliency
( ) condition” is what is needed to ensure nonlinear observabil-
x(σ) = z(σ) + εu1 z(σ), εσ, σ + O(ε2 ), ity. The explicit expression for S(µ, iγδ ) is obtained thanks
to (11). In the case of an unsaturated magnetic circuit this
where z(σ) is the solution of
matrix boils down to
dz ( )
1
= εg1 (z, εσ) + ε2 g2 (z, εσ) 0
dσ S(µ, iγδ ) = Mµ Ld
MµT
0 L1q
and ( L −L Ld −Lq
)
∫ Ld +Lq 1 + Ldd +Lqq cos 2µ Ld +Lq sin 2µ
1 T = 2Ld Lq Ld −Lq L −L
g1 (y, εσ) := f1 (y, εσ, s)ds = f 1 (y, εσ) Ld +Lq sin 2µ 1 − Ldd +Lqq cos 2µ
T
∫ σ0
( ) and does not depend on iγδ ; notice this matrix does not
v1 (y, εσ, σ) := f1 (y, εσ, s) − g1 (y, εσ) ds depend on µ for an unsaturated machine with no geometric
0
∫ σ saliency. Notice also (32) defines in that case two solutions
e
= f1 (y, εσ) f (s)ds on ] − π, π] for the angle θ since S(µ, iγδ ) is actually a
0
∫ function of 2µ; in the saturated case there is generically only
1 T
u1 (y, εσ, σ) := v1 (y, εσ, σ) − v1 (y, εσ, s)ds one solution, except for some particular values of iγδ .
T 0 There are several ways to extract the rotor angle informa-
= fe1 (y, εσ)F (σ) tion from (32), especially for real-time use inside a feedback

7611
 TABLE I (a)
R ATED AND SATURATION PARAMETERS OF THE TEST MOTOR . 0

Rated power 1500 W Ld 7.9 mH −20

Rated current In (peak) 5.19 A Lq 8.2 mH
Rated speed 3000 rpm α3,0 L2d In 0.0551 −40

Position in deg
Rated torque 6.06 Nm α1,2 Ld Lq In 0.0545
−60
n 5 α4,0 L3d In2 0.0170
R 2.1 Ω α2,2 Ld L2q In2 0.0249 −80
λ 0.155 Wb α0,4 L3q In2 0.0067
−100 θ − θc
b
θ − θc sat
−120
law. In this paper we just want to demonstrate the validity b
θ − θc no sat
−140
of (32) and we will be content with directly solving it through
0 20 40 60 80 100
a nonlinear least square problem; in other words we estimate
(b)
the rotor position as 2
2

% of rated Speed
eγδ
θb = θc + arg min eiγδ − S(µ, iγδ ) u . (33)
µ∈]−π,π] Ω 0
ω
C. Current demodulation ωc
−2
To estimate the position information using e.g. (33) it is 0 20 40 60 80 100
necessary to extract the low- and high-frequency components
iγδ and eiγδ from the measured current iγδ . Since by (31) % of rated Torque 200
(c)

iγδ (t) ≈ iγδ (t) + eiγδ (t)F (Ωt) with iγδ and eiγδ by construc-
tion nearly constant on one period of F , we may write 100
∫
1 t
iγδ (t) ≈ iγδ (s)ds τL
T t−T 0
∫t 0 20 40 60 80 100
i (s)F (Ωs)ds
eiγδ (t) ≈ t−T∫ γδ , (d)
T
0
F 2 (Ωs)ds 20
Voltage in V

2π
where T := Ω . Indeed as F is 2π-periodic with zero mean,
urd
∫ t ∫ t ∫ t 10 γ

iγδ (s)ds ≈ iγδ (t) e

ds + iγδ (t) F (Ωs)ds urd
δ
t−T t−T t−T 0
0 20 40 60 80 100
= T iγδ (t) time in sec
∫ t ∫ t
iγδ (s)F (Ωs)ds ≈ iγδ (t) F (Ωs)ds Fig. 1. Long test under various conditions: (a) measured θ − θc , estimated
t−T t−T
∫ t θb − θc with and without saturation model; (b) measured speed ω, reference
speed ωc ; (c) load torque τL ; (d) voltages urd
+ eiγδ (t) F 2 (Ωs)ds γδ .
t−T
∫ T
= eiγδ (t) F 2 (Ωs)ds. B. Validation of the rotor position estimation procedure
0
1) Control law: Since the goal is only to test the validity
IV. E XPERIMENTAL R ESULTS
of the angle estimation procedure, a very simple V /f open-
A. Experimental setup loop (i.e. Ωc and Uγδ do not depend on iγδ ) control law
The methodology developed in the paper was tested on is used for (15)–(19); a fast-varying (Ω := 2π × 500 rad/s)
a surface-mounted PMSM with parameters listed in table I. square voltage with constant amplitude is added in accor-
The magnetic parameters were estimated using the procedure dance with (24), resulting in
of [9, section III]. Notice this motor has little geometric dθc
saliency (Ld ≈ Lq ) hence the saturation-induced saliency dt
= ωc (t)
is paramount to estimate the rotor position. uγδ = urd eγδ f (Ωt).
γδ (t) + ωc (t)ϕm + u
The experimental setup consists of an industrial inverter
(400 V DC bus, 4 kHz PWM frequency), an incremental Here ωc (t) is the motor speed reference; urdγδ (t) is a filtered
encoder, a dSpace fast prototyping system with 3 boards piece-wise constant vector compensating the resistive voltage
(DS1005, DS5202 and EV1048). The measurements are drop in order to maintain the torque level and the motor
sampled also at 4 kHz, and synchronized with the PWM eγδ := (e
stability; u u, 0)T with ue := 15 V.
frequency. The load torque is created by a 4 kW DC motor. The rotor position θb is then estimated according to (33).

