Design of an axial flux machine

for an in-wheel motor application

Master of Science Thesis

CHRISTIAN DU-BAR

Department of Energy and Environment

Division of Electric Power Engineering
CHALMERS UNIVERSITY OF TECHNOLOGY
Göteborg, Sweden, 2011
 Design of an axial flux machine
for an in-wheel motor application

CHRISTIAN DU-BAR

Department of Energy and Environment

Division of Electric Power Engineering
CHALMERS UNIVERSITY OF TECHNOLOGY
Göteborg, Sweden, 2011
Design of an axial flux machine for an in-wheel motor application

c Christian Du-Bar, 2011

Examiner:

Sonja Lundmark
Department of Energy and Environment
Chalmers University of Technology
SE-412 96 Göteborg
Sweden

Supervisor:

Joachim Lindström
Volvo Technology

Chalmers Reproservice
Göteborg, Sweden 2011
Abstract
This master thesis deals with the design of an axial flux machine for an in-
wheel motor application. Four double-sided axial flux machine topologies
are compared to select the most suitable one for designing a high torque
density machine that fits into a predefined geometry (diameter: 260 mm,
length: 72.5 mm). The yokeless and segmented armature (YASA) topology
is chosen, due to its very compact structure. The design is made by using
analytic and 3D finite element analysis alternately.

The geometry of the electromagnetic circuit is optimized to find the maxi-

mum rated torque (≈80 Nm) for proposed values of magnetic flux and current
densities (1.7 T and 7 A/mm2 ). These densities are chosen with the intension
of reaching a high utilization of both the copper and the iron in the machine,
when their proportions are determined by the maximal product of electric
and magnetic loading.

When the geometry of the electromagnetic circuit is determined, it is held

fixed. Possible different torque speed characteristics of the machine are then
investigated, with respect to current and voltage limitations together with
the turn selections of the windings.

A final design is proposed and its theoretical characteristics are evaluated and
presented. The power density, at 2400 rpm and 7 A/mm2 , is calculated to
5.3 W/cm3 . Furthermore, some suggestions for future work and modifications
are made.
Acknowledgements
This master thesis started in late autumn 2010 at Volvo Technology and was
continued at Chalmers University of Technology during the spring of 2011.
I am very grateful to the experienced and helpful persons that I have met
during this time and I express extra thanks to:

Joachim Lindström, my supervisor, provided great support and guidance

during the whole project. Sonja Lundmark, my examiner, for valuable ad-
vice and feedback regarding the report. Martin West, Volvo Technology, who
shared clever ideas about electric machine design as well as a very positive
attitude. Maddalena Cirani, Volvo Technology, for advice about axial flux
machines. Johan Åström, Chalmers University of Technology, who gener-
ously shared his experience of FEA and electric machine modeling. Stefan
Lundberg, Chalmers University of Technology, for time spent discussing and
answering my questions about the machine modeling in the dq-coordinate
system. Magnus Dahlström, Höganäs, for supply of data for SMC materials.

Christian Du-Bar
Göteborg, Sweden
June, 2011
Contents
1 Introduction 1
1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Previous work . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.3 Objective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.4 Scope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2 Axial flux permanent magnet machines 4

2.1 Principles of permanent magnet machines . . . . . . . . . . . 4
2.1.1 Torque production . . . . . . . . . . . . . . . . . . . . 4
2.1.2 Wave forms . . . . . . . . . . . . . . . . . . . . . . . . 6
2.1.3 Limiting factors . . . . . . . . . . . . . . . . . . . . . . 8
2.2 Topologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2.1 Single-sided . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2.2 Double-sided internal rotor double stator . . . . . . . . 9
2.2.3 Double-sided internal stator double rotor . . . . . . . . 10
2.2.4 Multi-disc . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3 Comparison of suitable topologies . . . . . . . . . . . . . . . . 11
2.3.1 Selection of topology . . . . . . . . . . . . . . . . . . . 16

3 Analytic models 17
3.1 Machine characteristics in the dq-coordinate system . . . . . . 17
3.2 Voltage limit . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.3 Pole pairs and stator segment combinations . . . . . . . . . . 21
3.4 Winding factors . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.5 Torque calculations . . . . . . . . . . . . . . . . . . . . . . . . 24
3.6 Magnetic flux and permanent magnet length . . . . . . . . . . 24
3.7 Rotor core length . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.8 Stator segment . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.9 Inductances . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.10 Losses and efficiency . . . . . . . . . . . . . . . . . . . . . . . 29
3.10.1 Copper losses . . . . . . . . . . . . . . . . . . . . . . . 29
3.10.2 Iron losses . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.10.3 Efficiency . . . . . . . . . . . . . . . . . . . . . . . . . 30

4 Analytic design 31
4.1 Number of pole pairs and stator segments . . . . . . . . . . . 31
4.1.1 Winding configuration and inductance . . . . . . . . . 34
4.2 Optimization of the geometry for maximal torque . . . . . . . 36
4.3 Investigation of torque speed characteristics . . . . . . . . . . 38
5 Finite element analysis 44
5.1 Model and implementation . . . . . . . . . . . . . . . . . . . . 44
5.1.1 Symmetry planes . . . . . . . . . . . . . . . . . . . . . 44
5.1.2 Windings . . . . . . . . . . . . . . . . . . . . . . . . . 44
5.1.3 Core material . . . . . . . . . . . . . . . . . . . . . . . 45
5.1.4 Permanent magnets . . . . . . . . . . . . . . . . . . . . 47
5.1.5 Losses . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
5.2 Optimization of the geometry for maximal torque . . . . . . . 48
5.2.1 Output power, losses and efficiency at 3000 rpm . . . . 51
5.2.2 Final geometry . . . . . . . . . . . . . . . . . . . . . . 54
5.3 Flux linkage and back-EMF . . . . . . . . . . . . . . . . . . . 54
5.4 Computation of the inductances . . . . . . . . . . . . . . . . . 57
5.5 Computation of the torque . . . . . . . . . . . . . . . . . . . . 58
5.5.1 Average and ripple . . . . . . . . . . . . . . . . . . . . 58
5.5.2 Cogging torque . . . . . . . . . . . . . . . . . . . . . . 58
5.5.3 Overload and the spacing between the stator shoes . . 59
5.6 Losses and efficiency . . . . . . . . . . . . . . . . . . . . . . . 61

6 Discussion, conclusion and future work 64

6.1 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
6.2 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
6.3 Future work . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
Abbreviations
AC alternating current
AFM axial flux machine
AFIR axial flux internal rotor
AFIS axial flux internal stator
BLDC brushless direct current
DC direct current
FEA finite element analysis
FFT fast Fourier transform
PM permanent magnet
PMSM permanent magnet synchronous machine
YASA yokeless and segmented armature
Symbols
Latin
A area m2
B magnetic flux density T
B̂1 fundamental magnetic flux density peak T
Di inner air gap diameter m
Do outer air gap diameter m
f frequency Hz
fel electric frequency Hz
H magnetic field strength A/m
Hc magnetic coercivity A/m
Iph phase current rms A
J current density rms A/m2
kd distribution factor
kp pitch factor
kw winding factor
Kcu−f ill copper fill factor
l length or thickness m
lg air gap length m
lpm permanent magnet length m
lr rotor back thickness m
L inductance H
Ld direct-axis inductance H
Lq quadrature-axis inductance H
m1 number of phases
ma amplitude modulation ratio
n number of turns per stator segment, index of harmonics
Nph number of turns per phase
Ns number of stator segments
P power W
p number of pole pairs
R resistance Ω
Ri inner air gap radius m
Ro outer air gap radius m
T torque Nm
U voltage V
Udc dc-link voltage V
z number of coils in a group
Greek

φ magnetic flux Wb
λ ratio between inner and outer diameter
ω mechanical angular frequency rad/s
ωel electrical angular frequency rad/s
ωw wheel angular frequency rad/s
ψm magnetic flux linkage due to permanent magnets Wb
µ permeability H/m
µ0 vacuum permeability H/m
µr,pm relative permeability of permanent magnets
σ conductivity S/m
σcu copper conductivity S/m
η efficiency
1 Introduction
1.1 Background
The last years’ cold winters have caused problems in the traffic when heavy
duty vehicles have been stuck on slippery roads with traffic congestion as
a consequence. European regulations require that at least 25% of the total
weight of the vehicle is carried by driving axles [1]. If electric motors are
added to axels that do not receive torque from the existing drive train, it
may be possible to reduce the number of vehicles stuck with spinning wheels
and also to make sure that the regulations are fulfilled.

Another interesting possibility when electric machines are added to the vehi-
cles is that the drive cycle for the existing drive train can be made smoother.
The electric machines can be used for power assistance when going uphill
and for regenerative breaking when going downhill. This may lead to an
increased average speed, lowered fuel consumption and thereby also reduced
impact on the environment and cost savings.

The axial flux permanent magnet machine is very interesting for vehicle ap-
plications due to its short axial length together with the high efficiency, when
using permanent magnets. It can be engine integrated but the efficiency can
be further increased if the drive train is minimized. When suitable, regarding
the machine size and torque demand, direct driven in-wheel motors without
gears can be used to improve the efficiency, lower the maintenance and offer
independent control of each wheel [2].

1.2 Previous work

Volvo 3P is interested in electrical wheel motors for trucks in long-haul ap-
plications and financed a feasibility study during 2008-2009. An initial study
to investigate potential performance improvement and a rough evaluation of
the available space inside the rim was carried out by Volvo Technology [3].

A planetary gear with a gear ratio, of 6:1 to 10:1, was proposed to reduce
the torque requirements of the electric machine, since the torque production
is strictly limited by its volume. On the other hand the torque rating of the
planetary gear is directly influenced by its own size or volume and the gear
ratio is thereby also limited.

Different motor types were considered, permanent magnet and switched re-

1
luctance machines become the most interesting candidates due to their high
torque density and high efficiency. The permanent magnet machine was fi-
nally chosen as most suitable for the application due to higher torque density,
lower torque ripple and not so critically small air gap needed.

A DC-link voltage of 600 V from the battery is used. Variations of ±100 V

depending on the state-of-charge and load should be considered. It was also
indicated that an increase of the DC-link voltage was of interest.

There are no fixed requirements of power and torque ratings for the machine
but the performance investigation in [3] indicates the following targets for
one machine:

• Pmax ≈ 50 − 70 kW
• Tmax ≈ 400 Nm
• VDC−min = 500 V

Detailed investigations of possible mechanical solutions for integrating an

electrical motor into the front wheel of a truck, were made as a master thesis
during 2010 [4]. Two final concepts, one to fit a radial flux machine and
one to fit a transverse flux machine were stated. The radial flux machine
solution requires more changes of the existing hub configuration than the
more compact transverse flux machine solution does. Due to known issues
regarding the power factor of transverse flux machines, the very compact
hub concept originally designed for a transverse flux machine is considered
for this design of an axial flux machine, see Figure 1.

The hub concept is not presented in the public version of

this thesis. Contact Volvo Technology for further
information.

The machine geometry restrictions:

Maximal diameter: 260 mm
Maximal length: 72.5 mm

Figure 1: Available space for the electric machine and the power electronics.

