2017 Twelfth International Conference on Ecological Vehicles and Renewable Energies (EVER)

Comparison between a Double Excitation

Synchronous Machine and a Permanent Magnet
Synchronous Machine According to Various
Constant Power Speed Ranges

K. Hoang and M. Gabsi L. Vido F. Gillon

SATIE, ENS Cachan SATIE, Universite de Cergy Pontoise L2EP, Ecole Centrale de Lille
94235 Cachan, France 95000 Cergy-Pontoise, France 59650 Villeneuve d' Ascq Cedex, France
Email: [email protected] Email: [email protected] Email: [email protected]
[email protected]

Abstract-This paper presents a comparison between and PMSM types according to different prototypes and
a Double Excitation Synchronous Machine (DESM) and applications will help designers have a proper design. To
a Permanent Magnet Synchronous Machine (PMSM). the authors' knowledge, unfortunately, such comparisons
Optimal designs of both machine types are compared
seems to be scarce in the literature, even though there has
regarding different constant power speed ranges (CPSR).
been massive research efforts for each machine type in
The optimization objectives are to minimize the active
transportation applications, such as [5]-[8] for PMSMs,
material cost and losses. Results show that the DESM is
more advantageous over the PMSM counterpart in loss
and [9]-[11] for DESMs.
saving at a wide CPSR, and the PMSM is capable of This paper aims at comparing a DESM and a PMSM
reducing the machine cost. according to various ranges of speed. The main reason
Index Terms-Constant power speed range, double ex­ behind is to target directly the flux controlling capability,
citation, electric vehicle, optimization. which is claimed to be the key feature of a double
excitation machine. The comparison result is expected
1. INTRODUCTION to give a base to decide in which circumstances, using
DESMs is more beneficial over the PMSM counterparts
A Double (or hybrid) Excitation Synchronous Ma­
regarding the machine cost and losses.
chine (DESM) is constructed based on a combina­
Aiming at a fair comparison, the PMSM model is de­
tion between a permanent magnet synchronous machine
rived by removing some parts from the DESM. Optimal
(PMSM) and additional field windings [1]-[3]. The pur­
designs of both machine types are compared by using
pose behind is to bring together advantages of a PMSM,
multi-physics models. The objectives are to minimize
which are high power density and high efficiency, and a
the active material cost and total losses at the base and
good capability in air-gap flux control of a wound field
maximum speeds. The work is organized as follows:
synchronous machine. This advantage seems to make
Section II presents a multi-physics model that mainly
the double excitation machine type a very interesting
contains a coupling between electromagnetic and thermal
candidate for transportation applications, where variable
models. Section III addresses the optimization approach..
speeds are usually required. Having a machine using
Section IV analyses the comparison results followed by
additional field windings, however, would result in a
a conclusion.
complex geometry, and performances of existing per­
manent magnets (PMs) might be weakened in certain
II. MULTI-PHYSI CS MODEL
prototypes [4]. In addition, using field windings is an
effective solution to reduce the air-gap flux. However, at The DESM prototype is shown in Fig. I(a) [12] with
the same time a torque reduction would occur. Therefore, two global field windings. The corresponding permanent
the air-gap flux control in a double excitation machine is magnet type is displayed in Fig. 1(b) by removing
a complex issue. Specific comparisons between DESM some parts from the DESM model. Machine parameters

978-1-5386-1692-5/17/$31.00 ©2017 European Union

are detailed in the appendix section. Flux paths in where {L is friction coefficient, P is equivalent dynamic
the DESM model are truly three dimensional (3-D) at bearing load [N], d is bearing diameter [mm], and n is
some positions; therefore, the 3-D finite element method rotating speed [rpm].
(FEM) is usaully required to obtain highly accurate 2) Power electronics model: The three-phase inverter
results. This method, however, substantially increases topology using pulse width modulation (PWM) is shown
the computation time, and it is not practical for the in Fig. 2.
optimization process. In this paper, the electromagnetic
model is performed by a generalized equivalent magnetic +

circuit network (EMCN), which is detailed in [13].