7612
 (a)
2) Long test under various conditions (Fig. 1): Speed 400
and torque are slowly changed over a period of about two
minutes; the speed remains between ±2% of the rated speed 350

and the torque varies from 0% to 150% of the rated toque. 300
This represents typical operation conditions at low speed.

Position in deg
When the saturation model is used the agreement between 250

the estimated position θb and the measured position θ is very 200

good, with an error always smaller than a few (electrical)
degrees. By contrast the estimated error without using the 150

saturation model (i.e. with all the magnetic saturation pa- 100
rameters αij taken to zero) can reach up to 70◦ (electrical). θ
This demonstrates the importance of considering an adequate 50 b
θ
saturation model. 0
3) Very slow speed reversal (Fig. 2): The speed is very 0 2 4 6 8 10 12 14 16

slowly (in about 20 seconds) changed from −0.2% to +0.2% (b)

0.5

% of rated Speed
of the rated speed at 150% of the rated torque. This is
a very demanding test since the motor always remains in
the first-order unobservability region, moreover under high 0
ω
load. Once again the estimated angle closely agrees with the ωc
measured angle. −0.5
0 2 4 6 8 10 12 14 16
V. C ONCLUSION (c)
We have presented a new procedure based on signal injec- % of rated Torque 200

tion for estimating the rotor angle of a PMSM at low speed,

with an original analysis based on second-order averaging. 100

This is not an easy problem in view of the observability τL

degeneracy at zero speed. The method is general in the sense 0
0 2 4 6 8 10 12 14 16
it can accommodate virtually any control law, saturation
model, and form of injected signal. The relevance of the (d)
method and the importance of using an adequate magnetic 20
Voltage in V

saturation model has been experimentally demonstrated on a

urd
surface-mounted PMSM with little geometric saliency. 10 γ

urd
δ
R EFERENCES 0

[1] P. Jansen and R. Lorenz, “Transducerless position and velocity esti- 0 2 4 6 8 10 12 14 16

mation in induction and salient ac machines,” IEEE Trans. Industry
Applications, vol. 31, pp. 240–247, 1995.
[2] S. Ogasawara and H. Akagi, “An approach to real-time position
estimation at zero and low speed for a pm motor based on saliency,”
IEEE Trans. Industry Applications, vol. 34, pp. 163–168, 1998.
[3] M. Corley and R. Lorenz, “Rotor position and velocity estimation for a
salient-pole permanent magnet synchronous machine at standstill and
high speeds,” IEEE Trans. Industry Applications, vol. 34, pp. 784–789,
1998.
[4] T. Aihara, A. Toba, T. Yanase, A. Mashimo, and K. Endo, “Sensorless
torque control of salient-pole synchronous motor at zero-speed oper-
ation,” IEEE Trans. Power Electronics, vol. 14, pp. 202–208, 1999.
[5] J.-H. Jang, S.-K. Sul, J.-I. Ha, K. Ide, and M. Sawamura, “Sensorless
drive of surface-mounted permanent-magnet motor by high-frequency
signal injection based on magnetic saliency,” IEEE Trans. Industry
Applications, vol. 39, pp. 1031–1039, 2003.
[6] P. Guglielmi, M. Pastorelli, and A. Vagati, “Cross-saturation effects
in ipm motors and related impact on sensorless control,” IEEE Trans.
Industry Applications, vol. 42, pp. 1516–1522, 2006.
[7] N. Bianchi and S. Bolognani, “Influence of rotor geometry of an
interior pm motor on sensorless control feasibility,” in Industry Appli-
cations Conference, 2005, pp. 2553–2560.
[8] Y. Li, Z. Zhu, D. Howe, C. Bingham, and D. Stone, “Improved rotor-
position estimation by signal injection in brushless ac motors, ac-
counting for cross-coupling magnetic saturation,” IEEE Trans. Industry
Applications, vol. 45, pp. 1843–1850, 2009.
[9] A. Jebai, F. Malrait, P. Martin, and P. Rouchon, “Estimation of satu-
ration of permanent-magnet synchronous motors through an energy-
based model,” in IEEE International Electric Machines Drives Con-
ference, 2011, pp. 1316 –1321.
time in sec
Fig. 2. Very slow speed reversal: (a) measured θ, estimated θb with
saturation model; (b) measured speed ω, reference speed ωc ; (c) load
torque τL ; (d) voltages urd
γδ .
[10] D. Basic, F. Malrait, and P. Rouchon, “Euler-Lagrange models with
complex currents of three-phase electrical machines and observability
issues,” IEEE Trans. Automatic Control, vol. 55, pp. 212–217, 2010.
[11] D. Basic, A. K. Jebai, F. Malrait, P. Martin, and P. Rouchon, “Using
Hamiltonians to model saturation in space vector representations of
AC electrical machines,” in Advances in the theory of control, signals
and systems with physical modeling, ser. Lecture Notes in Control and
Information Sciences, J. Lévine and P. Müllhaupt, Eds. Springer,
2011, pp. 41–48.
[12] P. Vaclavek and P. Blaha, “Synchronous machine drive observability
analysis for sensorless control design,” in IEEE Int. Conf. on Control
Applications, 2007, pp. 1113 –1117.
[13] D. Zaltni, M. Abdelkrim, M. Ghanes, and J. Barbot, “Observability
analysis of PMSM,” in Int. Conf. on Signals, Circuits and Systems,
2009.
[14] J. A. Sanders, F. Verhulst, and J. Murdock, Averaging methods in
nonlinear dynamical systems, 2nd ed., ser. Applied Mathematical
Sciences. Springer, 2007, no. 59.

7613

View publication stats