2
1.3 Objective
The objective is to design an electrical axial flux permanent magnet machine
that fits into the limited space specified and, if possible, meets the targets
indicated. If it is not achieved, the obtained maximum power and torque
ratings should be presented.

1.4 Scope
This master thesis combines analytic design and finite element analysis (FEA)
based design of an axial flux permanent magnet machine. As it is difficult
to carry out a complete and accurate analytic design, in comparison with
a radial flux machine, FEA and analytic calculations are used alternately
during the design process. The analytic part is used to speed up the process,
calculate initial values etc., but also to verify the FEA solutions and to make
appropriate decisions about modifications. In general, the design process is
an iterative process which may be looped several times before a satisfactory
solution is reached.

Before the design process starts, some basic theory about permanent magnet
machines is presented. Four variations of two major topologies (internal ro-
tor double stator and internal stator double rotor) are compared to find the
most suitable one for the given requirements.

The design process can be subdivided into the following steps:

• Selection of topology

• Selection of pole number

• Optimization of the electromagnetic design

- Analytic before choice of material

- FEA after choice of material

• Investigation of torque speed characteristic

• Evaluation

3
2 Axial flux permanent magnet machines
The geometries of the permanent magnet axial flux machines in this report
are described in the cylindrical coordinate system shown in Figure 2. The
machine geometry lies centralized around the z-axis and the rotational parts
rotate around the z-axis in the ±θ-direction. The axial length is in the z-
direction, which is also called the axial direction. The machine quantities are
described in the radial direction, which is a direction from the z-axis that is
parallel to the xy-plane.
z

y
x
Figure 2: Axial flux machine in a cylindrical coordinate system.

2.1 Principles of permanent magnet machines

2.1.1 Torque production
The electrodynamic torque in an electric machine can be derived starting
from the Lorentz force [5] [6]. This is the force acting on a charge, q that
moves with velocity, v, in a magnetic field, B:

F q = q(v × B) (1)

If several charges are moving in a conductor, (1) can be rewritten as:

dF = Idl × B (2)

where dF is the force acting on an infinitely small line-segment, dl, carrying

the line-current, I, in the field, B, see Figure 3a.

4
 (a) (b)

Figure 3: Force on current carrying conductors in a magnetic field.

The geometry of a segment of a conductor in an axial flux machine is shown

in Figure 3b. The magnetic field in the air gap, B g , is oriented in the axial
direction and assumed to be constant. Only the conductor part in the radial
direction is of interest, as the force of interest is in the θ-direction, (2) can
therefore be rewritten as:

dFθ = Idr · Bg (3)

Hence, the torque on each conductor segment is

dTθ = rdFθ = rIdr · Bg (4)

where r is the distance from the rotor centrum. In the literature, Idr is
often expressed as A(r)dS, where A(r) is called the electric loading, which is
defined as the sum of the currents in all the conductors spread over a circle
with the radius r.
m1 Nph Iph
A(r) = (5)
πr

where m1 is the number of phases, Nph is the number of conductors in each

phase and Iph is the phase current.

An average air gap flux density, Bg−avg , together with (4) and (5) are then
integrated from the inner radius of the air gap, Ri = Di /2, to the outer
radius of the air gap, Ro = Do /2, with respect to r, [5]. The total average

5
torque from one rotor disc, when starting from the Lorentz force, becomes:
1
Tavg = m1 Nph kw1 Bg−avg Do2 (1 − λ2 )Iph (6)
4

where kw1 is the winding factor of the fundamental and λ is the fraction of
Di
the outer and inner diameter (λ = D o
).

Another common approach to calculate the electromagnetic force on the

rotor, shown in [5] and [7], is the product of the magnetic and electric loading
Bg−avg A(r = Ri ) multiplied by the active surface S. For a double-sided
machine S = 2π(Ro2 − Ri2 ) and the average electromagnetic torque hence:
Tavg−double = 2πBg−avg A(r = Ri )(Ro2 − Ri2 )Ri (7)

which also can be written as:

Tavg−double = 2πBg−avg A(r = Ri )Ro3 (λ − λ3 ) (8)
Starting from (8), taking the first derivative with respect to λ and equating
it to zero, one can show that the maximum torque should be received for
λ = √13 . In practice it has been shown that the maximum torque is given for
λ 6= √13 , [5],[7].

2.1.2 Wave forms

Permanent magnet machines are often divided into permanent magnet syn-
chronous machines (PMSM) and brushless direct current (BLDC) machines
depending on their wave form characteristics. The back-EMF and the cur-
rent wave form of a PMSM, sometimes called brushless AC machine, should
ideally be perfectly sinusoidal. For an ideal BLDC machine, the back-EMF
should be square or trapezoidal shaped and the currents to produce constant
torque should be square shaped.

In Figure 4, the wave forms are explained considering a BLDC machine since
its discretized characteristic may be easier to follow.

The air gap flux density produced by the permanent magnets is assumed to
be quasi-square, as illustrated in Figure 4a. The magnet width corresponds
to 2/3 of a pole pitch.

6
 a) Bm

b) Flux linkage

c) Back-EMF

d) Current

e) Torque

f) Torque phase B

g) Torque phase C

h) Total torque

0 60 120 180 240 300 360

Figure 4: Ideal wave forms of a BLDC machine.

The permanent magnet flux linkage, ψm , is defined as the magnetic flux from
the permanent magnets linking a winding. In Figure 4, a winding covers 60
degrees and the flux linkage can be expressed as:
Z θ+30
ψm (θ) = Bm (θ) dθ (9)
θ−30

The back-EMF wave form can be determined by Faraday’s Law, which states
that the induced voltage equals the time derivative of the flux linkage:
dψm
e= (10)
dt

which gives the square shaped back-EMF in Figure 4c.

DC-current is then switched on and off with the same polarity as the back-
EMF, see Figure 4d. This gives a pulsating but always positive product that
corresponds to the power and hence the torque produced by one phase, which
is shown in Figure 4e.

For a three phase machine, the two remaining phases produce torque that is
shifted 120 and 240 degrees respectively, see Figure 4f-g. Finally the torque

7
produced by the three phases sums up to the total torque produced by the
machine, which is constant and two times the pulsating torque produced by
one phase, see Figure 4h.

The sinusoidal flux linkage for a PMSM can be received by changing the dis-
cretized geometry of a BLDC machine. If the geometry is changed to form a
perfect sinusoidal air gap flux density, the same discretized and concentrated
winding configuration as for the BLDC machine can be used to get sinusoidal
flux linkage. A sinusoidal flux linkage can also be received if the air gap flux
density is kept square shaped but the windings are rearranged in a sinusoidal
distribution. In practice, neither the air gap flux density nor the distribution
of the windings are perfectly sinusoidal but they are combined to get as little
harmonic content as possible for a PMSM.

The torque densities of a PMSM and a BLDC machine of the same sizes
are compared by [8]. The iron losses of the two machines are assumed to be
equal and the torque density limited by the thermal dissipation due to the
copper losses. It was then shown that a BLDC machine could produce 15%
more torque. On the other hand, the higher harmonic content of the square
waves will contribute to the iron losses, and also cause extra copper losses
due to less even current distribution and thereby probably affect the total
losses significantly. The PMSM is expected to offer higher field weakening
capability and less torque ripple [8].

2.1.3 Limiting factors

In the general case, electric machines are limited by thermal constraints. The
torque or the power produced by a machine are thereby closely related to
the ability to cool the machine. Most machines have capacity to manage a
short time overload which may be several times higher than the rated torque
depending on the time period that is considered. The permanent magnets
and the insulation material of the windings are components that are sensitive
to high temperature.

Another limiting factor is the mechanical aspect, a component in the elec-

tromagnetic circuit of a machine must often also provide mechanical support
for the machine structure. Optimization of the electromagnetic circuit can
thereby often result in a theoretic geometry that is of little or no practical use.

The permanent magnets could be demagnetized by a rise in temperature

but also by the induced flux due to the current in the windings. This is of

8
particular interest when the machine is operating in field weakening, since
the induced flux is directed against the flux from the permanent magnets.

2.2 Topologies
Several basic axial flux machine topologies exist, they can be divided into:

• Single-sided

• Double-sided internal rotor double stator

• Double-sided internal stator double rotor

• Multi-disc

It is also possible to subdivide them further depending on: winding con-

figuration, slotted or slotless stator, iron core or coreless stator and rotor
etc.

2.2.1 Single-sided
With the single sided topology it is possible to get a very high ratio between
machine diameter and length, see Figure 5a. However, a significant attracting
axial force between the rotor and the stator will appear and the force can
not be balanced as for the other topologies where the outer rotors or stators
are mechanically connected. The bearings must therefore be dimensioned to
handle a greater axial force.

2.2.2 Double-sided internal rotor double stator

The internal rotor double stator topology is shown in Figure 5b and offers
some extra reliability since it is able to operate with only one stator if the
stators are connected in parallel. Preferably the stators are connected in
series to better balance the axial forces [5]. The stator losses are distributed
between the two stators and it may be easier to cool them by air. On the other
hand, it may be more complex to implement liquid cooling in two stators.
Theoretically, no rotor core is needed for this topology, as the magnetic flux
path is only in the axial direction through the rotor.

9
 (a)
(b)
Stator core

Rotor core

Permanent magnet

Stator winding

Bearing

(c) (d)

Figure 5: Four basic AFM topologies; a) Single-sided b) Double-sided internal

rotor double stator c) Double-sided internal stator double rotor d)
Multi-disc.

2.2.3 Double-sided internal stator double rotor

The axial forces can be balanced for the internal stator double rotor topology
as well. As for the internal rotor double stator, there exists a corresponding
design without a stator core, where the magnetic flux path through the stator

10
is in the axial direction only.

An interesting possibility of this basic topology is that the stator can be

wound toroidally. If so, a stator core is needed since the flux must go in the
θ-direction inside the stator.

2.2.4 Multi-disc
The increase of torque that can be achieved by enlarging the machine diam-
eter can be limited by: axial force supported by bearings, mechanical joints
between discs and shaft and the stiffness of the discs [5]. Instead, several
rotor discs and stators can be added together to increase the torque, which
is also a way of altering the ratio between machine length and diameter for
axial flux machines. This topology could be interesting if there is available
space left in the axial direction, when considering a double-sided machine.

2.3 Comparison of suitable topologies

Rough initial comparisons and calculations showed that neither the single-
sided nor the multi-disc topologies were of interest for the specified ratio
between diameter and length of the geometry. Four double-sided topologies
with balanced axial forces are considered as the most suitable, and thus com-
pared to motivate a final decision. The double-sided topologies to be studied
further are: axial flux internal rotor (AFIR), toroidally wound internal sta-
tor (TORUS), axial flux internal stator (AFIS) and yokeless and segmented
armature (YASA). Schematic figures of their respective flux paths and their
distributions of the main parts can be seen in Figure 6.

All four topologies have some differences and some similarities. The similari-
ties between TORUS, AFIS and YASA may be obvious as both TORUS and
YASA can be said to be variations of AFIS.

The required axial length for the TORUS machine is greater than for the
other topologies due to the extra flux path in the θ-direction, see Figure 6b.
Ideally, the flux in the axial direction is zero in the middle of the stator and
it is thereby possible to introduce a channel for fluid cooling inside the stator.

