Progress In Electromagnetics Research B, Vol.

33, 383–407, 2011

GENETIC ALGORITHM AND FINITE ELEMENT ANAL-

YSIS FOR OPTIMUM DESIGN OF SLOTTED TORUS
AXIAL-FLUX PERMANENT-MAGNET BRUSHLESS DC
MOTOR

A. Mahmoudi* , N. A. Rahim, and H. W. Ping

UMPEDAC, Level 4, Engineering Tower, University of Malaya, Kuala
Lumpur, Malaysia

Abstract—This paper presents an inside-out axial-flux permanent-

magnet brushless DC motor optimized by Finite Element Analysis
(FEA) and Genetic Algorithm (GA) that uses sizing equation. The
double-sided slotted-stator designed TORUS motor has sinusoidal back
EMF waveform and maximum power density. The GA obtained the
dimensions that gave the motor its highest power density. Field
analysis of the dimensions was then put through FEA, to obtain and
re-optimize the motor’s characteristics. Possible design parameters
were investigated via use of Commercial Vector Field Opera 14.0
software used in three-dimensional FEA simulation and of MATLAB
2010a in GA programming. Techniques such as modifying winding
configuration and skewing the permanent magnets were explored
to achieve the most-sinusoidal back-EMF waveform and minimized
cogging torque. The desired technical specifications were matched by
simulation results of the 3D FEA and the GA. The FEA and the GA
simulation results comparison of the flux density in different parts of
the designed motor at no-load condition agreed well.

1. INTRODUCTION

Permanent-magnet brushless DC motors attract interest through their

high performance [1]. Reduced prices of high-energy permanent
magnets and electronics that the motors operate with encourage their
use in a wide range of applications [2]. Brushless motors exist in various
geometries; among them, a disc-type or axial-flux permanent-magnet
(AFPM) motor, in various configurations [3–7]. AFPM motor’s
Received 2 July 2011, Accepted 15 August 2011, Scheduled 18 August 2011
* Corresponding author: Amin Mahmoudi ([email protected]).
384 Mahmoudi, Rahim, and Ping

high torque-to-volume ratio, suitable efficiency, and flat structure

attract military and transport applications, motivating researchers
into developing new approaches to AFPM machine design [8]. AFPM
machines can be single-sided or double-sided, with/without armature
slots/core, have internal/external permanent-magnet rotors, have
surface-mounted or interior permanent magnet, be single-stage or
multistage [9]. Their cogging torque is usually much larger than that
of conventional motors [10], but they could also earn potential to
applications such as ship propulsion and elevator direct drive [11, 12].
Double-sided AFPM motors are the most promising and widely used
type [13]. Topologies for double-sided AFPM machines are axial-flux
one-stator-two-rotor (TORUS) and two-stator-one-rotor (AFIR) [14];
either of two arrangements (external stator or external rotor) is
practical. External-stator arrangement uses few permanent magnets
but at the expense of winding. External-rotor arrangement is
considered particularly advantageous. Where space is limited, AFPM
machine size and shape matter, and compatibility crucial. A trend
in automotive and military applications is reduced size, weight, and
cost [15]. Double-sided slotted TORUS AFPM motors have the most
application among the other configurations; they are stronger and have
higher power density [16]. Slotted TORUS AFPM motor was used here
in modeling and simulation. Use of GA and FEA in the design process
maximized the motor’s power density and enhanced its operational
performance.
Huang et al. derived the general sizing and the power
density equations for radial-flux permanent-magnet machines, also
a systematic method comparing capabilities of machines of various
topologies [17]. In 1999, they developed the sizing equation for AFPM
machines but did not present the machines’ optimized sizes [18].
A general optimization process for an AFPM machine is possible
with shape modification, via geometrical parameters, deterministic
methods, or soft computing methods. M. Aydin presents optimum-
sized AFPM machines for both TORUS and AFIR topologies,
but only two parameters (diameter ratio and air-gap flux density)
were considered optimization variables, the optimization via shape
modification [19, 20]. In all the shape-modification methods trade-offs
were observed among the performance parameters and the methods are
inapplicable to multi-objective optimization problems. Soft computing
methods are based on artificial intelligent techniques. Heuristic,
probabilistic methods require good initial estimation, give global
optimum values and are highly pliable to multi-objective optimization
problems. Highly effective computer systems and new fast-computing
algorithms make soft computing methods the current choices for
Progress In Electromagnetics Research B, Vol. 33, 2011 385

optimization of AFPM machines. In this paper, GA was the

optimization tool minimizing AFPM machine size. Various parameters
were considered, making the problem multi-objective. Note that the
method is comprehensive and good for designing an arbitrary-capacity
arbitrary-parameter double-sided AFPM machine; However, only
slotted TORUS AFPM motor was optimized, and the equations used
were for slotted TORUS AFPM motor. Between past literatures and
the authors’ best knowledge, minimization of AFPM machine size with
various parameter considerations (winding turns, winding coefficient,
electrical loading, air-gap length, diameter ratio, and air-gap flux
density) has yet to be presented. There are standard techniques for
optimization [21] (some of them more popular than others in electrical
machines‘ design application): Random Search Method Hock and
Jeeves Method, Powell Method, and Genetic Algorithm. Random
Search Method needs a lot of time to converge and depends completely
on the starting point [22]. Hook and Jeeves Method is slower than
Random Search Method but is more accurate [22]. Powell Method is
able to quickly reach optimal solution, but may be not robust when the
problem is complicated or when the desired global minimum is hidden
among many local minimums [22]. The main feature of the genetic
algorithm is a population of points in parallel rather than a point to
be searched [23, 24]. Many design parameters of an AFPM machine
can be varied in parallel, affecting each other; the optimization is non-
linear, so GA is a suitable optimization method.
Presented is the design of a slotted TORUS AFPM motor with
sinusoidal back EMF and maximum power density. GA optimized
it, and FEA analyzed its performance. The paper is organized as
follows: sizing equation for the slotted TORUS AFPM motor was first
derived in Section 2 via a generalized sizing equation, to calculate the
motor’s power-production potential (it was used as fitness function
in the GA-optimized machine design, to minimize power-to-volume
ratio); Section 3 presents the GA optimization process including design
restrictions, requirements, chromosome representation, crossover, and
mutation; FEA was next performed for electromagnetic field analysis of
the proposed motor topology, calculating the air-gap flux distribution,
verifying results from analysis of the sizing equation, all in Section 4;
Section 5 discusses all the results and concludes.

