NORPIE 2004, Trondheim, Norway

Saliency Modeling in Radial Flux Permanent

Magnet Synchronous Machines
Sigurd Ovrebo (sigurd@smartmotor.no) Prof. Roy Nilsen, Prof. Robert Nilssen
Smartmotor AS, Stiklestadveien 1 Norwegian University of Science and
N-7041 Trondheim, Norway Technology, 7491 Trondheim, Norway

d − axis
Abstract—Sensorless control of Permanent Magnet
Synchronous Machines is popular for several reasons: cost saving
and system reliability. The basis for most low and zero speed
sensorless control is that there exist a difference in the direct and
quaderature inductance of the machines. This difference is also Ld
reefed to as saliency. In this paper the different contributors to
q − axis Lq
the resulting saliency is modeled. This type of modeling is useful
for several reasons: increased understanding of the fundamentals
of sensorless control, incorporating the combined effects of
leakage and main inductance variations, incorporating the effects
of loading of the machine and finally it gives valuable insight
when the saliency ratio is attempted increased by machine
design.
Figure 1 Salient pole synchronous machine
Index Terms—Sensorless control, saliency modeling, PMSM
II. SALIENCY MODELING
I. INTRODUCTION In Sensorless Control of PMSM the rotor inherent saliency or

S ensorless control of electric machines has been a research

topic for more than two decades. The objective for this
work is to eliminate the need for a position sensor by
using the machine as the position sensor. The position sensor,
cabling and connectors have been a source of failure for motor
control applications. For small drives the sensor contributes
considerable to the overall cost. The basis for most low and
zero speed sensorless control [1-5] is the presents of a
difference in the d and q inductances. This difference is
referred to as saliency. There are several sources of saliencies
in PM machines: rotor inherent saliency, saturation based
saliency (yoke, teeth) [3], rotor stator teeth harmonics [1],
lamination direction based saliency [7], eddy current based
saliency [8], rotor eccentricity based saliency. In this paper a
single saliency model is developed from the rotor inherent
saliency and the saturation induced saliency. The term
saliency can be understood when considering a salient pole
synchronous machine structure (Figure 1). In this structure the
rotor has an inherent saliency due to the shape of the rotor. In
the d-axis the air-gap is small while there is a large air-gap in
the q-axis. The d-axis inductance will in this case be larger
then the inductance in the q-axis. The injection based
sensorless control schemes [1-5] extract rotor position
information from the difference in the inductance.
the saturation based saliency is usually used as the basis for
the saliency. The presented saliency model is simplified as
only these sources of saliency are included. In order to
illustrate the saturation based saliency a simplified model of a
30 kW IPMSM machine was simulated in FEMLab. The
relative permeability function (from supplier) was included in
the FEM simulation. In Figure 2 the machine is unloaded and
only the flux from the permanent magnets contributes to the
field. Three regions of the machine can be seen to have low
relative permeability values: the permanent magnet (in rotor),
the stator teeth (in the main flux path) and the stator yoke (in
the main flux path). The stator teeth are more saturated
compared to the portions of the stator yoke that is saturated.
d − axis q − axis
Yoke saturation
Tooth saturation

Figure 2 Relative permeability, IPMSM (8-pole) FEM simulation

NORPIE 2004, Trondheim, Norway

The modeling of saliency is done by modeling the inductance

in a dq-rotor oriented reference frame. The inductance is
separated in two parts: the leakage inductance and the main
inductance. In this way the model can be used to incorporate
the saliency sensed by carrier signals that is concentrated in
the leakage flux paths or main flux path (or both). The
sensorless schemes [1-5] will in general sense the resulting
inductance but the high frequency carrier flux distribution
determines the components of main- or leakage flux. In the
first stage of the modeling the machine will be modeled at no
load (only flux from the permanent magnets). The load
dependency is included in the last section of the modeling.
Figure 3 illustrates why the main and leakage inductance
should be modeled separately: the main flux and the leakage
Figure 4 Relative permeability for 50,100,200 and 400 Hz excitation
flux are 90 electrical degrees spatially shifted. frequencies (Congent M235-35 lamination-supplier data).