11
 (a) (b)

S N

N S
N S

S N

(c) (d)

S N
S N

N S N S

Stator core Rotor core Permanent magnet

Figure 6: Schematic figures of the flux paths and distributions of a) AFIR, b)

TORUS and c) AFIS d) YASA in axial direction.

The YASA topology can also be said to be a special case of the AFIS topol-
ogy where the inter-polar stator core is eliminated, see Figure 6d. The first
YASA motors were developed in 2007, using stator segments of powdered iron
material, for the Morgan LIFEcar project1 and presented by [9]. In 2008 a
prototyped machine with laminated stator segment was presented by [2].

Schematic figures of the four topologies, to visualize the composition of the

main parts in the radial direction, using an axial cut are shown in Figure 7.
The protrusion of the end windings are expected to be greater for the AFIR
and AFIS topologies [10], see Figure 6a and 6c. A TORUS machine is ex-
pected to require less space for the end windings as it is toroidally wound
[7], [11]. For the YASA topology it is possible to shape the stator segment
shoes to hide the windings and thereby maximize the diameter of the active
1
Morgan LIghtweight Fuel Efficient Car, a project that aimed at producing the world’s
first hydrogen powered electric sports car.

12
area of the air gap, see Figure 7d.

a) (b) (c) (d)

Stator core Rotor core Permanent magnet Stator winding

Figure 7: Schematic figures of the distributions of a) AFIR, b) TORUS and c)

AFIS d) YASA in the radial direction.

As presented in section 2.1.1, the torque produced by an axial flux machine

is stated to be proportional both to the square, (6), and to the cube, (7), of
the air gap outer diameter. In Figure 8, the torque produced by a machine
with an outer diameter of 260 mm is put to 1 pu. The outer diameter is then
reduced as function of the end winding protrusions in the radial direction;
5 mm of end winding protrusion reduces the outer diameter by 10 mm. The
figure shows the importance of maximizing the outer diameter (minimizing
end winding protrusion) and the impact of the two different relations between
the outer diameter and the torque.

13
 1
2
P~Do
3
P~Do
0.9

0.8
Torque [p.u.]

0.7

0.6

0.5

0.4

0 5 10 15 20 25 30 35
Outer end winding protrusion [mm]

Figure 8: Torque production as function of end winding protrusion.

An attempt to compare the four topologies based on existing analytic designs

was carried out early in this work. This was problematic since it was difficult
to find reasonable values for the end winding protrusion without going into
details, such as pole numbers and winding configurations etc. As four parallel
machine designs were out of scope for this thesis, the expected properties
and consequences for the four topologies are instead, somewhat simplified,
summarized in Table 1.

14
Table 1: Expected distribution of the main parts along radial and axial directions
for TORUS, AFIS, AFIR and YASA.

Radial TORUS AFIS AFIR YASA

End Shorter end wind- Longer protrusion of the end windings The end windings
ings are expected is expected for conventional windings are more or less
windings
for toroidal wind- wound around the stator teeth. hidden behind the
ings wound around stator shoes.
the stator core.
Stator The outer radius of the stator is limited by the end winding The dimension of
protrusion inside a given machine radius. the stator shoes in
the radial direc-
tion is limited by
the space needed
to avoid flux leak-
age to the casing.
Rotor core Generally, a rotor core (or the material in between the PMs for AFIR) has the same
dimensions as the active area or the air gap in the radial direction.

Axial TORUS AFIS AFIR YASA

Stator core A stator core is Ideally, no stator A stator core is The stator con-
needed for the core is needed; needed for the sists of segments
magnetic flux the teeth can be magnetic flux path without a mag-
path in the θ- mounted back to in the θ-direction. netic connection
direction. Ideally back since the in between. The
the magnetic flux magnetic flux needed length of
should be zero in path should be in the segments is
the axial direction the axial direction influenced by the
in the middle of only. amount of copper
the stator core. in the machine.
Stator The lengths of the stator teeth are influenced by the electric The length of the
loading and the amount of copper. The expected shorter end stator shoes is
teeth
windings of the TORUS topology may indicate that it should influenced by the
be possible to make the stator slot and teeth configuration cross-sectional
tighter as well. area of the central
bar inside the
stator segments.
Rotor core A rotor core is needed for the magnetic Ideally, no rotor The same case as
flux in the θ-direction. Its thickness or core is needed for the TORUS
its axial length is depending on both the since there is no and AFIS topolo-
material properties and the magnetic flux path in the gies.
flux density, which together should be θ-direction.
chosen to avoid saturation in the rotor
core.

Air gap All the double-sided topologies include two air gaps, their length is mechanically
restricted by the desired tolerance of vibrations and bending forces.

PM The total axial length of the permanent magnets is mainly depending on the total
axial lengths of the air gaps.

15
2.3.1 Selection of topology
Both the basic AFIR and AFIS topologies are well proven topologies but the
space needed for the end windings in both cases is expected to significantly
limit their torque production. For a machine with a diameter of 260 mm,
10 mm of end winding protrusion in both ends of the active area decreases
the outer air gap diameter by 20 mm. The torque is then reduced to 80-85%
compared to a theoretic machine where the diameter is fully utilized, see
Figure 8.

The TORUS topology is expected to utilize the machine diameter better due
to its more compact end windings. On the other hand an extra flux path in
the θ-direction is introduced and the machine length is less effectively utilized.

Finally the YASA topology is expected to offer both the advantage of high
torque and a compact machine. Using stator segments of soft magnetic
composite (SMC), the stator segment shoes can overlap the windings and
maximize the diameter of the air gap. The iron in the stator can typically
be reduced to 50%, causing a torque density increase of around 20% when
compared to other axial flux machines [9].

When comparing the targets in terms of torque, power and the available space
for this design with the performance of other existing machines, it is obviously
a very demanding combination. Therefore the compact YASA topology is
considered the most interesting topology and chosen for this specific design.

16
3 Analytic models
Various analytic design methods, empirically developed, can be found in
scientific papers. Some of them are used for calculations to obtain an initial
approximate solution, to be further analyzed using finite element analysis
(FEA). Others are more accurate and based on lumped parameter magnetic
circuits, which can take 3D-geometry into account [12].

3.1 Machine characteristics in the dq-coordinate system

The machine characteristics can be derived starting from the equivalent cir-
cuit of a PMSM, described in the dq-coordinate system, as shown in Figure 9.
Amplitude invariant peak transformation is used and as a consequence all
quantities are referred to phase peak values.

- +
+

+ -
+
+ -

Figure 9: Dq-coordinate system equivalent PMSM circuit.

The expressions for the d- and q-components of the phase voltage for the
dynamic model then become:
did
ud = Rid + Ld − ωel Lq iq (11)
dt

diq
uq = Riq + Lq + ωel Ld id + ωel ψm (12)
dt

17
The amplitude of the phase voltage is calculated as:
q
u = u2d + u2q (13)

The time derivatives of the currents, in (11) and (12), are zero if a steady
state is considered. If, in addition, the d- and q-axis inductances are assumed
to be equal (Ld =Lq =L) and the phase resistance,R , is neglected, (11), (12)
and (13) can be combined and rewritten as:
 2  2
u 2 ψm
= iq + + id (14)
ωel L L
If the voltage, u, is considered as the maximal voltage available, (14) can
be visualized as voltage limit circles in the id iq -plane [13]; represented by
the dashed circles in Figure 10. The circles’ center point lies in −ψm /L on
the id -axle and their radii are inversely proportional to the electric angular
frequency, ωel , and given by ωelu L .

The d- and q-components of the current can be used to calculate the ampli-
tude of the phase current the same way as the phase voltage previously:
q
i = i2d + i2q (15)

Again, if the current, i, is considered as the maximal current available, equa-

tion (15) can be visualized as a current limit circle which is centered in origo
in the id iq -plane.

The electrodynamic torque produced by a PMSM can generally be expressed

as:
T = ψm iq + (Ld − Lq )iq id (16)

As Ld and Lq are assumed to be equal (Ld =Lq =L), the reluctance torque
term is zero. The remaining part of the torque is then proportional to the
q-component of the current, iq , and simply expressed as:

T = ψ m iq (17)

Constant torque can be visualized as horizontal lines in the id iq -plane, rep-

resented by the dotted lines plotted in Figure 10.

18
A permissible operating point at a certain speed must lie both inside the
current limit circle and inside the corresponding voltage limit circle in the
id iq -plane. When the center of the voltage limit circles, −ψm /L, lies inside
the current limit circle or on its boundary as in Figure 10, the machine can
in theory be operated at infinite speed. This since there will be a permissible
operating point for any speed.

OP1

OP2

Figure 10: Current and voltage limit circles in the id iq -plane.

The generalized torque speed characteristic of a machine with field weak-

ening possibilities is shown in Figure 11. Starting from zero speed in the
constant torque region, the machine is fed with iq = in and id = 0, which
gives rated torque up to rated speed. The rated torque and rated speed op-
erating point corresponds to OP1 in Figure 10.

To further increase the speed and to get the maximal possible torque at this
speed, a negative d-component current, id , must be applied to follow the
decreasing voltage limit circle. Hence, iq must be reduced to stay in the
current limit circle. OP2 in Figure 10 shows an example of a field weakening
operation point for maximum torque.

The region beyond rated speed is often called constant power region but the
power is changing since the increase of speed and the decrease of torque will
not result in a constant power product.

P = Tω (18)

19
Voltage phasor diagrams are plotted for three different operating points in
Figure 12. The phase displacements between the voltages and the currents
are denoted by ϕ. For an infinite speed machine, it can be shown that the
maximum power factor, |cos(ϕ)|, below rated speed is √12 = 0.71. As ϕ de-
creases for speeds beyond rated speed, the power factor starts to increase in
field weakening operation which can be seen in Figure 12c.

rated speed

constant torque
region
Torque

constant power
region

Speed
Figure 11: Generalized torque speed characteristics.

(a) (b) (c)

Figure 12: Voltage phasor diagrams: a) below rated speed, b) at rated speed, c)
field weakening operation.

20
3.2 Voltage limit
The machine characteristics are evaluated for the phase voltage that corre-
sponds to the minimum operational DC-link voltage of 500 V. For a three
phase inverter, it can be shown that the maximum line-to-line rms voltage
of the fundamental frequency can be written as :
√
3
ULL−1−rms = √ ma UDC (19)
2 2

where ma is the amplitude modulation ratio and the linear region is defined
as ma ≤ 1 [14]. By adding the 3rd harmonic to the phase voltage the
linear range of modulation can be increased by 15%. The expression for the
maximum phase peak voltage then becomes:
UDC
Ûph−1 = ma · 1.15 (20)
2

3.3 Pole pairs and stator segment combinations

The total number of poles, 2p, should be as close as possible to the number of
stator segments, Ns , to maximize the magnetic flux linkage. Equal numbers
can’t be used if the motor is to be able to produce torque in all rotor positions.
The feasible slot and pole number combinations are usually described by (21),

Ns = 2p ± 1 or 2p ± 2 (21)

where Ns must be divisible by 3 for a three phase machine. The phase-shift

angle, θps in electrical degrees, must be ±360k + 120, where k is a positive
integer [15],
 
Ns 180 2pβp Ns 180 360p
θps = βs = = (22)
3 βp Ns 3 βp 3

where βp is the magnet pole-pitch angle and βs = 2pβp /Ns is the slot-pitch
angle.