2. SIZING EQUATION

The main dimensions of each electrical machine are determined via

electrical-machine-output power equation. Assuming negligible leakage
386 Mahmoudi, Rahim, and Ping

inductance and resistance, rated power is expressed as [18]:

Z
m T
Pout = η e(t) · i(t)dt = mKp ηEpk Ipk (1)
T 0
e(t) is phase air-gap EMF, i(t) is phase current, η is machine efficiency,
m is number of machine phases, and T period of one EMF cycle. Epk
and Ipk are peaks of phase air-gap EMF and of current, respectively.
Kp is electrical power waveform factor, defined as:
Z Z
1 T e(t) · i(t) 1 T
Kp = dt = fe (t) · fi (t)dt (2)
T 0 Epk · Ipk T 0
where fe (t) = e(t)/Epk and fi (t) = i(t)/Ipk are expressions for
normalized EMF and current waveforms. For effect of current, the
current waveform factor (Ki ) is defined as:
Ipk 1
Ki = =r ³ ´2 (3)
Irms R
1 T i(t)
T 0 Ipk dt

where, Irms is phase-current rms value. Table 1 lists typical waveforms

and their corresponding power-waveform factor (Kp ) and current-
waveform factor (Ki ) [25]. The peak value of phase-air-gap EMF for
AFPM machine in Equation (1) is:
f¡ ¢
Epk = Ke Nph Bg 1 − λ2 Do2 (4)
p

Table 1. Typical prototype waveforms.

Model e(t) i(t)

t) Ki Kp

Sinusoidal 2 0.5cosϕ

Sinusoidal 2 0.5

Rectangular 1 1

Trapezoidal 1.134 0.777

Triangular 3 0.333
Progress In Electromagnetics Research B, Vol. 33, 2011 387

Ke is EMF factor incorporating winding distribution factor (Kw ) and

per-unit portion of air-gap area-total spanned by machine’s salient
poles (if any); Nph is number of winding turns per phase; Bg is flux
density in air-gap; f is converter frequency; P is machine pole pairs;
λ is AFPM machine’s diameter ratio Di /Do ; Do is the diameter of
machine outer surface; Di is the diameter of machine inner surface.
Equation (1)’s peak phase current is:
1 + λ Do
Ipk = Aπ Ki (5)
2 2m1 Nph
where m1 is the number of phases of each stator, and A is the
total electrical loading. A general-purpose sizing equation for AFPM
machines takes the following form:
µ ¶
1 m π f ¡ 2
¢ 1+λ
Pout = Ke Ki Kp KL ηBg A 1−λ Do2 Le (6)
1 + K ϕ m1 2 P 2
Le is the motor’s effective axial length; Kϕ is the electrical loading ratio
on rotor and stator; KL is the aspect ratio coefficient pertinent to a
specific machine structure, considering effects of losses, temperature
rise, and the design’s efficiency requirements. Machine power density
for volume total is defined as:
Pout
Pden = π 2 (7)
4 tot Ltot
D
Dtot and Ltot respectively are total of the machine’s outer diameter and
total of the machine’s length including stack’s outer diameter and end-
winding protrusion from radial and axial iron stacks. The generalized
sizing equation approach can easily be applied to a double-sided AFPM
TORUS-type machine. The outer surface diameter (Do ) can be written
as: s
Pout
Do = 3 π m f ¡ 1+λ ¢ (8)
2
2 m1 Ke Kp Ki ABg η p (1 − λ ) 2

Machine outer diameter total Dtot for the TORUS motor is given by:
Dtot = Do + 2Wcu (9)
where Wcu is protrusion of end winding from iron stack, in radial
direction. For back-to-back wrapped winding, protrusions exist
towards the machine’s axis and towards its outsides, and can be
calculated as: q
Di − Di2 − 2AD ave
Kcu Js
Wcu = (10)
2
388 Mahmoudi, Rahim, and Ping

where Dave is the machine’s average diameter, Js is current density,

and Kcu is copper-fill factor. Machine’s axial length Le is given by:
Le = Ls + 2Lr + 2g (11)
Lr is rotor’s axial length, and g is air-gap length. Stator’s axial length
Ls can be written as:
Ls = Lcs + 2Lss (12)
Note that for slotted machines, stator slot depth is Lss = Wcu . Stator
core’s axial length Lcs can be written as:
Bg παp Do (1 + λ)
Lcs = (13)
4pBcs
where Bcs is stator-core flux density, and αp is ratio of average-air-
gap flux density to peak-air-gap flux density. Rotor’s axial length Lr
becomes:
Lr = Lcr + Lpm (14)
Lpm is permanent-magnet length; rotor core’s axial length Lcr is:
Bu πDo (1 + λ)
Lcr = (15)
8pBcr
where Bcr is flux density in rotor disc core, and Bu is attainable flux
density on surface of permanent magnet, whose length Lpm can be
calculated as:
µr Bg
Lpm = ³ ´ Kc g (16)
K
Br − Kfd Bg
where µr is magnet’s recoil relative permeability, Br is permanent-
magnet material residual-flux density, Kd is leakage flux factor, Kc is
Carter factor, Kf = Bgpk /Bg is peak-value corrected factor of air-gap
flux density in radial direction of AFPM machine.