The thermal model is accomplished by using a lumped
parameter thermal network. In order to more accurately
calculate the total losses, mechanical and inverter losses
are also computed.
A. Mechanical and inverter losses
1) Mechanical model: Two types mechanical losses Fig. 2. Three phase inverter topology

presented in an electrical machine are the windage loss

in the air gap, which is due to the rotor's rotation, and The calculations of the inverter losses are fundamen­
the friction losses in the bearings. A proper calculation tally based on the sinusoidal load current assumption.
of these losses not only help designers have a more Moreover, a linearized loss model is assumed. The
accurate value of machine's efficiency but also provide switching loss energies Es is linearized given by (4a),
heat sources input to the thermal model. According and the conduction losses Pc for a single semiconductor
to [14], the windage loss of a cylinder rotating in a is calculated by (4b) [15].
concentric cylinder is computed based on the following
iv
assumptions: No axial flow exists, the gap is small Es = EST' --
Vv
.-.
Vr'ej ZTej- (4a)
compared to the radius and axial length, and the air in
the gap is homogeneous. Pc =
TT
va . Zv + r . Z·2v
'
(4b)
With those assumptions, the windage loss is calculated
as (1): where:

Wwindage nCdpR4w 3L (1) EST is rated switching loss energy given for the reference
commutation voltage and current Vr'ej and iTej
=

where p is air density in the air gap [kg/m3], R is average

Vv and iv indicate the actual commutation voltage and
air-gap diameter [m], w is rotor angle speed [rad/s], L
current, respectively
is axial length [m], and Cd is skin friction coefficient.
Va and r are semiconductors threshold voltage and
The friction coefficient Cd is defined as (2a) for the
differential resistance, respectively.
laminar flow, and (2b) for the transition and turbulence
The conduction loss in each IGBT and each free­
cases.
wheeling diode are expressed by (5a) and (5b), respec­
2
Cd (2a) tively.
=

Re
-

� = 2.04 + 1.768 . In ( Re � ) (2b)

(5a)
In a rotating machine, two bearings are usually located at
the driving and non-driving ends. The friction losses in (5b)
the bearings is apparently subject to rotor's properties
(the rotating speed and rotor weight) and bearing's where:
properties (the bearing size and friction coefficient). With iL: peak load current
a bearing using lubricant oil, after a certain time in use, A M cos ¢, M is modulation index, and ¢ is the
=

the friction coefficient increases due to the lubricant oil displacement angle between the load current and the
deterioration. It is, therefore, complicated to obtain an fundamental component of the phase voltage
accurate calculation of this friction loss. According to VCE, rCE and Vp, rp: threshold voltages and differential
SKF (a bearing manufacturer), the bearing friction losses resistances of the IGBT and diode, respectively.
can be computed by (3):
The current supply principle for a field winding is
Wbearing = 0.525 . 1O-4{LPdn (3) shown in Fig. 3.
 Field winding

End shield (solid)

Rotoric flux collector (solid)

Rotor core (laminated)

Stator core (laminated)

(a) (b)

Fig. 1. Machines for comparisons. (a) DESM model. (b) Corresponding PMSM model

r Ambient temperature
+ = Empirical coefficient
Rambicnt
_ Simple coefficient (conduction)

1--1" Outer stator yoke

II Rfield.yoke

1--1" Field windings 1--1" Inner stator yoke

Rend.field

Fig. 3. DC-DC converter for field windings

Phase winding\

Rend-active Rp ha."e-leelh RaiJ:-gap

The switching losses of the MOSFE T and diode are
expressed by (6a) and (6b), respectively. Rend winding - rotor

.gw.MOSPET = (Eon + EOff) . .Is (6a)