All possible combinations, up to p=16, derived from (21) and (22) are listed
in Table 2. It can be noted in Table 2 that no combinations where p is
divisible by 3 exist, which is a consequence of (21). Ns and p for a given
combination of Ns /p can be multiplied by a positive integer to get a new

21
 Table 2: Feasible pole and slot number combinations for a 3-phase machine.

Number of pole pairs, p Ns = 2p ± 1 Ns = 2p ± 2

1 3 -
2 3 6
4 9 6
5 9 12
7 15 12
8 15 18
10 21 18
11 21 24
13 27 24
14 27 30
16 33 30
. . .
.. .. ..

feasible combination.

Using a Ns = 2p ± 1 combination results in an odd number of coils per phase

which will limit the possibilities of series and parallel connections of the coils.
It will also cause unbalanced axial forces between the rotors and the stator.
All the Ns = 2p ± 2 combinations give an even number of coils per phase.
Hence, the coils can be connected both in series and in parallel configura-
tions. The attracting forces between a rotor and the stator are in balance,
since all coils consist of pairs which are displaced by 180 mechanical degrees.

3.4 Winding factors

The winding factor is defined as the distribution factor multiplied by the
pitch factor which can be seen in (23).

kwn = kdn kpn (23)

Both the distribution factor and the pitch factor depend on the type of
winding that is used. The following equations; (24)-(27) refer to double-
layer, non-overlap and slotted/segmented iron-cored windings [5],
nπ
sin( 2m )
kdn = nπ
1
(24)
zsin( 2m 1z
)

22
 where n is the harmonic number, m1 the number of phases and z the number
of coils in a group. The number of coils in a group can be calculated as:
Ns
z= (25)
m1 F

where F is the greatest common divisor (GCD) of the number of poles, 2p,
and the number of stator segments, Ns .

F = GCD(2p, Ns ) (26)

The pitch factor can be calculated according to (27),

nπp θmn
kpn = sin( ) = sin( ) (27)
Ns 2

where θmn corresponds to the coil span angle.

The winding factors for the feasible combinations of Ns /p up to 24/13 are

shown in Table 3.

Table 3: Winding factors for the 1st to 13th harmonic for feasible pole pair and
stator segment combinations up to p = 13 for a 3-phase YASA machine.

Ns /p Harmonic number
1 2 3 4 5 6 7 8 9 10 11 12 13
6/2 0.866 0.866 0 -0.866 -0.866 0 0.866 0.866 0 -0.866 -0.866 0 0.866
6/4 0.866 -0.866 0 0.866 -0.866 0 0.8660 -0.866 0 0.866 -0.866 0 0.866
12/5 0.933 0.433 -0.500 -0.433 0.067 0 -0.067 0.433 0.500 -0.433 -0.933 0 0.933
12/7 0.933 -0.433 -0.500 0.433 0.067 0 -0.067 -0.433 0.500 0.433 -0.933 0 0.933
18/8 0.945 0.289 -0.577 -0.289 0.140 0 0.061 0.289 0 -0.289 -0.061 0 -0.140
18/10 0.945 -0.289 -0.577 0.289 0.140 0 0.061 -0.289 0 0.289 -0.061 0 -0.140
24/11 0.950 0.217 -0.604 -0.217 0.163 0 0.096 0.217 -0.104 -0.217 0.017 0 -0.017
24/13 0.950 -0.217 -0.604 0.217 0.163 0 0.096 -0.217 -0.104 0.217 0.017 0 -0.017

23
3.5 Torque calculations
More than one way of analytically calculating the torque production of an
axial flux machine exist, previously presented in (6) and (7). As discussed by
[6], the widespread acceptance for the Lorentz force method seems surpris-
ing, since it should be valid for a non-slotted machine but not for a slotted
machine. In the slotted machine, the main part of the magnetic flux goes
trough the stator core and not trough the conductors at all. A variation2
of (7), which considers the product between the air gap areas, the magnetic
loading and the electric loading, is used in this thesis:
Do + Di Do − D i
T = πBQ Di (28)
2 2
where B is the magnetic loading and Q is the electric loading expressed as:
2Nph Iph
Q = kw1 (29)
πDi
It was later verified that this torque expression gives a good correlation be-
tween the analytic and FEA results for this particular machine topology and
size of geometry.

3.6 Magnetic flux and permanent magnet length

A very simple approach is to model the magnetic flux distribution as a square
shaped distribution; see the dashed line in Figure 13, where the permanent
magnets are assumed to cover 120 electrical degrees or 2/3 of the pole pitch.
Using Fourier analysis, the peak value of the corresponding fundamental, B̂1 ,
can be expressed as:
√
4 3
B̂1 = Bm (30)
2π
If the reluctances of the iron parts are neglected, the required length of the
permanent magnets can be calculated as:
µr,pm lg
lpm = Br,pm
(31)
Bm
−1

where lg is the axial length of the air gap, µr,pm is the relative permeability
of the permanent magnets and Br,pm is the residual flux density of the per-
manent magnets.

2
From discussions with Martin West at Volvo Technology in January 2011.

24
 Figure 13: Air gap flux density.

As can be seen in Equation (31), in order to design a very compact machine

with high magnetic loading, it is important to keep the air gap as small as
possible; to minimize the air gap length itself and to minimize the length of
the permanent magnets.

3.7 Rotor core length

If both the leakage flux and the induced flux are neglected, the total air gap
flux per pole can be calculated as the product of the flux density and the
area of a permanent magnet:

2 R2 − Ri2
φg = Bpm Apm = Bm π o (32)
3 2p
where a magnet is assumed to cover 2/3 of a pole area. As illustrated in
Figure 14, the magnetic flux density in the θ-direction trough the rotor back
can be expressed as:

φr = Ar Br = lr (Ro − Ri )Br (33)

where Br is the magnetic flux density in the rotor back.

Hence 2φr = φg , as half of the flux per pole goes trough the rotor back in
the θ-direction and the other half in the −θ-direction, which can be seen in
Figure 6d. The required rotor core length can be calculated as:
Bm π
lr = (Ri + Ro ) (34)
Br 6p

25
 Rotor core

Permanent magnet

Figure 14: Schematic figure of the cross-sectional area, Ar , of the rotor back.

3.8 Stator segment

A schematic figure of a stator segment is shown in Figure 15. It can be
divided into three parts: two shoes and one central bar. The shoes are wider
in both the radial- and the θ-direction to overlap the windings and maximize
the active surface. This results in a three dimensional flux path (axial-,
radial- and θ-direction), therefore the shoes can not be made of laminated
steel. In the central bar, on the other hand, the flux path is only in the
axial direction and it is possible to use laminated steel or soft magnetic
composite (SMC). SMC has lower permeability but also lower losses at higher
frequencies compared to those of laminations. Hence, the choice of material
for the central bar can be motivated by the electric frequency. However,
using two different materials will add extra complexity due to the non-ideal
contact surfaces. Entire SMC stator segments are therefore considered.

Shoes

Central bar

Figure 15: Schematic figure of a stator segment.

The cross-sectional area of the central bar, As illustrated in Figure 16, re-
quired for the desired magnetic flux density, Bs , can be expressed as:

φs 2 Bm Ro2 − Ri2
As = = π (35)
Bs 3 Bs 2p

26
Figure 16: Schematic figure of the central bar and the winding behind the stator
segment shoe.

The relation between the width of the copper winding, Ww , around the cen-
tral bar and the area of the winding, Aw−a , seen in the axial direction in
Figure 16 can be approximated as:
 
πDo πDi
Aw−a = 2(Ro − Ri ) + ( − 2Ww ) + ( − 2Ww ) Ww (36)
2p 2p
Finally, the cross-sectional area of the central bar, As , and the area of the
winding, Aw−a , must be chosen to fit into the area that is available for each
stator segment:

air gap area π(Ro2 − Ri2 )

Astatorsegment = = = As + Aw−a (37)
Ns Ns

3.9 Inductances
A very simplified way of calculating the inductance per phase is to trans-
form the machine geometry to a an equivalent C-core, see Figure 17. The
inductance of a C-core can easily be approximated if the flux leakage and the
fringing effects are neglected, i.e. lg−C ≪ Ag−C , and if also the permeabil-
ity is assumed to be much greater inside the core than in the air gap. The
approximated inductance is then:
µ0 n2 Ag−C
L= (38)
lg−C

which is shown by [16] for example.

When the number of stator segments is chosen as Ns = 2p ± 2 according to

section 3.3, the number of stator segments per phase is always an even num-
ber. As mentioned previously, the stator segments in each phase can then be
divided into pairs, displaced 180◦ mech. The coils around those pairs should
be connected in series to make sure that the attracting forces between the
rotors and the stator are balanced (two axial forces of the same magnitude

27
 (a) (b)

Figure 17: C-core transformation to calculate the inductance.

displaced 180◦ mech.). To further simplify the situation, all coils are assumed
to be connected in series.

A stator segment pair is shown in Figure 17a, drawn side by side to reduce
the figure. The corresponding lg−C can be calculated as:

lg−C = 4(lpm + lg ) (39)

and the corresponding air gap area, Ag−C , can be approximated as:

πRo2 π
Ag−C = − Ri2 tan( ) (40)
Ns Ns

which is the area of the stator segment shoe that faces the air gap.
The fact that a three phase machine is considered together with (38)-(40)
gives the expression for the phase inductance:

µ0 (πR02 − Ns Ri2 tan( Nπs ))n2

Ls = (41)
6(lg + lpm )

where n is the number of turns around each stator segment.

28
3.10 Losses and efficiency
The electromagnetic losses in the machine can be subdivided into copper and
iron losses. In this project, the copper and the iron losses are calculated in the
FEA part. The iron losses are calculated in the time domain according to the
method suggested by [17]. This should give a more accurate prediction of the
hysteresis losses than the classic method that is calculated in the frequency
domain. The classic analytic methods are briefly described in this section to
explain the mechanisms that cause the losses.

3.10.1 Copper losses

The winding resistance for direct current can be calculated using (42):

n · lav
Rs−DC = (42)
ap · A · σcu

where n is the number of turns, ap is the number of parallel conductors within

each turn, lav is the average length of a turn, A is the cross-sectional area of
the conductor and σcu is the conductivity of copper. A constant, K, can be
introduced to compensate for the temperature and the content of harmonics
in the current:

Pcu = m1 · i2s · Rs = m1 · i2s · K · Rs−DC (43)

where m1 represents the number of phases. Various analytical methods that
considers temperature dependence and the skin effect due to the harmonics
were investigated in [5] and [18].

Eddy currents are induced in the conductors since these are exposed to an
alternating magnetic field in the air gap. This can generally be ignored for
slotted AFPM but should be considered for slotless AFPM [5].

3.10.2 Iron losses

Accurate computation of the iron or core losses is a complicated task. The
iron losses are subdivided into three parts [18]:

• Hysteresis losses

• Eddy current losses

• Excess losses

29
The hysteresis losses are caused by the energy needed to flip the magnetic
dipoles within the soft magnetic material [16]. Material with high perme-
ability and low coercivity should be chosen to minimize the hysteresis losses.