3. GENETIC ALGORITHM AND OPTIMIZATION

The design procedure for highest-possible-power-density AFPM motor

is a multi-dimensional optimization problem of achieving a goal
function within constraints. As Equations (1) to (16) show, many
parameters affect each other, varying in parallel (for example: Di ,
Do , and λ); the AFPM machine’s optimization is thus a non-
linear problem. GA, a powerful tool, is capable of solving various
complex and non-linear optimization problems [26]. It is a search
algorithm inspired by life’s natural-selection mechanisms and natural
genetics [27, 28]. Parameters are first coded in the solution area, to
Progress In Electromagnetics Research B, Vol. 33, 2011 389

specific-length arrays (each array has a definite fitness that depends

on application). GA then searches for an optimal solution; it includes
chromosome representation of the solution, initializing of the first
generation, cross over, and mutation [29]. In GA, inputs and their
scopes of change are first determined as genes, which create the
chromosomes making up a population [30]. The algorithm features
a population of points in parallel rather than a point to be searched.
Most of the other algorithms are not parallel and can solve the problem
in just one direction, concurrently. In this case, if a solution is a local
optimal solution or a subset of the original answer, all the procedures
should be repeated. GA has various starting points, so it can search
the solution in different directions simultaneously. If an archived
solution is defeated, other available solutions could be continued and
more resources are provided. In non-linear problems, variation in one
parameter may have a disharmonic effect or a significant change on the
whole system. Its parallelism and the various directions considered in
solution selection make GA appropriate for the optimization problems
with big domain and non-linearity leading to proper answer in a rather
short time [31]. This property also makes GA a good option for
multi-objective optimization problems because it can change many
parameters simultaneously. GA may thus provide more than one
answer for an optimization problem; each by considering a special
parameter. There are also other methods able to expedite the
solution process and improve answer accuracy; they are applicable
when knowledge about domain interval increases. Another advantage
of GA is that it does not need any knowledge about differentiability
and can be used in noisy environments.
A difficulty in genetic algorithm programming is how to select
the fitness function leading to the best solution of the problem [32].
An inappropriate fitness function may lead to the wrong answer. To
select the fitness function, other parameters, too, should be considered:
number of population, chromosome representation, mutation, and
crossover. Another possible problem arises when one of the genes
created is rather better than the other genes [33]; the answer may
go towards the local solution. This is overcome by selecting a large
number of populations. The following sections present the first and
foremost parts of GA optimization: selection of the fitness function,
number of the chromosome’s genes, and design limitations.

3.1. Design Restrictions and Requirements

An optimum design would have maximum power density incorporated
with desired sinusoidal back-EMF maintained within design restric-
tions and requirements. Some of the motor’s parameters and charac-
390 Mahmoudi, Rahim, and Ping

teristics cannot vary much, inherently or owing to material and ap-

plication limits. Besides maximum power density and sinusoidal back
EMF, other limitations should be considered. Table 2 lists limits of
the design procedure.
Figure 1 shows GA optimization algorithm for highest power
density of the AFPM motor. GA begins with population, an
initial set of random solutions. A population contains chromosomes,
string-structured concatenated lists of binary digits that code the
control parameters of a given problem. In this paper, a population
of 1400 strings was created randomly and the chromosomes were
normalized. The chromosomes evolve from generation to generation
through successive iterations, each generation evaluated by a measure
of fitness. Here, to create the next generation, half the genes were
selected to breed a new generation, the other half eliminated. Machine
power density (Equation (7)) is selected as fitness function and is
in each step calculated for each chromosome. Individual solutions
are fitness-selected, fitter solutions measured by the fitness function
likelier to be selected. The stopping criterion is then checked; if it

Table 2. Design restrictions and requirements.

Dimensional Constraints
Stator Outer Diameter (Do ) Do ≤ 300 mm
Inner to Outer Ratio (λ) 0.4 ≤ λ ≤ 0.75
Total Axial-Length (Le ) ≤ 500 mm
Air-Gap Length (g) 0.5 ≤ g ≤ 2.5
Material Limitations
Stator and Rotor Core
Bcs , Bcr ≤ Bmax = 1.5 T
Flux Density (Bcs , Bcr )
Permanent Remanence 1.3 T
Requirements
Rated Line-to-Line Voltage (rms) 100 V ≤
Input Phase Current (rms) ≤ 20 A
Air-gap Flux Density (Bg ) 0.35 T ≤ Bg ≤ 0.95 T
Electrical Loading (A) 1000 ≤ A ≤ 30000
Output Power (Pnom ) 1 kW
Pole Pairs (P ) 2
Motor Efficiency (η) η ≥ 80%
Frequency (f ) 50 Hz
Number of Phases (m) 3
Progress In Electromagnetics Research B, Vol. 33, 2011 391

Start

Random generation of
the first population

Fitness Function Calculation

Goal Function Calculation

Selection of Superior genes
Elimination of weak genes
Regeneration using
crossover and mutation
No Stopping
Criterion

Yes

Genes Selection

End

Figure 1. Optimization via GA.

is satisfied, the algorithm stops and the final genes are selected; else,
new chromosomes called off-springs are created. New generation is
formed by selecting, according to fitness value, some parents and some
off-springs; others are rejected to maintain population size. Half the
genes from previous steps are omitted, and new generation is created
through application of crossover and mutation on selected genes. For
each two selected genes, two children are created, replacing omitted
genes. So, a new generation with the same population as before (1400)
is created. All the previous steps are applied on the new generation,
and after several generations, the algorithm is ended when the stopping
criterion is satisfied. Finally, suitable selected genes give the motor
optimal or near optimal dimensions for the highest power density.

3.2. Chromosome Representation

Genes are effective variables for fitness function and motor
performance. As mentioned, many parameters affect AFPM motor
392 Mahmoudi, Rahim, and Ping

0.8

Gnorm

Bg λ g A Kw Nph 0.2

Gmin G Gmax

Figure 2. Chromosome repre- Figure 3. Real coding of the

sentation (1 × 6 array). genes (linear normalization).

operation and they depend on each other. Fig. 2 shows the 1 × 6 array
of each chromosome of the proposed GA; Bg , λ, g, A, Kw and Nph are,
respectively, air-gap flux density, inner to outer diameter ratio, air-
gap length, electrical loading, winding coefficient, and winding turns
in each phase. Chromosome population in each generation is 1400 and
is selected randomly in the first generation.
Genes or chromosome variables are real values, so real coding is
applied to each gene’s normalization (see Fig. 3). Linear normalization
is the result of:
0.8 − 0.2
Gnormal = · (G − Gmin ) + 0.2 (17)
Gmax − Gmin
where G is the chromosome gene varying between Gmin and Gmax .

3.3. Crossover
Crossover specifies how Genetic Algorithm combines two individuals or
parents to form a crossover child for the next generation. Methods of
gene selection and elimination are roulette wheel selection, tournament
selection, elitist selection, etc. The elitist method was used here, as
selection operator for two-point crossover (see Fig. 3). Two random
numbers in the interval between “1” and chromosome length “−1” were
first generated (1 ≤ Random Number ≤ chromosome length“−1”).
Each chromosome was then cut from the indicated points in Fig. 5, and
the corresponding sections were exchanged.