In order to simplify the modeling some simplifications are magnetic field and the time constant for the material to return
made: to its original sate is very large.
• Only fundamental components are considered
• The flux from the magnets and the windings are
assumed sinusoidal distributed over the air gap. The A. No load model
distributed windings are simplified by representing In the no load model only the flux from the permanent
them as a two axis model. magnets are considered. The inductance will be modeled in
• Saturation effects are assumed to be sinusoidal dq-rotor reference frame and the main and leakage inductance
distributed (superposition applied). will be modeled separately.
• Relative permeability assumed to wary linear with 1) Main inductance
flux density In the first stage the machine is considered as a uniform
structure with no magnets or saturation.
These simplifications may seem crude. Some lamination
materials may have extremely non linear relation between Ld −main = L0 (1)
relative permeability and flux density. The machine in the test
setup has laminations from Congent (M235-35). The relative Lq−main = L0 (2)
permeability is derived from the BH curves from the supplier.
Figure 4 shows the relative permeability for the laminations The sources of saliency for a no load condition is shown in
(measurement errors in the 100 and 200 Hz curves). From the Figure 5. These effects are included in the model as:
50 Hz curve an approximate linear relation can be obtained in
the region 0.4 to 1.3 T. The non linearity in the lower region is Ld −main = L0 − L0 rd − L0ts − L0 ys (3)
related to the first magnetization curve of the material. When
the machine model is considered this effect is neglected as the
laminations have been exposed to a L0 - main inductance (symmetrical machine)
s
β L0rd - rotor design (low relative permeability in magnets
and induced currents in magnets)
L0ts - stator tooth saturation
ψs
Is L0 ys - yoke saturation

The yoke saturation is assumed to have minimal impact on the

Leakage q-axis inductance. If high frequency carrier is used there may
flux ξ αs be induced currents in the magnets due to the carrier flux. The
regions effect of induced currents in the magnets is reduced d-axis
inductance. This effect is included in the model by the term
L0rd (combined effect of low permeability and induced
currents). The induced current is highly dependent on the
carrier frequency [8].

Figure 3 Current distribution, current and flux space vectors

NORPIE 2004, Trondheim, Norway

βs stator teeth Stator

saturation Leakage Flux

q
d
permanent
magnets

θ αs Rotor
stator yoke
saturation Figure 6 Leakage flux path

3) Resulting no load model

In the resulting no load model the leakage and main
inductance are summed up in each axis:

Figure 5 Sources of saliency – no load Ld = Ld −main + Ld −leakage

(8)
2) Leakage inductance Ld = L0 − L0 rd − L0ts − L0 ys + Lσ − Lσ ys
If we consider a flux vector (space vector) the leakage flux is Lq = Lq−main + Lq −leakage
located 90 electrical degrees spatially shifted compared to the (9)
main flux distribution. When the dq- inductances are modeled Lq = L0 + Lσ − Lσ ts
the reference system is referred to the main flux. The effects
on the leakage inductance are there fore shifted 90 electrical The difference between Ld and Lq is the basis for saliency
degrees. The basis for the leakage inductance model is also a in the machine. The position dependent parts of the inductance
uniform structure with no saturation: can be illustrated by expressing the inductance on vector
form:
Ld −leakage = Lσ (4)
Lq−leakage = Lσ (5)
L = ΣL + ∆1Le j 2θ (10)
where:
The leakage flux paths are illustrated in Figure 6. Three
regions may lead to modeling of the inductance: stator teeth,
stator yoke and rotor surface. Variations (or saturation) in the
θ - rotor position
rotor surface is neglected in this model as the surface of the
test machine is uniform and the saturation level is modest. If Ld + Lq
we consider a stator flux vector oriented in the magnet d-axis ΣL =
2
the leakage flux path for this main vector can be influenced by (11)
yoke saturation (saturation due to flux from the permanent 2 L0 − L0 rd − L0 ts − Lys + 2 Lσ − Lσ ts − Lσ ys
ΣL =
magnets). This effect is included in the model as a reduction 2
of the d-axis leakage inductance by a factor Lσ ys . If a stator Lq − Ld
current vector is oriented in the q-axis the leakage inductance ∆1L =
2
may be influenced by tooth saturation (saturation due to flux (12)
from the permanent magnet). This effect is included in the − Lσ ts + Lσ ys + L0 rd + L0 ts + L0 ys
∆1L =
model as a reduction of the q-axis leakage inductance by a 2
factor Lσ ts :
If the main inductance variation is used as the reference the
variation in the leakage inductance influence the magnitude of
Ld −leakage = Lσ − Lσ ys (6)
position dependent inductance. If the leakage yoke saturation
Lq−leakage = Lσ − Lσ ts (7) term ( Lσ ys ) in (12) is larger than the leakage tooth saturation
Lσ - leakage inductance (symmetrical machine) term ( Lσ ts ) the leakage terms adds to the main inductance
Lσ ys - yoke saturation effect on leakage inductance saliency. On the other hand if Lσ ts > Lσ ys the leakage terms
Lσ ts - tooth saturation effect on leakage inductance oppose the position dependent terms from the main
inductance.
NORPIE 2004, Trondheim, Norway

B. Load dependency
Stator
Several factors may change from an unloaded to a loaded
machine: flux and saturation levels, flux and saturation
orientation and saturation regions. In this model the Load
orientation of the resulting stator flux is considered in the first dependent
Saturated
stage. In the next stage saturation from the load current is Leakage Flux
Regions
included. In order to simplify the model the control optimal
control of the PMSM is not considered. The control strategy
of the machine is assumed to be id = 0 and iq = is* .
Ψs
I q Lq
∆θ Rotor

Ψm Figure 8 Saturation due to leakage flux from load current

Figure 7 Load dependency on stator flux
In this model the saturation is assumed sinusoidal distributed
The stator flux increases when the machine is loaded (Figure and linear. The different terms may be combined to form a
7). The resulting flux vector is shifted by a term ∆θ due to resulting single saliency:
the loading of the machine.
L = ΣL' + ∆ 4 Le jξ (20)
⎡ iq Lq ⎤
∆θ ≈ a sin ⎢ ⎥ (13) sin( ∆θ ) ∆ 3 L
⎢ ψ M 2 + (iq Lq ) 2 ⎥ ξ = a tan( ) (21)
⎣ ⎦ ∆ 2 L + cos( ∆θ ) ∆ 4 L
The shift in the stator flux results in a shift of the stator yoke
( ∆ 2 L + cos( ∆θ )∆ 3L )
and teeth saturation regions. In addition to the main flux 2
∆4L = + (sin( ∆θ ) ∆ 3 L) 2 (22)
effects the leakage effects must also be included. Figure 8
illustrates typical saturation regions due to the leakage flux
from the load current. This type of saturation will result in a The resulting model in (20), (21) and (22) describes the
relation between the different sources of saliency in the
load dependent term in the q-axis leakage inductance ( Llds ):
IPMSM and load current in the q-axis.
Lq−leakage = Lσ − Lσ ts − Llds (14)