Eddy currents are induced in the iron parts due to the alternating magnetic
flux from the stator windings and the field from the rotating permanent
magnets. The induced currents can be limited by lamination of the core.
The core is therefore constructed of thin sheets separated by a material with
lower conductivity. The eddy current losses increase with frequency, which
is related to the current harmonics and the speed. A few simple methods of
predicting the losses per unit volume are shown by [18]. An example can be
seen in (44),

σF e d 2
PF e = khys B̂ 2 f + π 2 (B̂f )2 + 8.67 · kex (B̂f )1.5 (44)
6
where khys is the hysteresis loss coefficient, kex is the excess loss coefficient,
d is the thickness of the steel sheets and σF e is the electric conductivity of
the ferromagnetic material. The loss coefficients can be determined from loss
curves as a function of frequency, which may be provided by the manufac-
turers.

3.10.3 Efficiency
The efficiency of the electric machine is defined as:
Pout Pout
η= = (45)
Pin Pout + Pd
where Pout is the mechanical output power, Pin is the electrical input power
and Pd is the total power losses in the machine.

30
4 Analytic design
The intention of this section is to describe the analytic part of the design
process. As explained before, this is not a complete analytic design procedure
as the analytic design and FEA are made alternately in this thesis. Therefore,
values from the FEA, in section 5, are sometimes used as input to the analytic
design part in this section.

4.1 Number of pole pairs and stator segments

The number of pole pairs, p, should be chosen considering several aspects.
High pole numbers give:

- Smoother torque

- Higher winding factor/torque

- Better magnetic balance

- Possible reduction of the amount of iron for the same magnetic loading,
shown in (34) and (35). The electric loading and the torque may then
be increased.

On the other hand the pole pair number is limited by:

- Increased switching frequency and switching losses

- Increased iron losses

- Flux leakage

- Practical/mechanical aspects (constructions comprised by many small

parts)

The mechanical system is schematically visualized in Figure 18. The funda-

mental of the electric frequency, fel , at a given vehicle speed can be calculated
using (46).
ωw
fel = · gR · p (46)
2π
where ωw is the angular frequency of the wheel, gR is the gear ratio of the
planetary gear and p is the pole pair number. Using the wheel radius of
0.492 m together with (46), the electric frequency as function of vehicle speed
is plotted in Figure 19 for four different gear ratios and four different pole

31
 G
W
A H
E
F E
A
M E
R
L

Figure 18: Schematic figure of the mechanical system.

pair and stator segment configurations.

During this work it was indicated, by people working with the planetary gear,
that a one-stage gear ratio is limited to 6-7. The loss cross over frequency
is assumed to be around 1kHz, as the losses for laminated core materials
increase faster than the losses for SMC materials with respect to the electric
frequency. It can then be seen in Figure 19 that use of SMC cannot be
motivated for reduction of the iron losses for any combination.

Ns=12 p=5 Ns=12 p=7

500 700
gR=10 gR=10
600
400
Electric frequency [Hz]

Electric frequency [Hz]

gR=8
500 gR=8
300
gR=6
400 gR=6
gR=5
200 gR=5
300

100
200

0 100
40 50 60 70 80 90 100 40 50 60 70 80 90 100
Vehicle speed [km/h] Vehicle speed [km/h]

Ns=18 p=8 Ns=18 p=10

800 1000
gR=10 gR=10
700
800
Electric frequency [Hz]

Electric frequency [Hz]

600 gR=8
gR=8

500 600
gR=6
gR=6 gR=5
400 400
gR=5
300
200
200

100 0
40 50 60 70 80 90 100 40 50 60 70 80 90 100
Vehicle speed [km/h] Vehicle speed [km/h]

Figure 19: Electric frequency for four different gear ratios (5, 6, 8, 10) and pos-
sible combinations of pole pairs and stator segments.

32
As described in section 3.3, the number of stator segments is directly influ-
enced by the number of poles. A Ns = 2p ± 2 configuration is preferable due
to its balanced characteristics, the possible choices are previously shown in
Table 2.

The outer width, Wo , and the inner width, Wi of a permanent magnet and a
stator segment shoe according to Figure 20 can be calculated using (47) and
(48).

Wo

Wi

Figure 20: Outer and inner width of a permanent magnet or a stator segment
shoe.

πD
Ws−segment = (47)
Ns
where D is the diameter and Ns is the number of stator segments.

πD 120
Wpm = (48)
p 360
where p is the number of pole pairs and a magnet is assumed to cover 120
electrical degrees.

The resulting widths, Wo and Wi , for six different Ns /p combinations are

calculated and presented in Table 4. Some extra space between the stator
segments is of course needed in order to prevent flux leakage between the
segments.

Two examples of YASA machine designs are presented in [2] and [9]. An
Ns /p configuration of 12/5 is used in both cases. Their geometries are slightly
smaller but in the same order of magnitude (machine diameters of 186 mm
and 204 mm compared to 260 mm). The Ns /p=12/5 configuration is used
as a starting point for this design. The lower numbers give very unfavorable
winding factors.

33
 Table 4: Magnet and stator segment widths for different Ns /p combinations.

Magnet width Stator segment

[mm] width [mm]
Ns /p Wi Wo Wi Wo
12/5 34.9 54.5 43.6 68.0
12/7 25.0 38.9 43.6 68.0
18/8 21.8 34.0 29.0 45.4
18/10 17.4 27.2 29.0 45.4
24/11 15.8 24.8 21.8 34.0
24/13 13.4 21.0 21.8 34.0

4.1.1 Winding configuration and inductance

Using the coil span angle in (27), the phase displacement of the induced
fundamental back-EMF in the coils can be expressed as:
2πp
θemf −coil = ncoil · (49)
Ns
where ncoil represents the index of the coil.

The phase orientation of the fundamental of the induced back-EMF in the

twelve coils is plotted in Figure 21. It can be seen that the two coils that
form a pair, displaced 180 mechanical degrees, also are shifted 180 electrical
degrees. As mentioned in section 3.3, the pairs should be connected in series
to make sure that the axial forces are balanced. The windings are arranged
in phase groups that consist of coils with the smallest possible phase dis-
placement in between. They are connected in series, the circuit diagram is
shown in Figure 22

The phase inductance that is calculated according to (41) becomes 1.47µH/n2 .

34
 10
5 3

12 8

7 1

2 6

9 11
4

Figure 21: Induced back-EMF phase displacement between the coils.

1 2 7 8
Phase A

3 4 9 10
Phase B

5 6 11 12
Phase C

Figure 22: Circuit diagram of the winding configuration.

35
4.2 Optimization of the geometry for maximal torque
As mentioned previously, the theoretic ratio between outer and inner diam-
eter for maximum torque is given by λ = √13 = 0.58. Optimal ratios for
different topologies and machine sizes are suggested by [2], [7], [9], [10] and
[19]. The highest torque density can be reached for ratios between 0.6 and
0.7 according to one approach in [7]. For the YASA machine designs in [2]
and [9], ratios of 0.58 and 0.65 are used respectively. A ratio of 0.64, which is
included in the range above, is therefore assumed for the design in this thesis.

The product of magnetic and electric loading should be maximized in order

to obtain the highest possible torque according to (28). The idea is to match
the copper and iron quantities to get a high utilization of both materials,
thereby avoiding a bottleneck situation when one of them is saturated too
long before the other one is fully utilized.

A maximum flux density of 1.7 T and a maximum current density of 7 A/mm2

rms are assumed to be reasonable values for high utilization of the materials.
The magnetic flux density is chosen to stay within the saturation limit of
the iron, which is about 2 T. The current density has been chosen based on
values for existing machines, but may be a bit conservative when comparing
with the YASA machine design in [9]. The air gap length is kept constant at
1 mm, which is assumed to be a practical length to give mechanical clearance.
The axial length of the stator segment shoe is assumed to be 5 mm, while all
the other geometric variables are varied in order to find the maximal torque
within the given volume.

The peak flux density of the fundamental in the air gap, B̂1 in (30), is used
as an input to an optimization script. The value of B̂1 together with the
assumed maximum flux density of 1.7 T in the iron are then used to deter-
mine the geometry of the magnetic circuit according to previously presented
relations.

The space that is left (Aw−a in Figure 16 together with the length of the
central bar) when the geometry of the stator segments is determined, cor-
responds to the total cross-sectional area of the winding, Aw . This area is
filled with copper using an assumed fill factor of 70%; copper fill factors up
to 78% for pre-pressed windings in segmented fractional slot machines are
reported according to [20].
Acu
Kcu−f ill = (50)
Aw

36
The electric loading can be calculated using the copper area, Acu in (50),
together with the assumed maximum current density of 7 A/mm2 . The re-
sulting torque is calculated based on magnetic loading and electric loading
using (28).

Lower values of magnetic loading give a rotor back thickness that is too small
to make sense. If the calculated rotor back thickness is below 9.5 mm it is
simply forced to 9.5 mm which is assumed to be a reasonable starting value.

The magnetic and electric loading for different geometries is shown in Fig-
ure 23 when varying the air gap flux density while the core magnetic flux
end current densities are kept constant.

0.8

0.6
B1rms [T]

0.4

0.2

0
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
B̂1 [T]
60

50
J1rms [kA/m]

40

30

20

10
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
B̂1 [T]

Figure 23: Magnetic and electric loading when the geometry is changed to keep
the current density in the copper and magnetic flux density in the
iron constant while the air gap flux density, B̂1 , is varied.

The analytically calculated torque is presented in Figure 24, where the lower
torque for low values of B̂1 corresponds to geometries that give a mismatch
between electric and magnetic loading in favor of the electric loading. The
maximum for the analytically calculated torque is received for B̂1 ≈ 0.75 T.

37
 80

70

60
Torque [Nm]

50

40

30

20
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
B̂1 [T]

Figure 24: Analytically calculated torque when the geometry is changed to keep
the current density in the copper and magnetic flux density in the
iron constant while the air gap flux density, B̂1 , is varied.

4.3 Investigation of torque speed characteristics

This subsection describes how the machine torque speed characteristics can
be changed based on the number of turns for a given maximum voltage. All
quantities needed in order to find the torque speed characteristics accord-
ing to section 3 are expressed as functions of the number of turns, n. The
machine geometry (the geometry that corresponds to B̂1 =0.85 T in section
5.2) is fixed as it has been optimized to give the maximum torque for the
assumed current and flux densities of 7 A/mm2 and 1.7 T.

The maximum phase peak voltage is calculated to 273 V according to (20),

where an amplitude modulation ratio, ma , of 95% is used to stay within the
linear region of the inverter with reasonable margin.

The total current in a winding around a stator segment can be calculated as:

Itot−winding = Jcu · Acu = Jcu · Aw · Kcu−f ill (51)

where Jcu is the current density in the conductors (7 A/mm2 ), Acu is the total
conductor cross-section area and Aw is the cross-section area of the winding.