3.4. Mutation
Mutation options specify how Genetic Algorithm makes small random
changes in population individuals to create mutation children.
Progress In Electromagnetics Research B, Vol. 33, 2011 393

g w ph

' ' ' '

g' ' w ph

' '
g w ph

' ' '

g w ph

Figure 4. Two-point crossover.

Mutation provides genetic diversity and enables GA’s search for

broader space. For each method of coding, distinct mutation operators
are usually defined. Mutation length is important and must be
controlled. Mutation operation is executed with probability Pm
(0.005 ≤ Pm ≤ 0.05) and should result in a valid chromosome. For
example, in real coding, genes are randomly chosen: a random value
is chosen from the interval mentioned, then added to, or reduced,
from the gene pool. GA uses various conditions to determine when to
stop the algorithm. Presented algorithm stops when fitness function
value for best current-population point is less than, or equals, fitness
limit (Gn+1 − Gn ≤ ε). Table 2 lists each gene’s optimization-allowed
variations.

3.5. GA Results
For a 3-phase 2-pole-pair AFPM motor, possible number of slots is
assumed to be 9, 12, 15, 18, 21, and 24; a program for stators with
those numbers of slots was executed. The number of slots in each
pole, per phase, for 9, 15, 18, and 21 slots, is fractional. Winding
configuration of the fractional slot-pitch is not as easy as full slot-pitch
but all the values are considered valuable because they reduce current,
voltage harmonics, and cogging torque. AFPM brushless DC motor
can have any even number of magnet poles (2P ) and any number of
slots (Ns ). From this infinite set, only a small number of magnet
pole and slot count combinations maximize use of stator slots and
lead to efficient torque production. Appendix A presents the brushless
DC AFPM motor’s optimized winding configurations simulated in this
paper.
Table 3 lists various parameters of the motor’s design, with the
different number of stator slots obtained from GA optimization. Fig. 5
shows the MATLAB-programming fitness-function variation during
optimization of the various-slot-count stators [34].
394 Mahmoudi, Rahim, and Ping

Table 3. Dimensions of the motor, with highest power density

obtained via GA.

No. of P den Do A G Lpm Lcs Lcr Bg λ Ds

3 Nph
slots (W/cm ) (mm) (A/m) (mm) (mm) (mm) (mm) (mm) (mm) (mm)
9 0.35 157 71 16089 1.2 4.6 13.44 12.9 0.49 0.52 16.9
12 0.35 166 64 14370 1.07 2.73 12 11.7 0.4 0.46 16.64
15 0.36 170 90 14115 1.03 2.22 14 11 0.46 0.47 16
18 0.36 158 76 16757 1.26 3 13 12.5 0.46 0.55 17.25
21 0.36 152 69 16503 1.29 3.76 13.3 12.8 0.51 0.5 17.9
24 0.36 162 67 15070 1.21 2.63 12.99 12.4 0.46 0.5 16
Power Density (W/cm3)

0.36
0.35
0.34
0.33
0.32 9 Slots
0.31 12 Slots
0.3 15 Slots
18 Slots
0.29 21 Slots
0.28 24 Slots
0.27
0.26
0 10 20 30 40 50 60 70 80 90 100 110 120
Generation

Figure 5. Fitness function vari- Figure 6. Magnet skew diagram

ation during GA optimization. for cogging torque and reduction
of undesired harmonics.

4. FINITE ELEMENT ANALYSIS (FEA)

GA is used to obtain maximum power density, so dimensions obtained

via GA are considered raw data; they need further analysis to be
mature enough for final design. 3D FEA is used to analyze the double-
sided, TORUS AFPM motor’s magnetic circuit and power density,
giving an overall picture of the saturation levels in various parts of the
motor and extracting the motor’s characteristics. Advantage of the
3D FEA approach is that different components of flux density can be
calculated with high accuracy [35–39].
Usually, magnet skewing is applied to reduce cogging torque in
electric machinery. It reduces back-EMF and eliminates some of the
undesired harmonic components. Maximum, skewing angle should
equal slot pitch, not exceed it. Fig. 6 is a diagram of the magnet’s
geometric skewing relative to stator teeth and slots. Skew angle θi
is defined as the angle in which the rotor pole is skewed relative to
the stator teeth. GA analysis produced stator dimensions for each
slot count. FEA gave THD of the back-EMF waveform, in various
Progress In Electromagnetics Research B, Vol. 33, 2011 395

skew angles; see Fig. 7. The minimum THD was obviously for the
motor with 15 slots in 9 degrees. Also, the flux density obtained from
FEA was a little less than that calculated theoretically via GA, owing
to core magnetic reluctance having been neglected. In real conditions,
however, flux density of the different core parts decrease via MMF drop.
Optimality can be possible via slight changes to magnet thickness, air-
gap length, and lengths of stator yoke and rotor yoke; the best design is
achieved with the utmost skill of the FEA with extreme difficulty of the
changing magnet thickness, air-gap length, and lengths of stator yoke
and rotor yoke, for several times. Table 4 lists the machine design’s
final dimensions and specifications.
Figure 8 shows only an eighth part of the motor, which was used
to model the FEA-designed AFPM motor’s structure: 90 degrees of
the entire motor structure and 1 pole, fulfilling symmetry conditions.
Total Harmonic Distortion (THD)

10 9 Slots
12 Slots
9 15 Slots
18 Slots
8 21 Slots
24 Slots
7
6
5
4
3
2
0 5 10 15 20 25 30
Magnet Skew Angles θ i (in Degrees)

Figure 7. Back-EMF THD variation against skew angles for motors

with different number of slots.

Surface contour: BMOD

1.500000E+000

1.400000E+000

1.200000E+000

1.000000E+000

8.000000E-001

6.000000E-001

4.000000E-001

2.000000E-001

0.000000E-001

(a) (b)
Figure 8. Field analysis of an AFPM motor, in vector field opera 14.0
software. (a) 3D auto-mesh generation, (b) flux-density plot.
396 Mahmoudi, Rahim, and Ping

Table 4. The motor’s final design dimensions and specifications.