III. EXPERIMENTAL WORK

Lq = L0 + Lσ − Lσ ts − Llds (15)
A 30 kW radial flux IPMSM (motor data in Table 1) was used
The spatial shift in the stator flux is incorporated in the model in the experimental work. The machine was designed for high
by shifting the saturation based saliency by the angle ∆θ : fundamental flux (500-600 Hz) operation during field
L = ΣL' + ∆ 2 Le − j 2θ + ∆ 3 Le − j 2(θ +∆θ ) (16) weakening.
Where the upper index ‘ indicates that there may be a change
Table 1 IPMSM parameters
in the saturation level due to the loading (as the stator flux
PN UN IN nN Rs [Ω] Ld[mH] Lq[mH] p
increase) and the terms in (16) can be expressed as: [kW] [V] [A] [rpm]
' L'd + L'q 30 110 181 2500 0.011 0.123 0.381 4
ΣL =
2
(17)
1 ⎛ 2 L0 − L0 rd − L0 ts − Lys + ⎞
' ' The self and mutual inductance was estimated by applying a
ΣL = ⎜
'
⎟ transient excitation [4] to the machine. The excitation
2 ⎜⎝ 2 Lσ − L'σ ts − L'σ ys − Llds ⎟⎠ sequence (t1-t5) is illustrated in Figure 9. The time interval
between t1/t2 and t4/t5 was 12.5 us and t2/t3 and t3/t4 was 25
L us.
∆ 2 L = 0 rd (18)
2
− L'σ ts + L'σ ys + L'0 ts + L'0 ys − Llds
∆3L = (19)
2
NORPIE 2004, Trondheim, Norway

011 100 011 100

Ia
a
IPMSM
A A

inverter
V Ua
Control signals to inverter leg A,B,C

B n

V Uc V Ub
b

C c

Figure 10 Measurements of self and mutual inductance

In the saliency modeling in the previous section variations in

t1 t2 t3 t4 t5 the leakage inductance where assumed. If these variations are
t
included in the model the self inductance can be expressed as:
Figure 9 Transient excitation in the zero vector

Figure 10 shows the connection between the inverter and the Laσ (θ ) = Laσ 0 − ( Laσ ts − Laσ ys ) cos(2θ ) (28)
IPMSM. The pulses were applied across the a-phase and
La (θ ) = Laσ 0 + La 0 + ( Laσ ts − Laσ ys − Lg ) cos(2θ ) (29)
neutral point while the b and c phase was open. The current in
the a-phase, the applied voltage and the induced voltage in the
b and c-phase was measured. Current and voltage Where:
measurements where sampled at time instances t2, t3 and t4.
From these measurements the self and mutual inductance can Laσ 0 -average leakage inductance phase a
be derived as:
Laσ ts -tooth saturation dependent term phase a
t4 − t2
Laa = U a (23) Laσ ys -yoke saturation dependent term phase a
I a (t2 ) − 2 I a (t3 ) + I a (t4 )
t 4 − t2
Lab = U b (24) The b and c-phase did not conduct any current and the
I a ( t 2 ) − 2 I a ( t3 ) + I a ( t 4 ) leakage inductance in these phases where not measured. The
t4 − t 2 expression in (27) is therefore assumed to be correct for the
Lac = U c (25) mutual inductance term.
I a (t2 ) − 2 I a (t3 ) + I a (t4 )
The classical modeling uses the average inductance as the
The rotor was turned by hand and the inductance as a function basis for the model. In this paper a maximum inductance was
of the position was estimated by current and voltage used. In order to relate the two models (29) was used in the
measurements. In Figure 11 there is a large difference in the stationary inductance matrix and transformed to the dq-
position dependent part of the self and mutual inductance. In representation. By comparing the dq-inductance the relation
classical machine modeling [9] the self and mutual inductance between the models where found (Table2).
are modeled as:
0.4
la [mH]

La (θ ) = Laσ + La 0 − Lg cos(2θ ) (26)

0.3
0.2
0.1
L
Lab (θ ) = − a 0 − Lg cos(2θ − 120)
0
(27) 200 300 400 500 600 700 800 900

2 0.1
0
lab [mH]

-0.1

Where: -0.2
-0.3
-0.4
200 300 400 500 600 700 800 900
Laσ - leakage inductance phase a 0.1

La 0 -average inductance (minus leakage) phase a

0
lac [mH]