38
The maximum phase peak current is then:
√
ˆ 2
Iph−max = Itot−winding (52)
n

Hence, with the current density fixed:

1
Iˆph−max ∼ (53)
n

The relation between the phase flux linkage due to the permanent magnets
(open circuit) and the number of turns around each segment is:
ψm ∼ n (54)

Considering the C-core analogy in section 3.9, the relation between the in-
ductance and the number of turns is:

L ∼ n2 (55)

The relation between phase resistance and number of turns is:

R ∼ n2 (56)
which may cause confusion but both the cross-sectional area of the conductor
and the total length of the conductor in each coil are affected by the number
of turns, see (42). This to keep the current density, Jcu , the total current,
Itot−winding , and the geometry constant while the number of turns are varied.

Values from the FEA part in section 5 are used to analytically calculate the
torque speed characteristics of the machine. The values are summarized in
Table 5.
Table 5: Values that are determined in the FEA and used for analytic calculations
of torque speed characteristics.

Ûph−1 273.13 V
Itot−winding 1.60 kA
ψm 0.0054 W b/n
Ld 2.56 µH/n2
Lq 2.14 µH/n2
R 45.9 µΩ/n2

39
In Figure 25, the torque as a function of speed is plotted for n=20,n=30,n=40
and n=50. As expected, the constant torque region is decreased for higher
number of turns and the slope of the torque becomes steeper in the constant
power region.

100

90

80

70 n=20

60
Torque [Nm]

n=30

50 n=40
n=50
40

30

20

10

0
0 1000 2000 3000 4000 5000 6000 7000 8000
Speed [rpm]

Figure 25: Torque as function of speed for n=20,n=30,n=40 and n=50. The
dotted lines show the characteristics for a DC-link voltage of 700 V
instead of 500 V.

In Figure 26, the output power as function of speed is shown for the four
variations of n. A machine with a lower number of turns is able to produce
more output power, since it enters its field weakening region at a higher speed
compared to a machine with higher a number of turns.

40
 70

60

50
n=20
Output power [kW]

40
n=30
30
n=40
n=50
20

10

0
0 1000 2000 3000 4000 5000 6000 7000 8000
Speed [rpm]

Figure 26: Output power as function of speed for n=20,n=30,n=40 and n=50.
The dotted lines show the characteristics for a DC-link voltage of
700 V instead of 500 V.

The corresponding power factors are plotted as function of speed in Fig-

ure 27. One can see that greater field weakening performance and output
power production are received at the cost of a poor power factor.

Below base speed, the power factor is approximately 0.75. This indicates
that the machine geometry is close to a theoretical infinite speed machine,
which has a theoretic maximum power factor of √12 = 0.71 in the constant
torque region. This implies that the value of ψLm is close to the maximum
current, i, according to the theory in section 3.

One can also observe the equal power factors in their constant torque re-
gion, explained by the fact that both ψLm and i are inversely proportional to
the number of turns, n. Therefore, the center of the voltage limit circles in
relation to the current limit circle, seen in Figure 10, cannot be moved by
changing the number of turns, they are scaled together.

41
 0.95

n=50
0.9 n=40
n=30
Power factor

0.85 n=20

0.8

0.75

0.7
0 1000 2000 3000 4000 5000 6000 7000 8000
Speed [rpm]

Figure 27: Power factor as function of speed for n=20,n=30,n=40 and n=50.
The dotted lines show the characteristics for a DC-link voltage of
700 V instead of 500 V.

The nominal phase currents for the different selections of the number of turns
are summarized in Table 6. In section 5.4 the inductances in positive and
negative d- and q-direction are calculated per number of turns in square. The
phase inductance is plotted as a function of the number of turns in Figure 28.

Table 6: Nominal phase rms current for n=20,n=30,n=40 and n=50.

Number of turns Nominal phase rms current [A]

20 80.2
30 53.5
40 40.1
50 32.1

42
 7

L
+q
6
L
−q
L
+d
5
L
Phase inductance [mH]

−d

0
20 25 30 35 40 45 50
Number of turns around each stator segemnt

Figure 28: Phase inductance at 7 A/mm2 as a function of the number of turns

around each stator segment, n.

As no requirements in terms of rated speed, rated torque etc. were stated

for the machine, the choice of the number of turns is based on an indicated
maximal speed of around 3000 rpm. When comparing n=20 and n=30,
the output power and the torque at 3000 rpm can be increased by 6.7%
(with n=20) at the cost of a 50% higher current. The inductance is about
two times higher for n=30 compared to n=20 according to Figure 28. The
current ripple is thereby two times lower using the same switching frequency
and DC-link voltage. Hence, the ripple is lower for n=30 also in relation to
the amplitude of the fundamental of the current. Therefore n=30 is chosen,
this results in a nominal current of 53.5 A and a nominal speed of around
2400 rpm.

43
5 Finite element analysis
The finite element analysis is made in Ansoft Maxwell 3D version 14, there-
fore some terminology from this software is used in the report.

5.1 Model and implementation

5.1.1 Symmetry planes
The model of the whole machine is shown in Figure 29a. Two symmetry
planes exist in the geometry, the xy-plane and the xz-plane.

The machine is first split and reduced along the xy-plane using an ”Even
(flux normal)” symmetry boundary. This boundary condition is also called
Neumann’s condition and forces the flux lines to be normal to the symmetry
surface, see Figure 29b. It is equivalent to having an external material with
infinite permeability just outside the surface [21].

The geometry is then split along the xz-plane and further reduced using
”Master” and ”Slave” boundaries, which is also called periodic boundary
condition and can be seen in Figure 29c. It is split in two parts since 2 is
the greatest common divisor of Ns = 12 and 2p = 10. The computational
domain is now reduced to a quarter of the initial machine.

In Maxwell 3D, most of the machine properties can be calculated using the
quarter machine while at least the half machine has to be used for some prop-
erties. The model of the whole machine has been used to verify the results
from the reduced models for a limited number of simulations.

5.1.2 Windings
The windings or the coils are modeled as solid objects to simplify the geom-
etry. A coil terminal is introduced as a surface, the cross-sectional area of
the solid winding, of which the normal is in the current direction. As the
real coils consist of several turns of a conductor, each turn carrying the same
current, the current density is expected to be more or less evenly distributed.
The windings are therefore modeled as ”Stranded” conductors, which forces
the current to be evenly distributed in the solid object.

44
 (a)

(b) (c)

Figure 29: Reduction of the 3D-model using symmetry planes.

5.1.3 Core material

Höganäs AB3 kindly supplied data of suitable SMC materials. Four materials
were suggested for the application and their BH-curves can be seen in Fig-
ure 30. Somaloy 1000 3P was chosen due to its preferable BH-characteristics
for high magnetic flux densities. Its loss characteristics, shown in Figure 31,
were slightly less favorable than the other materials’ (not shown here) but
thanks to its BH-characteristic, it was regarded as suitable for maximizing
3
http://www.hoganas.com/

45
the torque and power densities of the machine.

2.2

2
Somaloy 700 3P
1.8 Somaloy 700HR 3P
Somaloy 700 5P
1.6 Somaloy 1000 3P

1.4
B [T]

1.2

0.8

0.6

0 2 4 6 8 10
H [A/m] x 10
4

Figure 30: Four different SMC materials suggested by Höganäs (Source: Höganäs
”Data adjusted for use in Finite Element modeling”).

The BH-curve for Somaloy 1000 3P in Figure 30 together with its loss char-
acteristics, shown in Figure 31, are extrapolated and used to simulate the
non-linear behavior of the SMC material in Maxwell 3D. The SMC material
is used both in the stator segments and in the rotor cores.

46
 350

300

250
Core loss [W/kg]

200

150

100

50

0
0 0.5 1 1.5
Induction [T]

Figure 31: Core loss for Somaloy 1000 3P, from bottom: f=50, 60, 100, 200,
300, 400, 500, 600, 700, 800, 900 and 1000 Hz (Source: Höganäs
”Datasheet of Somaloy 1000 3P (0.3% 3P Lube), 800 MPa”).

5.1.4 Permanent magnets

Sintered Neodymium-Iron-Boron (NdFeB) magnets called N35EH were cho-
sen to model the permanent magnets. They offer high magnetic output at a
reasonable cost, and were recently used in another machine design by Volvo
Technology. A simple linear model is used in the FEA-model, where data
from [22] is used.

The properties of the permanent magnets change significantly depending on

temperature. A temperature of 120◦ C has been assumed, which yields a
relative permeability, µr,pm , of 1.04 and a coercivity, Hc , of 795.57 kA/m
according to the datasheet.

If the temperature is assumed too low, there is a risk that the magnetic
loading and thereby the torque is overestimated. On the other hand, if it
is assumed too high, there is a risk that the required amount of iron is
underestimated.

47
5.1.5 Losses
Copper
As the coil terminal is modeled as a surface that covers the whole cross-
sectional area of the winding, the conductivity for the winding object is
scaled using the copper fill factor:

σw = σcu · Kcu−f ill (57)

The conductivity in (57) together with the total current in (51) are used to
get the correct resistive copper losses when stranded conductors are used in
3D. The average length of a turn is determined by the geometry of the wind-
ing object, which gives the DC-resistance according to (42). Thereby the
expected frequency dependence of the copper losses is not taken into account
in this model.

Iron
Core losses are often calculated in the frequency domain, as shown in section
3.10.2. A dynamic model, in time domain, using the time derivatives of
the flux density together with loss data for the material is used to calculate
the core losses in Maxwell 3D 4 . In this way, both the frequency and the
magnetic flux density dependences together with the non-linear BH-curve
are taken into account in the model.

5.2 Optimization of the geometry for maximal torque

The same procedure as in section 4.2 was carried out in the FEA part to
find the geometry that gives the maximum torque. The torque for five FEA
variations can be seen in Figure 32, together with the previous analytically
calculated torque. As can be seen, higher magnetic loading which is achieved
at the cost of electric loading is required in order to obtain the maximal
torque.

The difference can be explained as a consequence of the simplifications made

to analytically calculate the magnetic loading. The permeability of the iron
was set at infinity and the leakage was neglected. The magnetic loading is
thereby overestimated.

4
Accordning to ”Maxwell 3D user’s guide”, provided by Ansoft.

48
 90

80 Analytic
FEA

70
Torque [Nm]

60

50

40

30

20
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
B̂1 [T]

Figure 32: Torque from FEA and analytic calculations when the geometry is
changed to keep the current density in the copper and magnetic flux
density in the iron constant while the air gap flux density, B̂1 , is
varied.

As the idea was to keep the maximum magnetic flux density in the iron parts
constant (1.7 T) while the geometry was varied, the peak values of the flux
densities in the middle of the stator segments were measured in the FEA
model. The flux densities at open circuit tests are plotted as a dashed line in
Figure 33, where the magnetic flux density is about 1.4 T which is below the
target. Again, this is a consequence of the simplifications that were made in
the analytic models. It can also be seen that it is almost constant for the
two lowest values of B̂1 while it starts to increase for higher values of B̂1 .
This may be caused by the model used to calculate the required length of
the permanent magnets, (31), which approaches infinity when Bm =Br−pm .
It is then probable that the required length of the permanent magnets is
overestimated for the higher values of magnetic flux density.