Nominal Voltage Vnom 70 V

Nominal Power Pnom 1 kW
Number of pole pairs P 2
Number of phases m 3
Drive Frequency f 50
Efficiency η 87.8 %
Outer Diameter Do 170 mm
Inner Diameter Di 80 mm
Inner to Outer Diameter’s Ratio λ 0.47
Magnet’s axial length Lpm 2.5 mm
Pole Pitch γp 118◦
Stator-yoke thickness 2 × Lcs 30 mm
Rotor-yoke thickness Lcr 11 mm
Slot Width Ws 10 mm
Slot Depth Ds 16 mm
Number of slots Ns 15
Number of winding turns per phase Nph (15 × 18)/3
Air-Gap Flux Density Bg 0.47 T
Air-gap length g 1 mm

The whole machine comprises 15 slots and 2 pole-pairs. Fig. 8(a)

(generated on Vector Field Opera 14.0 software) is a three-dimensional
auto-mesh: tetrahedral elements with 6 nodes fitting circular shape of
layers starting from shaft to outer diameter of the AFPM motor [40].
Fig. 8(b) is distribution of the magnetic flux density in different sectors
of the AFPM motor.
Magnetic flux density evaluation in different sectors of an AFPM
machine is important because if flux density of core or teeth goes to
saturation, machine efficiency reduces, affecting operation. Fig. 9 is
air-gap flux density distribution, in average radius. Maximum flux
density is obviously 0.9 Tesla, averaging 0.5 Tesla.
Figure 10 shows the magnetic flux density in stator yoke and
teeth at average radius. The maximum magnetic flux was 1.45 Tesla,
averaging 1.2 Tesla. Fig. 11 is the magnetic flux density distribution
at average rotor yoke and magnet surface radius. Table 5 lists average
magnetic flux densities from the FEA simulation results and the
Progress In Electromagnetics Research B, Vol. 33, 2011 397

0.9

Flux Density (Tesla)

0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1

0 90 180 270 360

Rotor Position (in mechanical degrees)

Figure 9. Magnetic flux density distribution of air-gap, for average

radius.
Flux Density (Tesla)

1.4 1.1

Flux Density (Tesla)

1.3 1.0
1.2 0.9
1.1 0.8
1.0 0.7
0.9 0.6
0.8 0.5
0.7 0.4
0.6 0.3
0.5 0.2
0.4 0.1
0 90 180 270 360 0 90 180 270 360
Rotor Position (in mechanical degrees) Rotor Position (in mechanical degrees)
(a) (b)

Figure 10. Stator magnetic flux density, for average radius. (a) Stator
yoke, (b) stator teeth.
1.4
Flux Density (Tesla)

0.8
Flux Density (Tesla)

1.3 0.7
1.2 0.6
1.1 0.5
1.0 0.4
0.9 0.3
0.8 0.2
0.7 0.1
0.6
0 90 180 270 360 0 90 180 270 360
Rotor Position (in mechanical degrees) Rotor Position (in mechanical degrees)
(a) (b)

Figure 11. Magnetic flux density distribution for average radius,

on (a) rotor yoke, (b) magnet surface.

sizing equation analysis of various parts of the motor design’s no-load

condition.

4.1. Back-EMF Waveform

The aim is to design an AFPM motor with sinusoidal waveform, i.e.,
the back EMF should be as sinusoidal as possible. Fig. 12 shows the
back EMF at 1500 rpm, also the FEA-calculated THD and back-EMF
RMS.
398 Mahmoudi, Rahim, and Ping

Table 5. Magnetic flux density compared among various parts of the

motor, at no-load condition.

Air-gap Stator yoke Rotor yoke Magnet surface Teeth

Bg Bcs Bcr Bm
Bt
Ave. Max. Ave. Max. Ave. Max. Ave. Max.
FEA 0.47 0.94 1.20 1.5 1.4 1.45 0.5 0.9 0.36 0.59 1.1
Sizing Eq. 0.48 0.95 1.25 1.5 1.42 1.48 0.5 0.9

60 0.3

Cogging Torque(N.m)
40 0.2
Back EMF (Volt)

Phase A-After Skew

Phase B-After Skew Skew 9 degrees
Phase C-After Skew without skew
20 Phase A-Before Skew 0.1
Phase B-Before Skew
Phase C-Before Skew
0 0
0 30 60 90 120 150 180 210 240 270 300 330 360 0 10 20 30 40 50 60
−20 − 0.1
− 40
−0.2
− 60 THD =2.5 %
RMS BackEMF = 41V −0.3
Electrical Angle (in degrees) Rotor Position (inmechanical degrees)

Figure 12. Back EMF at speed Figure 13. Cogging torque,

1500 rpm. with, and without, skewing.

4.2. Torque
Undesired torque ripple due to unwanted harmonics in output torque is
one of the important concerns in designing an electric machine. Factors
of undesired torque in AFPM machines are: ripple owing to permanent
magnet harmonics and cogging torque. Reciprocal effect of permanent
magnet and harmonic components is owed to winding distributions and
current harmonic components from drive circuit. Stator’s magnetic
reluctance changes owing to its teeth; interaction of its magnetic
reluctance with permanent magnet creates an oscillatory output torque
called cogging torque, resulting from permeance difference between
that of the stator teeth and slots and that of the permanent magnet.
If rotor rotates freely, it stops at the position of minimum magnetic
reluctance. Fig. 13 shows cogging torque of the AFPM motor with,
and without, skewing. Pre-skewing, peak cogging torque was 0.29 Nm.
Skewed magnets reduce cogging torque; at 9-degree skewing, peak
cogging torque reduced to 0.17 Nm (a 40% reduction). Stator magnetic
reluctance interaction with permanent magnet varies when stator’s
magnetic reluctance changes owing to teeth’s presence.
Progress In Electromagnetics Research B, Vol. 33, 2011 399

4.3. Efficiency
For accurate assessment of machine efficiency and thermal behavior,
calculation of the losses is crucial. Machine efficiency is:
Pout
η= (18)
Pout + Pcu + Pcor + Prot
where Pcu , Pcor , Prot are respectively copper loss, core loss, and
rotational loss components. Copper losses (Rs × I 2 ) make up most of
the loss total. Stator resistance (Rs ) depends on load and on winding
temperature [41].
2Nph-s (l + le )
Rs = (19)
σT Nph-p scu
Nph-s is number of winding turns in series per phase, Nph-p is number
of winding turns in parallel per phase, σT is electric conductivity of
wire at temperature T , and scu is cross-section area of wire. Thin
parallel wires minimized skin effect, eliminating its consideration in
Equation (19). l and le are coil length and end-winding length,
respectively.
FE-AC analysis was repeated for every space harmonic component
(up to the 49th order) and every current waveform’s simulated time
harmonic component, to get the eddy current losses in the stator steel.
Core loss for stator laminated 0.1 mm thick, calculated via FE-AC
analysis, was 20 W. Fig. 14 shows the motor’s efficiency in various
speeds. Rotational loss (which includes windage and friction losses)
was estimated from [42]:
1 ¡ ¢¡ ¢
Prot = cf ρ πn3 Do5 − Di5 (20)
2

100%
90%
80%
70%
Efficiency

60%
50%
40%
30%
20%
10%
0%
0 250 500 750 1000 1250 1500
Speed (rpm)

Figure 14. Efficiency versus speed.