-0.1
-0.2
Lg -position dependent inductance phase a -0.3
-0.4
200 300 400 500 600 700 800 900

rotor position[el.deg]
Figure 11 Self and mutual inductance measured with transient excitation

Table 2 Relation between Theoretical model and measurements

NORPIE 2004, Trondheim, Norway

Theoretical model Modified standard V. REFERENCES

model(measurements) [1] M.W. Degner; “Flux, Position, and Velocity Estimation in AC Machines
Using carrier signal injection”, Ph.D. Dissertation, Dep. Of Mechanical
L0 + Lσ 3La 0 Engineering, University of Wisconsin - Madison 1998
+ Laσ [2] M.J. Corley, R.D. Lorenz; “Rotor position and velocity estimation for a
2
salient-pole permanent magnet synchronous machine at standstill and
L0 rd + L0ts + L0 ys Lg high speeds” Industry Applications, IEEE Transactions on , Volume: 34
Issue:4,July-Aug.1998 Page(s): 784 -789
[3] M. Schroedl; “Sensorless Control of A.C. Machines”, Ph.D. thesis,
Lσ ts Laσ ts Wien, 1992
2 [4] S. Ovrebo, R. Nilsen; “New self sensing scheme based on INFORM,
heterodyning and Luenberger observer”, IEMDC'03. IEEE
Lσ ys Laσ ys International, Volume: 3 , 1-4 June 2003,Pages:1819 - 1825 vol.3
[5] J. Ha,K. Ide, T. Sawa, S. Sul, “Sensorless position control and initial
2 position estimation of an interior permanent magnet motor” Thirty-Sixth
IAS Annual Meeting. Conference Record of the 2001 IEEE
[6] M.L. Aime, M.W. Degner, R.D. Lorenz; “Saturation measurements in
From the measurements done in this work it is difficult to AC machines using carrier signal injection”, IAS, 1998. The 1998 IEEE
estimate all the components in the saliency model. Table 3 , Volume: 1 , 12-15 Oct. 1998 Pages:159 - 166 vol.1
shows the terms that can be derived from the measurement. [7] T Wolbank, J Machl “Anisotropy in Induction Machine Lamination and
its Influence on Mechanical Sensorless Control and Conditioning
Monitoring” EPE Toulouse 2003-10-22
Table 3 Measured inductance terms
[8] M. Leksell, L. Harnefors, H.P. Nee; “Machine Design Considerations
La 0 Laσ Laσ st − Laσ yt Lg Ld Lq for Sensorless Control of PM Motors”, Proceedings International
Conference on Electrical Machines, Istanbul, September 1998
[mH] [mH] [mH] [mH] [9] A.E. Fitzgerald, C. Kigsley, S.D. Umans; “Electrichal Machinery”,
[mH] [mH]
McGraw-Hill Inc. 1983
0.125 0.025 0.0375 0.0875 0.1 0.325

The magnitude of the position dependent terms in the mutual

and self inductance indicates that there is a position
dependency in the leakage inductance. These variations can
be incorporated in the machine model by using saliency
modeling as presented in this paper. Table 3 shows the
estimated inductance terms for the test machine used in this
work obtained by transient excitation (Figure 9). The term
( Laσ st − Laσ yt ) describes the position dependent term in the
leakage inductance. In classical machine modeling this term
will equal zero.

IV. CONCLUSION
A new saliency model is presented in this paper. The model
incorporates the combined saturation effect in the stator
teeth/yoke, induced currents in the magnets and load
dependency. The model suggests that the leakage inductance
should be modeled with a position dependent part.
Measurements on an IPMSM show that there can be a
difference in the position dependent term in the self and
mutual inductance when transient excitation is used to
estimate the inductance. This difference can be incorporated in
the machine model by adding position dependent part to the
leakage inductance.

The new saliency model describes: the different sources of

saliency, the combined effect of saliency in main and leakage
inductance, saliency due to induced currents in magnets and
the load dependency in the machine.