P,w.diode = Qr"T . Vd . .Is (6b)
Rbcaring
Shaft
where Eon and EOff are on and off switching energies of
the MOSFE T, .Is is switching frequency, Qr"T is reverse
Fig. 4. Thermal network. Thermal resistances: Rfield.yoke - between
recovery charge, and Vd is knee voltage of the diode. field winding and outer stator yoke, Rend.field - between armature end
The conduction losses of the MOSFE T and diode are and field windings, Rend-8£uve - between armature end and active parts,
Rend-rotor - between armature end winding and rotor surface, R hase-tccth
expressed by (7a) and (7b), respectively. p
- between armature active winding and stator teeth, Rair-gap - air gap,
2 RheariJlg - bearing. Losses: PCuF - field winding copper losses, PcuP
PCon.MOSFET = TCE . IS.rms (7a)
. - armature copper loss, Arons I - outer stator yoke core loss, Arons2
2
PCon.Diode = Tp ' ID•rrns + VP•o· (1 - D)zJ (7b) - inner stator yoke core loss, PironS3 - stator teeth core loss, Aronl{ -
rotor core loss
where TCE and Tp are differential resistance of the
MOSFE T, Is.rms is rms value of current flowing through
the MOSFE T, D is duty cycle, and VP•o is threshold
voltage of the diode. As it can be seen in Fig. 4, two types of thermal
resistance are presented: one uses a simple coefficient,
B. Lumped parameter thermal network analysis and the others need empirical coefficients. The first type
A lumped parameter thermal model is to benefit from can be easily computed based on the object geometry,
the fast computation while maintaining a good accuracy while the second one is more challenging, and the
for overall temperatures. The thermal network is shown coefficients based on experiments or design experiences
in Fig. 4, which is capable of considering the transient are usually required. All of thermal coefficients and
process by introducing thermal heat capacitors. experimental validations are detailed in [4].
 III. OPTIMIZATION APPROACH TABLE I
GEOMETRY VARIABLE RANGES
A. Optimization problem formulation
The optimization objectives are to minimize the ma­ Variable Description Range [mm] Prototype [mm]
chine's active material and the total losses at two operat­
Xl Stack length 30 -:- 60 40
ing points, which are the base (nb) and maximum (nm)
X2 Bridge thickness 3 -:- 12 7
speeds. nm is actually the maximum achievable speed.
X3 Field winding height 0.05 -:- 9 9
At this speed, the produced torque should be close to
X4 Tooth width 3 -:- 6.5 5.5
zero as shown in Fig. 5.
X5 Tooth length 10 -:- 25 16.5
X6 Azimuth PM thickness 4.5 -:- 10 6
X7 Azimuth PM length 10 -:- 32 24
Xs Shaft radius 10 -:- 20 17
Xg Side PM thickness 3 -:- 10 6

set to minimize the total losses at each operating point.

Speed
The upper level is to find machines with the lowest
cost and smallest loss (found in the lower level). This
approach implies that the lower level is a mono-objective
Fig. 5. Two operating points for optimization
optimization (only minimizing the total losses), and
The constant power speed range (CPSR) is character­ all constraints are dealt with at this level. Due to the
ized by a ratio, between the maximum and base speeds presence of field windings in the DESM, three control
given by (8). variables representing field and armature currents are
nm set. Whereas with the PMSM configuration, only two
, = ­ (8)
nb variables representing the armature current are presented.
The upper level is a unconstrained multi-objective opti­

{
The optimization problem is formulated as (9).
mization. In this research, both optimization levels will
j(X) _
Total losses use particle swarm optimization (PSO) technique [16],
mlmmlze _

x Co� [17].

s.t. Vmax � 150 [V] IV. RESULT AND DISCUSSION

(9)
(}max � 155 [0C] The machine cost is calculated based on normalized
T T* [Nm] values by assuming unit cost (cost per kilogram active
material) are I, 6 and 7 for iron, copper and ferrite
=

n n* [rpm]
=
permanent magnet (PM), respectively [18]. The loss is
where Vmax and (}max are maximum allowable voltage the sum of the copper losses, core losses, and power
and winding temperature, respectively. T and n are the electronics losses at the base (1547 rpm) and maximum
machine torque and speed, T* and n* are required torque speeds.
and speed. Targeting the flux controlling capabilities of both
With the DESM type, nine geometry variables are machine types, different CPSRs (i.e. different , ratios)
chosen for the upper optimization level as marked in will be examined. In this research, three values of ,
Fig. 6. With the PMSM type, only six variables Xl, X4, ranging from 3 to 5 will be evaluated.
X5, X6, X7 and Xs are used. Fig. 7 shows the comparison between optimal pareto­
The width of the field winding windows are fixed, fronts of the two machine types.
and the field windings are allowed to only change their As revealed in Fig. 7, the DESMs are capable of
heights ( X3 ) . In case the optimal solution does not prefer reducing the total losses compared to the PMSMs. This
to use field windings, X3 approaches zero. In order to advantage is more obvious when increasing ,. This
avoid computation errors, the lower limit of X3 is set to is due to the good flux weakening capability of the
0.05 mm. Variable ranges are summarized in Table. I. DESMs using two available options by field and arma­
ture windings, while a PMSM can only use negative
B. Bi-level optimization approach d-axis currents from the armature windings. Strong
In order to handle the optimization, a bi-level approach field weakenings with a PMSM requires highly negative
is used. The lower level is to find the optimal control d-axis currents; therefore, increasing much the copper
 Fig. 6. Geometry variables: Xl - stack length, X2 - bridge thickness, X3
- field winding height, X4 - tooth width, X5 - tooth length, X6 - azimuth PM thickness, X7 -azimuth PM length, Xs - shaft radius and Xg -
side PM thickness.