400 Mahmoudi, Rahim, and Ping

where cf is friction coefficient, ρ is density of the rotating part, and n is

rotation speed (in ‘rotation per second’). Efficiency of the laminated-
stator motor, obtained with full loading, was 87.8%.

5. CONCLUSION
Presented was an optimized AFPM brushless DC motor, the design
was aided by Genetic Algorithm and Finite Element Analysis. GA
sought to obtain the maximum power per volume for a 1 kW two-
pole-pair AFPM slotted TORUS motor. 3D-FEA then changed
and moderated the design parameters based on the electromagnetic
field analysis. This approach’s advantages include the capability
to study different components of flux density, and to handle more
complicated core and winding geometries. Different constructions and
winding configurations were examined and compared and the best
one was chosen. Various characteristics of the proposed design were
investigated and compared with the ones desired. The simulated
and the desired values agreed. Flux density of various parts of the
optimal motor compared between the 3D-FEA and the sizing equation
analysis, without-load, agreed. The method is comprehensive and good
for designing an arbitrary-capacity arbitrary-parameter double-sided
AFPM motor.

ACKNOWLEDGMENT
The authors acknowledge University of Malaya’s provision of the High
Impact Research Grant No. D000022-16001 funding the Hybrid Solar
Energy Research Suitable for Rural Electrification.

APPENDIX A.
A method described in [43] was used to place the coils. There are
infinite possibilities for pole and slot count combinations as there are
for windings layouts; assumptions are necessary, either for focus or for
scope limitation, so desirable windings can be found. The assumptions
were:
a) Three-phase motor.
b) All slots filled; the number of slots is thus a multiple of the number
of phases (i.e., Ns = k × Nph ); for three-phase motors, the number
of slots is thus always a multiple of three.
c) Two coil-sides in each slot, the winding can be classified as double-
layer winding.
Progress In Electromagnetics Research B, Vol. 33, 2011 401

d) Balanced-windings only, i.e., only pole and slot count combina-

tions that result in back EMF of phases B and C being 120oE
offset from back EMF of phase A.
e) Coils have equal number of turns, all spanning equal number of
slots, implying same-sized coils and therefore same resistance and
same inductance.
The assumptions routinely lead to motors capable of high
performance, and to motors that are readily wound. Motors can be
wound violating one or more of the assumptions, but they may be more
difficult to wind; such winding could also lower performance. Fig. A1
shows the coil arrangements (9, 12, 15, 18, 21, and 24 slots) that gave
the best sinusoidal waveforms. A, B, and C represent the phases, and
+ and − represent direction of the windings.

Slot
No. 1 2 3 4 5 6 7 8 9
up +A -B -A +B -C -B +C -A -C
down +A +C -A +B +A -B +C +B -C
(a)
Slot
No. 1 2 3 4 5 6 7 8 9 10 11 12
up +A -B +C -A +B -C +A -B +C -A +B -C
down +A -B +C -A +B -C +A -B +C -A +B -C
(b)
Slot
No. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
up +A -B +C -A +B +B -C +A +A -B +C -A +B -C -C
down -B -B +C +C -A -C -C +A -B +C -A -A +B +B +A
(c)
Slot
No. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
up +A -B -B +C -A +B +B -C +A +A -B +C +C -A -A +B -C -C
down +A -B +C +C -A -A +B -C -C +A -B -B +C -A +B +B -C +A
(d)
Slot
No . 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
up +A -B -B +C +C -A -A +B -C -C +A +A -B -B +C -A -A +B +B -C -C
down +A -B -B +C +C -A +B +B -C -C +A +A -B +C +C -A -A +B +B -C +A
(e)
Slot
No. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
up +A +A -B -B +C +C -A -A +B +B -C -C +A +A -B -B +C +C -A -A +B +B -C -C
down +A +A -B -B +C +C -A -A +B +B -C -C +A +A -B -B +C +C -A -A +B +B -C -C
(f)
Figure A1. Stator winding constructions for 9, 12, 15, 18, 21, and 24
slots. (a) 9-slot double-layer stator winding (coil span = 2), (b) 12-slot
double-layer stator winding (full-pitch), (c) 15-slot double-layer stator
winding (coil span = 2), (d) 18-slot double-layer stator winding (coil
span = 4), (e) 21-slot double-layer stator winding (coil span = 5),
(f) 24-slot double-layer stator winding (full-pitch).
402 Mahmoudi, Rahim, and Ping

Table A1. Possible winding configurations and number of slots in

each pole, per phase.
Number of slots
Configuration Number of Coil Pitch/Pole
in each pole per
Number Slots Pitch
phase (Nspp )
1 9 2/2.25 0.75
2 12 2/3 1
3 12 full-pitch 1
4 15 2/3.75 1.25
5 15 3/3.75 1.25
6 18 3/4.5 1.5
7 18 4/4.5 1.5
8 21 3/5.25 1.75
9 21 4/5.25 1.75
10 21 5/5.25 1.75
11 24 4/6 2
12 24 5/6 2
13 24 full-pitch 2

The number of winding configuration options can also be increased

by short-pitching the fractional-slot structures. The 15-slot stator was
designed with a 3-slot coil span, but a 2-slot coil span is possible,
reconfiguration for it easy. For an 18-slot structure, 3-slot coil span,
and for 21-slot structure, both 3-slot and 4-slot coil spans can be
considered. Considering the 13 stator configurations in Table A1, and
possible magnet spans, their losses, back-EMF harmonic content and
pulsating torque components investigated. Efficiencies were found to
not differ much except at lower speeds, where the differences were more
pronounced, owing to copper losses. The worst structure in terms of
copper losses was found to be 24-slot full-pitched; the best was the
15-slot, either 2 or 3.75, short-pitched structure.