In Figure 33, the flux density is also plotted when the machine is fed with
rated current. As can be seen in the figure, the induced flux has a significant
impact on the magnitude of the total flux. Actually, it is the total flux that
is of interest when the utilization of the iron is evaluated. Ideally, the ge-
ometry for constant flux density of the total flux should therefore be found

49
iteratively. This was not possible due to the restricted time for this mas-
ter thesis, where the geometry for B̂1 =0.85 T is considered as the optimum
design, as it gives the maximum torque according to Figure 32 and a flux
density close to 1.7 T according to Figure 33.

One could discuss if the amount of iron should be reduced a bit for the ge-
ometries where B̂1 >0.85 T, as they are not ”fully utilized” to 1.7 T. The
electric loading could then be slightly increased and the maximum of the
obtained torque in Figure 32 further moved towards higher values of B̂1 .

1.8

1.6

1.4

1.2 7 A/mm2
0 A/mm2
[T]

0.8

0.6

0.4

0.2

0
0.75 0.8 0.85 0.9 0.95
B̂1 [T]

Figure 33: The flux density in the middle of the central bar when the geometry
is changed to keep the current density in the copper and magnetic
flux density in the iron constant while the air gap flux density, B̂1 , is
varied.

Figure 34 shows the analytically calculated geometric variations that are used
in the FEA model, in the optimization procedure.

50
 0.04

0.035

0.03

Rotor back length

0.025
Central bar length
Copper winding width
[mm]

0.02 PM length

0.015

0.01

0.005

0
0.75 0.8 0.85 0.9 0.95
B̂1 [T]

Figure 34: Geometric variables that are used for keeping the current density in
the copper and magnetic flux density in the iron constant while the
air gap flux density, B̂1 , is varied.

5.2.1 Output power, losses and efficiency at 3000 rpm

Transient5 simulations considering a steady state operation point at 3000 rpm
were carried out for the five variations to compare their characteristics in
terms of output power, losses and efficiency. Output power can be seen in
Figure 35, and it is increased by approximately 15% when comparing the
lowest and highest values that are obtained.

As described in section 5.1.5, the losses are subdivided into copper losses and
iron losses, as shown in Figure 36. The copper losses decrease with decreased
electric loading as expected. As shown in Figure 33, the induced magnetic
flux contributes significantly to the total magnetic flux. The iron losses are
therefore reduced even though the magnetic loading and the iron volume are
increased, see Figure 36.

When efficiency, copper losses and iron losses are taken into account, it is
shown that an optimization with respect to torque does not give an optimiza-
tion of the efficiency, see Figure 37. It may also indicate that the geometry
5
Time stepping simulations are made in transient simulation mode in Maxwell 3D.

51
 25.5

25

24.5
Output power [kW]

24

23.5

23

22.5

22
0.75 0.8 0.85 0.9 0.95
B̂1 [T]

Figure 35: The output power at 3000 rpm for the five variations of the geometry.

Figure 36: The losses at 3000 rpm for the five variations of the geometry.

should be chosen in favor of magnetic loading if efficiency is the optimization

target, when current and flux densities of 7 A/mm2 and 1.7 T are used.

52
 97.2

97

96.8

96.6
Efficiency [%]

96.4

96.2

96

95.8

95.6
0.75 0.8 0.85 0.9 0.95
B̂1 [T]

Figure 37: The efficiency at 3000 rpm for the five variations of the geometry.

53
5.2.2 Final geometry
The geometry corresponding to B̂1 =0.85 T which gives the maximum torque
is chosen as the final geometry. It has been determined by: the initial re-
quirements, some assumptions and the torque optimization procedure (FEA)
previously described. Table 7 presents the variables and how they were es-
tablished.
Table 7: Variable summary for the machine design.

Variable Description Value Requirement Assumption Calculations

Do Outer diameter 260 mm ×
λ Diameter ratio 0.64 ×
Di Inner diameter 166.4 mm ×
ltot Machine length 72.5 mm ×
p Number of pole pairs 5 ×
Ns Number of stator segments 12 ×
lair−gap Air gap length 1 mm ×
lshoe Stator shoe length 5 mm ×
lshoe−space Stator shoe space 4 mm ×
lpm Permanent magnet length 3.4 mm ×
lr Rotor core length 10.1 mm ×
lcentral−bar Stator segment central bar 33.5 mm ×
Ww Width of copper winding 9.8 mm ×

5.3 Flux linkage and back-EMF

The flux linkage and back-EMF wave forms of the machine are calculated
without current excitation at 3000 rpm and with one turn around each stator
segment, n = 1.

Figure 38 shows the flux linkage of the three phases. As can be seen in the
figure, they are sinusoidal with low harmonic content. This is also confirmed
by the FFT of the flux linkage in Figure 39, where the amplitudes of all the
higher harmonics that are shown are less than 0.5% of the fundamental. The
fundamental of the flux linkage, ψm , is determined to 5.36 mWb/n.

The back-EMF wave forms for the three phases are displayed in Figure 40.
As can be expected, their wave forms show a higher harmonic content than
the flux linkage wave forms shown previously. Again, this is confirmed by
the FFT of the back-EMF which is shown in Figure 41. The 11th harmonic,
which has the greatest amplitude of the harmonics that are shown in the

54
 6

4
Phase flux linkage [mWb/n]

−2

−4

−6
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
Time [ms]

Figure 38: Per turn flux linkage calculated at 3000 rpm.

Figure 39: FFT of the flux linkage in Figure 38.

figure, is about 4% of the fundamental.

55
 10

4
Phase back−EMF [V/n]

−2

−4

−6

−8

−10
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
Time [ms]

Figure 40: Per turn back-EMF calculated at 3000 rpm.

Figure 41: FFT of the back-EMF in Figure 40.

56
5.4 Computation of the inductances
The direct-axis and the quadrature-axis inductances of a machine with sur-
face mounted magnets and without rotor saliency are often assumed to be
equal. This is also the expected result when calculating the inductances
without taking into account the magnetic flux due to the magnets or the
slightly higher permeability. The method with ”demagnetized” or neglected
permanent magnets is described in [21]. The FEA result of this method but
with the permanent magnets’ permeability taken into account is shown in
Table 8. One can observe that the inductances in the positive and negative
d-direction is about 0.5% higher. This is a consequence of the about 4%
higher permeability in the permanent magnets.

The inductances are often considered as non-linear due to saturation of the

iron. The level of saturation is determined by the magnitude of the current.
However, as seen in Figure 33, the flux from the permanent magnets con-
tributes significantly to the total magnetic flux. In fact, the rotor position
(or the current angle) has a clear impact on the inductance, as is shown in
Table 9. As expected, the inductance is at its highest in the negative d-
direction and at its lowest in the positive d-direction, as the induced flux is
counteracting respectively aligning with the flux from the permanent mag-
nets.
Table 8: Inductances with ”demagnetized” permanent magnets.

Phase Rotor position Phase inductance @

position [mech.degrees] 7A/mm2 [µH/n2 ]
-d 67.2 2.30
q 85.2 2.29
+d 103.5 2.30
-q 121.2 2.29

Table 9: Inductances with the magnetic flux due to the permanent magnets taken
into account.

Phase Rotor position Phase inductance @

position [mech.degrees] 7A/mm2 [µH/n2 ]
-d 67.2 2.56
q 85.2 2.14
+d 103.5 1.65
-q 121.2 2.22

57
5.5 Computation of the torque
5.5.1 Average and ripple
The average torque produced by the machine when applying the nominal
current, 53.5 A, at nominal speed, 2400 rpm, is 80.76 Nm, see Figure 42.
The torque ripple band is visualized by the dotted lines at 78.38 Nm and
83.23 Nm, which gives a peak-to-peak value of 4.85 Nm. This corresponds
to 6% of the average torque.

100

90

80

70

60
Torque [Nm]

50

40

30

20

10

0
0 1 2 3 4 5 6 7
Time [ms]

Figure 42: Torque production at nominal speed and nominal current.

5.5.2 Cogging torque

The cogging torque is the torque that one can feel when an unplugged BLDC
or PMSM shaft is rotated by hand. It is a consequence of the fact that the
magnetic energy that is stored in the air gap changes depending on the rotor
position. For some positions the reluctance of the magnetic circuit is mini-
mized and the energy thereby minimized. To reach the next minimum the
energy in the air gap must be increased before it drops again. The character-
istic of the cogging torque is therefore an alternating torque and its average
value ideally sums up to zero.

One period of the cogging torque is plotted in Figure 43. Its peak value is
1.32 Nm which is about 1.6% of the average torque.

58
 1.5

0.5
Torque [Nm]

−0.5

−1

−1.5
0 1 2 3 4 5 6
Rotor position [mech. degrees]

Figure 43: Cogging torque.

5.5.3 Overload and the spacing between the stator shoes

As mentioned in section 2.1.3, most electric machines are able to manage a
short time overload which can be several times the rated load. The spacing
between the stator segment shoes is expected to affect the overload capacity,
since greater spacing prevents the flux leakage between the shoes.

The short time overload is therefore investigated for up to five times the nom-
inal current and for two values of the spacing between the stator segments.
The difference in torque for the nominal current is negligibly small, as can be
seen in Figure 44. When the current is increased to two times the nominal
current, the difference is around 5 Nm. For higher currents, the difference is
then increased to around 8 Nm and remains constant within the investigated
range.

When the machine is fed by two times the nominal current, the torque is in-
creased by 62% for the assumed spacing of 4 mm. If the spacing is increased
to 6 mm the machine will produce 64% more torque than rated, when it is fed
by two times the nominal current. It was previously shown in Figure 33 that
the magnetic flux density was increased from 1.4 T to 1.7 T when comparing
the open circuit flux density and the nominal current flux density. As the
iron is almost fully utilized already at rated load conditions, there is simply

59
not much capacity left for an overload.

220

200

180

160
4mm
140 6mm
Torque [Nm]

120

100

80

60

40

20

0
0 1 2 3 4 5
⋅In

Figure 44: Overload for up to 5 times the nominal current and for a spacing of
4 mm and 6 mm respectively between the stator shoes.

The spacing between the stator segment shoes is varied in four steps to
investigate how it affects the torque. The result is shown in Figure 45; an
increase from 4 mm to 7 mm gives a 2.6% higher average torque while the
torque ripple is incresed by 70%.

60
Figure 45: Average and ripple torque when the spacing between the stator shoes
is varied.

5.6 Losses and efficiency

The machine characteristic, in terms of torque production, copper and iron
losses, is calculated in 30 operation points. The operation points are selected
using current steps corresponding to 25% of the rated current and speed
steps of 800 rpm. Both copper and iron losses are then interpolated between
the calculated operation points, which are plotted as the black dots in the
following figures.

When calculating the copper losses, the resistivity of copper has been put
to 2.39·10−8 Ωm corresponding to a temperature of 120◦ C. The expected
square relation between current and copper losses can be seen as the square
relation between torque and copper losses in Figure 46, since the torque is
proportional to the current. As mentioned previously, the frequency depend-
ing part of the copper losses such as: proximity effect, induced eddy currents
and skin effect are not included, since the windings are modeled as stranded
conductors. Therefore, the expected speed dependence cannot be seen in the
copper losses shown in Figure 46.

61
 500

400

300
PCu [W]

200

100

0
100
4000
50 3000
2000
1000
Torque [Nm] 0 0
Speed [rpm]

Figure 46: Copper losses.