55 55 55
DESM DESM DESM
PMSM 45 PMSM PMSM
Normalized cost

45 45
Normalized cost

Normalized cost
D1
D2 D3
35 35 35
P1 P2 P3
25 25 25

15 15 15
0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1
Loss [kW] Loss [kW] Loss [kW]
(a) (b) (c)
Fig. 7. Comparison results according to various speed ranges. (a) I = 3. (b) 1=4. (c) 1=5

(a) (b) (c)

(d) (e) (f)

Fig. 8. Machines on the pareto-fronts: (a) D1. (b) D2. (c). D3. (d) P1. (e) P2. (f). P3

loss and possibly lead to a thermal limit violation as machine type, a machine at the middle of the pareto­
well. But using additional field windings, and moreover front is extracted and shown in Fig. 8
bigger volume would increase the machine cost as shown
Fig. 9 shows the field winding thickness (of the DESM
in Fig. 7 that DESM costs are higher in comparison
type) variation along the pareto-front in the direction of
with the ones of the PMSM type. In addition, PMSM
increasing the total losses. As it can be seen, in order
machine type finds it difficult to reach very high speeds
to reduce the total losses, a double excitation machine
(partly proved as a limited number of machine found
prefers to use more field windings.
when I 5 as in Fig. 7c)). In each case of I and
=

Optimal current controls are shown in Fig. lO. Several

 10
,. ,=5 TABLE II
0 ,=4 DESM PROTOTYPE CONFIGURATION
S �� 0 b-**
<IID1!ilI!
o�
§ 8
Ii> �o b. ,= 3
'" & "'1l@'oao8(jJ)
** Parameters Value
'" ,.,.
'"
" b. 0 o�
o§ 0 0
�
II b.b. Number of phases 3
.0. *
:s 6
b. b. Number of turns per phase 33
bJ)
" 0 Number of turns per field winding 150
0
;a b. b. �

.� b. b.b.
,f b. 1S.'! ,.,. ,.
Number of poles 12
"0
4
t b.
b. 0
0 * ,. ,. Machine length lIS mm
V
t£ b. b. &. @o
d' ,.
Outer stator diameter 92 mm
0 0 Inner stator diameter 57.5 mm
2
Number of slots 36
0.4 0.5 0.6 0.7 0. 8 0.9
Air-gap length 0.5 mm
Loss [kW]
PM residual flux density 0.4 T (ferrite PM)
Fig. 9. Field winding thickness along the pareto-front (in increasing
loss direction) according to various speed ranges

REFERENCES
remarks would be observed:
[1] Y. Amara, L. Vido, M. Gabsi, E. Hoang, A. Hamid Ben Ahmed,
I) A PMSM only uses the d-axis current to reduce
and M. Lecrivain, "Hybrid excitation synchronous machines:
the air-gap flux; therefore, a big amount of this Energy-efficient solution for vehicles propulsion," Vehicular
current is required. Whereas with double excitation Technology, IEEE Transactions on, vol. 58, no. 5, pp. 2137-
principle, a DESM can use both d-axis and field 2149, Jun 2009.
[2] J. Tapia, F. Leonardi, and T. Lipo, "Consequent-pole permanent­
currents. As a result, it offers a good opportunity
magnet machine with extended field-weakening capability,"
by a certain combination between these two cur­ Industry Applications, IEEE Transactions on, vol. 39, no. 6,
rents to minimize the total losses. pp. 1704-1709, Nov 2003.
2) There is no clear pattern for the trend of the [3] J. S. Hsu, "Direct control of air-gap flux in permanent-magnet
machines," IEEE Transactions on Energy Conversion, vol. IS,
current control in a DESM, and noises might be
no. 4, pp. 361-365, Dec 2000.
presented due to the combination of armature and [4] T.-K. Hoang, "Design optimization of double excitation
field currents, since certain combinations might synchronous machines in railway traction," PhD thesis,
lead to a same or almost the same optimum. Universite Paris-Saclay, Nov. 2016. [Online]. Available:
https:/lhal.archives-ouvertes.fr/tel-01408271
3) DESM machines tend to use the field current to
[5] Z. Yang, F. Shang, 1. P. Brown, and M. Krishnamurthy, "Com­
control the air-gap flux. parative study of interior permanent magnet, induction, and
switched reluctance motor drives for ev and hev applications,"
V. CONLUSLION IEEE Transactions on Transportation Electrification, vol. I,
no. 3, pp. 245-254, Oct 2015.
A comparison addressing the total losses and active [6] P. Lazari, J. Wang, and L. Chen, "A computationally efficient
material cost minimizations between a classical PM and design technique for electric-vehicle traction machines," IEEE
double excitation machines has been presented. The Transactions on Industry Applications, vol. 50, no. 5, pp. 3203-
3213, Sept 2014.
comparisons was done at different speed ranges. The
[7] A. Vagati, G. Pellegrino, and P. Guglielmi, "Comparison be­
total losses were computed at the based and maximum tween spm and ipm motor drives for ev application," in The
speeds. DESM machines, with more complex structures XIX International Conference on Electrical Machines - ICEM
due to the additional field windings, can provide flexible 2010, Sept 2010, pp. 1-6.
combinations between armature and field currents. This [8] Y. Dai, L. Song, and S. Cui, "Development of pmsm drives
for hybrid electric car applications," IEEE Transactions on
makes them capable of reducing the total losses at Magnetics, vol. 43, no. I, pp. 434-437, Jan 2007.
an expense of a bigger volume and more cost. The [9] E. Hoang, M. Lecrivain, S. Hlioui, and M. Gabsi, "Hybrid exci­
advantage of the field windings is more noticeable at tation synchronous permanent magnets synchronous machines
optimally designed for hybrid and full electrical vehicle," in
very high speed, where a PMSM has to struggle keeping
Power Electronics and ECCE Asia (ICPE ECCE), 2011 IEEE
a low air-gap flux. In a perspective, a more elaborate 8th International Conference on, May 20II, pp. 153-160.
comparison between these two machines on a specific [10] N. Patin, L. Vido, E. Monmasson, J. P. Louis, M. Gabsi, and
application will be carried out. M. Lecrivain, "Control of a hybrid excitation synchronous gen­
erator for aircraft applications," IEEE Transactions on Industrial
ApPENDIX Electronics, vol. 55, no. 10, pp. 3772-3783, Oct 2008.
[11] D. Fodorean, A. Djerdir, 1. A. Viorel, and A. Miraoui, "A
See Table. II. double excited synchronous machine for direct drive applica-
 o o ,= 3
,= 3 ,= 3
'
. '