APPENDIX B. NOMENCLATURE

A electrical loading total [A] Bcr rotor-disc flux density [T]

Bcs stator-core flux density [T] Bg air-gap flux density [Wb/m2 ]
Bgpk peak value of air-gap Br permanent-magnet
flux density [Wb/m2 ] residual-flux density [T]
Bu flux density on Dave machine stator
permanent-magnet surface [T] average diameter [m]
Progress In Electromagnetics Research B, Vol. 33, 2011 403

Di machine stator Do machine stator

inner diameter [m] outer diameter [m]
Ds slot depth Dtot machine outer
diameter total [m]
Epk peak value of G gene value
phase-air-gap EMF
Gmax gene maximum value Gmin gene minimum value
Gnormal normalized chromosome Ipk phase current peak value [A]
Irms phase current rms value [A] Js current density [A/m2 ]
Kc Carter factor Kcu copper fill factor
Kd leakage-flux factor Ke EMF factor
Kf peak value corrected Ki current waveform factor
factor of air-gap flux density
KL aspect ratio coefficient Kp electrical power
waveform factor
Kw winding distribution factor Kϕ electrical loading ratio
Lcr rotor-core axial length [m] Lcs stator-core axial length [m]
Le effective axial Lpm permanent-magnet
length of motor [m] length [m]
Lr rotor axial length [m] Ls stator axial length [m]
Lss stator slot depth [m] Ltot machine axial
length total [m]
Nph number of winding Nph-p number of winding turns
turns per phase in parallel per phase
Nph-s number of winding Ns Number of stator slots
turns in series per phase
P number of motor pole pairs Pnom nominal power
T period of one EMF cycle [s] Vnom nominal voltage
Wcu end-winding protrusion Ws slot width
from iron stack [m]
cf friction coeficient e(t) phase-air-gap EMF [V]
f electrical frequency [Hz] fe (t) normalized
EMF waveforms
fi (t) normalized g air-gap length [m]
current waveforms
i(t) phase current [A] l coil length [m]
le end winding length [m] m number of
machine’s phases
m1 number of n rotation speed [rs−1 ]
phases of each stator
scu cross-section αp average air-gap
area of wire [m2 ] flux density to
its peak value ratio
γp pole pitch η motor efficiency
θi permanent magnet λ diameter ratio
skew [in degrees]
µr recoil relative ρ density [kg/m3 ]
permeability of magnet
σT electric conductivity of wire [S/m]
404 Mahmoudi, Rahim, and Ping

REFERENCES

1. Hadef, M., M. R. Mekideche, A. Djerdir, and A. Miraoui,

“An inverse problem approach for parameter estimation of
interior permanent magnet synchronous motor,” Progress In
Electromagnetics Research B, Vol. 31, 15–28, 2011.
2. Touati, S., R. Ibtiouen, O. Touhami, and A. Djerdir,
“Experimental investigation and optimization of permanent
magnet motor based on coupling boundary element method with
permeances network,” Progress In Electromagnetics Research,
Vol. 111, 71–90, 2011.
3. Cho, C. P. and B. K. Fussel, “Detent torque and axial force effect
in a dual air-gap axial-field brushless motor,” IEEE Transaction
on Magnetics, Vol. 6, 2416–2418, November 1993.
4. Mendrela, E. A., R. Beniak, and R. Wrobel, “Influence of stator
structure on electromechanical parameters of torus-type brushless
DC motor,” IEEE Transaction on Energy Conversion, Vol. 18,
No. 2, 231–237, June 2003.
5. Lukaniszyn, M., M. Jagiela, and R. Wrobel, “A disc-type motor
with co-axial flux in the stator-influence of magnetic circuit
parameters on the torque,” Electrical Engineering, Vol. 84, No. 2,
61–71, 2002.
6. Chalmers, B. J. and E. Spooner, “An axial-flux permanent-magnet
generator for a gearless wind energy system,” IEEE Transaction
on Energy Conversion, Vol. 14, 251–257, June 1999.
7. Barakat, G., T. El-Meslouhi, and B. Dakyo, “Analysis of the
cogging torque behavior of a two-phase axial flux permanent
magnet synchronous machine,” IEEE Transaction on Magnetics,
Vol. 37, 2803–2805, July 2001.
8. Liu, C. T. and S. C. Lee, “Magnetic field modeling and
optimal operational control of a single-side axial-flux permanent
magnet motor with center poles,”Journal of Magnetism and
Magnetic Materials, Vol. 304, No. 1, 454–456, September 2006.
9. Mahmoudi, A., N. A. Rahim, and W. P. Hew, “Axial-flux
permanent-magnet machine modeling, design, simulation, and
analysis,” Scientific Research and Essay (SRE), Vol. 6, No. 12,
2525–2549, June 2011.
10. Barakat, G., T. El-meslouhi, and B. Dakyo, “Analysis of the
cogging torque behavior of a two-phase axial flux permanent
magnet synchronous machine,” IEEE Transaction on Magnetics,
Vol. 37, No. 4, 2803–2805, July 2001.
Progress In Electromagnetics Research B, Vol. 33, 2011 405