The calculated iron losses together with the interpolation are shown in Fig-
ure 47. It can be seen in the figure that the iron losses depend upon both
torque and speed. This since the magnitude of the current affects the flux
density and the electric frequency is proportional to the speed. A more
distinct square relation between iron losses and electric frequency than in
Figure 47 may be expected. For the selected number of turns, the machine
is in field weakening operation for speeds above 2400 rpm. The magnetic
flux density is thereby decreased for higher speeds and the rise in iron losses
reduced.

62
 600

500

400
PFe [W]

300

200

100

0
100
4000
50 3000
2000
1000
Torque [Nm] 0 0
Speed [rpm]

Figure 47: Iron losses.

The iron losses and the copper losses are added together to calculate the
machine efficiency, which is shown in the contour plot in Figure 48. For the
rated speed and the rated torque operation point, the efficiency is 96%.

80

70

60

50
Torque [Nm]

96.5%
40 96%

30 95%
94%
20 90%

10

0
0 500 1000 1500 2000 2500 3000 3500 4000
Speed [rpm]

Figure 48: Calculated efficiency map, copper and iron losses taken into account.

63
6 Discussion, conclusion and future work
6.1 Discussion
The chosen ratio between outer and inner diameter, λ, is based on other
existing machine designs. It is probable that it could be tuned to slightly
increase the power and torque production. As the outer diameter was limited
by the available space, the inner diameter could have been varied to find an
optimum using FEA. However, this would have been very time-consuming,
as variation of the inner diameter affects other geometric variables; it was
therefore not considered in this thesis.

SMC material for the stator segments was chosen to maximize the power
and torque production by maximizing the active area of the air gap only,
since the electric frequency was too low to take advantage of the lower iron
losses at higher frequencies. Another way of reaching maximal power and
torque production could have been to not maximize the active area but in-
stead choose laminated steel due to an expected wider linear region before its
non-linear region up to saturation. As shown in Figure 30 in section 5.1.3,
the presented SMC materials enter their non-linear region already at flux
densities of 1.2-1.4 T. When using laminated steel, the overload capacity is
also expected to be higher up to a few times the nominal current and then
saturate more distinctly compared to the SMC, which was shown in Figure 44
in section 5.5.3.

In section 4.3, the number of turns was chosen to n=30 which does not
strictly give the highest possible power and torque production according to
the objective in section 1.3. In this case, the significant current increase (for
n=20) and the overall performance of the machine are given priority.

The machine volume is 3.8 liters and the weight, considering the active parts
only, is 14.8 kg. Using the average torque of 80.76 Nm and the nominal speed
of around 2400 rpm, the torque density is calculated to 5.5 Nm/kg and the
power density to 5.3 W/cm3 . Both values should be considered carefully
as both the torque density and power density are at this stage calculated
without feedback from thermal analysis. They are therefore presented only
to point out their order of magnitude and not used for any comparison with
other machines.

64
6.2 Conclusion
In this master thesis, the YASA topology was chosen as the most suitable
axial flux machine topology to meet the very demanding combination of
requirements indicated. The design process has been carried out by using
analytic calculations and FEA alternately, focusing on the FEA. As thermal
analysis was out of scope for this thesis, the assumed magnetic flux and cur-
rent densities have been the limiting factors. Using the presupposed values
of 1.7 T and 7 A/mm2 , the nominal torque has been determined to about
80 Nm. Considering the weight of the active parts only, this corresponds to
a torque density of 5.5 Nm/kg.

The machine torque speed characteristics have been investigated based on

the number of turns of the windings. The selected torque speed character-
istic, using a DC-link voltage of 500 V and n=30, gives a power density of
5.3 W/cm3 . More information about the expected operation of the machine
and the rating of the inverter can be used to make a more appropriate choice
of the number of turns to optimize the whole drive system.

It has been shown that the optimization of the rated torque influences the
efficiency of the machine; the efficiency for rated torque at rated speed is
96%. The optimization of the rated torque also affects the short time overload
capacity of the machine, due to the high level of flux density already required.
It is then possible that the amount of iron should be increased and the copper
reduced. Thereby the efficiency and the short time overload capacity can be
improved at the cost of a lowered rated torque.

6.3 Future work

Calculations considering thermal coupling in the FEA would be a more ac-
curate strategy for finding the thermal limit of the machine together with
a cooling system. It is then probable that the assumed current density of
7 A/mm2 will be altered since it is directly related to the thermal limit. The
assumed magnetic flux density of 1.7 T on the other hand is more related
to the non-linear behavior of the iron, but an altered value could also be of
interest for adjusting the iron losses.

It could also be of interest to calculate the demagnetization current, the max-

imum current that can be applied to the machine without risk of losing the
magnetization of the PMs. It may be a good idea to perform this calculation
when the temperature in the machine is known, as the demagnetization field

65
strength of the PMs is related to their temperature [22].

Depending on the material of the machine casing, the machine geometry

might have to be slightly further decreased to introduce a ”magnetic insu-
lation” that prevents magnetic flux leakage to the casing. The size of this
barrier probably has an impact on the short time overload capacity of the
machine.

It is also recommended that the shape of the stator segment shoes be further
investigated. Its geometry has been assumed based on existing designs of
YASA machines. In section 5.5.3, the spacing between the stator segment
shoes was investigated but their axial length has not been evaluated.

Other limits which have not been considered are mechanical strength and
fabrication. This could be of particular interest regarding the stator segment
shoes but also the permanent magnets, as the shapes of their edges and cor-
ners may be unnecessarily sharp in the presented design.

The chosen pole pair and stator segment combination, p/Ns = 5/12, is also
related to the mechanical strength. As an example, the axial length of the
rotor cores , lr in (34), could be reduced for a higher number of poles. The sta-
tor segments could then be made longer and the electric loading and thereby
the torque be increased. It must then be investigated how small the parts
could be made and yet be able to withstand the greater forces.

66
References
[1] Council Directive 85/3/EEC of 19 December 1984 on the weights, di-
mensions and certain other technical characteristics of certain road ve-
hicles, OJ L 2, 3.1.1985, p. 1418 (DA, DE, EL, EN, FR, IT, NL) , Celex
document number 31985L0003, Available at http://eur-lex.europa.eu/

[2] Fei, W.; Luk, P.; Jinupun, K.; , ”A new axial flux permanent mag-
net Segmented-Armature-Torus machine for in-wheel direct drive appli-
cations,” Power Electronics Specialists Conference, 2008. PESC 2008.
IEEE , vol., no., pp.2197-2202, 15-19 June 2008

[3] Cirani M. & Lindström J. ”Electric hub wheel motors: An initial study
for long haul applications” Technical report, Volvo 3P, 2009

[4] Landfors P. & Unfors F. ”A concept study of electrical wheel motors for
the front axle of a semi truck tractor”, Chalmers University of Technol-
ogy, Göteborg, 2010

[5] Gieras J.F.; Kamper M.J. and Wang R-J. ”Axial Flux Permanent Mag-
net Brushless Machines 2nd Edition” Springer Science & Business Media
B.V. 2008

[6] Hendershot J.R. & Miller TJE ”Design of brushless permanent-magnet

motors” Magna Physics Publishing and Clarendon Press, Oxford 1994

[7] Cirani M. ”Analysis of an innovative design of an axial flux Torus-

Machine”, Roal Institute of Technology, Stockholm 2002

[8] Pillay, P.; Krishnan, R.; , ”Application characteristics of permanent

magnet synchronous and brushless DC motors for servo drives,” Industry
Applications, IEEE Transactions on , vol.27, no.5, pp.986-996, Sep/Oct
1991

[9] Woolmer, T.J.; McCulloch, M.D.; , ”Analysis of the Yokeless And Seg-
mented Armature Machine,” Electric Machines & Drives Conference,
2007. IEMDC ’07. IEEE International , vol.1, no., pp.704-708, 3-5 May
2007

[10] Surong Huang; Aydin, M.; Lipo, T.A.; , ”TORUS concept machines:
pre-prototyping design assessment for two major topologies,” Industry
Applications Conference, 2001. Thirty-Sixth IAS Annual Meeting. Con-
ference Record of the 2001 IEEE , vol.3, no., pp.1619-1625 vol.3, 30
Sep-4 Oct 2001

67
[11] Aydin, M.; Surong Huang; Lipo, T.A.; , ”Design and 3D electromag-
netic field analysis of non-slotted and slotted TORUS type axial flux
surface mounted permanent magnet disc machines,” Electric Machines
and Drives Conference, 2001. IEMDC 2001. IEEE International , vol.,
no., pp.645-651, 2001

[12] Kano, Y.; Tonogi, K.; Kosaka, T.; Matsui, N.; , ”Torque-Maximizing
Design of Double-Stator, Axial-Flux, PM Machines Using Simple Non-
Linear Magnetic Analysis,” Industry Applications Conference, 2007.
42nd IAS Annual Meeting. Conference Record of the 2007 IEEE , vol.,
no., pp.875-881, 23-27 Sept. 2007

[13] Soong, W.L.; Miller, T.J.E.; , ”Theoretical limitations to the field-

weakening performance of the five classes of brushless synchronous AC
motor drive,” Electrical Machines and Drives, 1993. Sixth International
Conference on (Conf. Publ. No. 376) , vol., no., pp.127-132, 8-10 Sep
1993

[14] Mohan N.; Undeland T.M. and Robbins W.P ”Power Electronics - Con-
verters, Applications and Design” John Wiley & Sons, Inc. 2003

[15] Jiabin Wang; Howe, D.; , ”Tubular modular permanent-magnet ma-

chines equipped with quasi-Halbach magnetized magnets-part I: mag-
netic field distribution, EMF, and thrust force,” Magnetics, IEEE Trans-
actions on , vol.41, no.9, pp. 2470- 2478, Sept. 2005

[16] Furlani E.P. ”Permanent magnet and electromagnetical devices” Else-

vier Inc. 2001

[17] Lin, D.; Zhou, P.; Fu, W.N.; Badics, Z.; Cendes, Z.J.; , ”A dynamic core
loss model for soft ferromagnetic and power ferrite materials in transient
finite element analysis,” Magnetics, IEEE Transactions on , vol.40, no.2,
pp. 1318- 1321, March 2004

[18] Parviainen A. ”Design of Axial-Flux Permanent-Magnet Low-Speed Ma-

chines and Performance Comparison Between Radial-Flux and Axial-
Flux Machines” Lappeenranta University of Technology, Lappeenranta
2005

[19] Surong Huang; Jian Luo; Leonardi, F.; Lipo, T.A.; , ”A comparison of
power density for axial flux machines based on general purpose sizing
equations,” Energy Conversion, IEEE Transactions on , vol.14, no.2,
pp.185-192, Jun 1999

68
[20] EL-Refaie, A.M.; , ”Fractional-Slot Concentrated-Windings Syn-
chronous Permanent Magnet Machines: Opportunities and Challenges,”
Industrial Electronics, IEEE Transactions on , vol.57, no.1, pp.107-121,
Jan. 2010

[21] Bianchi N. ”Electrical machine analysis using finite elements” Taylor &
Francis 2005

[22] Arnold magnetic technologies ”N35EH datasheet” Available at

http://www.arnoldmagnetics.com

69