0 ,=4
,=4 0 ,=4
"
,=5
,=5
" ,=5

� -3 � -3

;:
-4 � -4
.
.

O&,

-5'---c------::----=---
� -5 L-----�� -5L-----=----=----=--=-��___=_�_=_��.
_=_-=-=:-
-=- -:--::------::---::---
__:__
-c--
0.4 0.5 0.6 0.7 0. 8 0.9 1.0 1.1 0. 4 0.5 0.6 0.7 0. 8 0.9 1.0 1.1 0.4 0.5 0.6 0.7 0. 8 0.9 1.0 1.1
Loss [kW] Loss [kW] Loss [kW]
(a) (b) (c)

Fig. 10. Current controls along the pareto-front (in increasing loss direction). a) d-axis current in PMSM. b) d-axis current in DESM. c)
Field current in DESM

tion: Design and prototype tests," Energy Conversion, IEEE P ESC 04. 2004 IEEE 35th Annual, vol. 4, 2004, pp. 2836-2842.
Transactions on, vol. 22, no. 3, pp. 656-665, Sept 2007. [16] J. Aubry, H. B. Ahmed, and B. Multon, "Bi-objective sizing
[12] L. Vido, M. Gabsi, M. Lecrivain, Y. Amara, and F. Chabot, "Ho­ optimization of a pm machine drive on an operating profile,"
mopolar and bipolar hybrid excitation synchronous machines," in Electrical Machines (ICEM), 2010 XIX International Con­
in Electric Machines and Drives, 2005 IEEE International ference on, Sept 2010, pp. 1-7.
Conference on, May 2005, pp. 1212-1218. [17] J. Aubry, "Optimisation du dimensionnement d'une chaine
[13] K. Hoang, L. Vido, F. Gillon, and M. Gabsi, "Modeling of de conversion electrique directe incluant un systeme de lis­
double excitation synchronous motors using nodal based gen­ sage de production par supercondensateurs. application au
eralized equivalent magnetic circui." ISEF, September 2015. houlogenerateur searev," These, Ecole normale superieure de
[14] J. E. Vrancik, "Prediction of windage power loss in alternators," Cachan - ENS Cachan, Nov. 2011.
NASA, Tech. Rep., October 1968. [18] A. Saxena, "Performance and cost comparison of pm bldc
[15] M. Bierhoff and F. Fuchs, "Semiconductor losses in voltage motors for ceiling fan," in Power Electronics, Drives and Energy
source and current source igbt converters based on analytical Systems (PEDES), 2014 IEEE International Conference on, Dec
derivation," in Power Electronics Specialists Conference, 2004. 2014, pp. 1-5.