11. Caricchi, F., F. Crescimbini, and O. Honrati, “Modular axial-

flux permanent-magnet motor for ship propulsion drives,” IEEE
Transaction on Energy Conversion, Vol. 14, No. 3, 673–679,
September 1999.
12. Ficheux, R. L., F. Caricchi, F. Crescimbini, and O. Honorati,
“Axial-flux permanent-magnet motor for direct-drive elevator sys-
tems without machine room,” IEEE Transaction on Industry Ap-
plications, Vol. 37, No. 6, 1693–1701, November–December 2001.
13. Gieras, J. F., R. J. Wang, and M. J. Kamper, Axial Flux
Permanent Magnet Brushless Machines, Springer Verla, 2008.
14. Mahmoudi, A., N. A. Rahim, and W. P. Hew, “TORUS and
AFIR axial-flux permanent-magnet machines: a comparison via
finite element analysis,” International Review on Modelling and
Simulations, Vol. 4, No. 2, 624–631, April 2011.
15. Liu, C. T., S. C. Lin, and T. S. Chiang, “On the analytical flux
distribution modeling of an axial-flux surface-mounted permanent
magnet motor for control applications original research,” Journal
of Magnetism and Magnetic Materials, Vol. 282, 346–350,
November 2004.
16. Gholamian, S. A., “Optimum design and manufacturing of axial
flux permanent magnet motor for electric vehicle application,”
Ph.D. Dissertation, K. N. Toosi Univ. Technology, Tehran, Iran,
January 2008.
17. Huang, S., J. Luo, F. Leonardi, and T. A. Lipo, “A general
approach to sizing and power density equations for comparison of
electrical machines,” IEEE Transaction on Industry Applications,
Vol. 34, No. 1, 92–97, January–February 1998.
18. Huang, S., J. Luo, F. Leonardi, and T. A. Lipo, “A comparison of
power density for axial flux machines based on the general purpose
sizing equation,” IEEE Transaction on Energy Conversion,
Vol. 14, No. 2, 185–192, January 1999.
19. Aydin, M., S. Huang, and T. A. Lipo, “Design and 3D
electromagnetic field analysis of non-slotted and slotted TORUS
type axial flux surface mounted permanent magnet disc
machines,” IEEE International Electric Machines and Drives
Conference, January 17–20, 2001.
20. Aydin, M., S. Huang, and T. A. Lipo, “Optimum design and 3D
finite element analysis of nonslotted and slotted internal rotor type
axial flux PM disc machines,” IEEE Power Engineering Society
Summer Meeting, July 15–19, 2001.
21. Rao, S. S., Optimization: Theory and Application, Wiley Eastern
Limited, New Delhi, 1985.
406 Mahmoudi, Rahim, and Ping

22. Rao, S. S., Engineering Optimization: Theory and Practice, Wiley,

1996.
23. Pemha, E. and E. Ngo Nyobe, “Genetic algorithm approach and
experimental confirmation of a laser-based diagnostic technique
for the local thermal turbulence in a hot wind tunnel jet,” Progress
In Electromagnetics Research B, Vol. 28, 325–350, 2011.
24. Gen, M. and R. Cheng, Genetic Algorithms and Engineering
Optimization, Wiley, New York, 2000.
25. Rahim, N. A., W. P. Hew, and A. Mahmoudi, “Axial-flux
permanent-magnet brushless dc traction motor for direct drive of
electric vehicle,” International Review of Electrical Engineering,
Vol. 6, No. 2, 760–769, April 2011.
26. Tokan, F. and F. Gunes, “The multi-objective optimization of
non-uniform linear phased arrays using the genetic algorithm,”
Progress In Electromagnetics Research B, Vol. 17, 135–151, 2009.
27. Goldberg, D. E., Genetic Algorithms in Search, Optimization &
Machine Learning, Addison-Wesley, January 1989.
28. Conn, A. R., N. I. M. Gould, and L. Toint, Ph., “A globally
convergent augmented lagrangian algorithm for optimization
with general constraints and simple bounds,” SIAM Journal on
Numerical Analysis, Vol. 28, No. 2, 545–572, April 1991.
29. Conn, A. R., N. I. M. Gould, and Ph. L. Toint, “A
globally convergent augmented lagrangian barrier algorithm for
optimization with general inequality constraints and simple
bounds,” Mathematics of Computation, Vol. 66, No. 217, 261–288,
January 1997.
30. Üler, G. F., O. A. Mohammed, and C. S. Koh, “Utilizing, genetic
algorithms for the optimal design of electromagnetic devices,”
IEEE Transactions on Magnetics, Vol. 30, No. 6, November 1994.
31. Üler, G. F., O. A. Mohammed, and C. S. Koh, “Design
optimization of electrical machines using genetic algorithms,”
IEEE Transactions on Magnetics, Vol. 31, No. 3, May 1995.
32. Wurtz, F., M. Richomme, J. Bigeon, and J. C. Sabonnadiere, “A
few results for using genetic algorithms in the design of electrical
machines,” IEEE Transactions on Magnetics, Vol. 33, No. 2,
May 1997.
33. Mirzaeian, B., M. Moallem, V. Tahani, and C. Lucas,
“Multiobjective optimization method based on a genetic algorithm
for switched reluctance motor design,” IEEE Transactions on
Magnetics, Vol. 30, No. 3, May 2002.
Progress In Electromagnetics Research B, Vol. 33, 2011 407

34. MATLAB 2010a, User Guide, MathWorks Inc., 2010,

http://www.mathworks.com.
35. Mahmoudi, A., N. A. Rahim, and W. P. Hew, “Analytical
method for determining axial-flux permanent-magnet machine
sensitivity to design variables,” International Review of Electrical
Engineering, Vol. 5, No. 5, 2039–2048, September–October 2010.
36. Vaseghi, B., N. Takorabet, and F. Meibody-Tabar, “Transient
finite element analysis of induction machines with stator winding
turn fault,” Progress In Electromagnetics Research, Vol. 95, 1–18,
2009.
37. Torkaman, H. and E. Afjei, “FEM analysis of angular
misalignment fault in SRM magnetostatic characteristics,”
Progress In Electromagnetics Research, Vol. 104, 31–48, 2010.
38. Torkaman, H. and E. Afjei, “Comparison of two types of dual
layer generator in field assisted mode utilizing 3D-FEM and
experimental verification,” Progress In Electromagnetics Research
B, Vol. 23, 293–309, 2010.
39. Torkaman, H. and E. Afjei, “Magnetio static field analysis
regarding the effects of dynamic eccentricity in switched reluctance
motor,” Progress In Electromagnetics Research M, Vol. 8, 163–180,
2009.
40. Opera Version 14.0 User Guide, Vector Fields, 2011,
http://www.cobham.com.
41. Wang, R. J., M. J. Kamper, and K. V. D. Westhuizen, “Optimal
design of a coreless stator axial flux permanent magnet generator,”
IEEE Transaction on Magnetics, Vol. 41, No. 1, 55–64, January
2005.
42. Saari, J., “Thermal analysis of high-speed induction machines,”
Ph.D. Dissertation, Helsinki Univ. Technology, Helsinki, Finland,
January 1998.
43. Hanselman, D. C., Brushless Permanent Magnet Motor Design,
McGraw-Hill, New York, 1994.