Copyright (c) 2013 IEEE. Personal use is permitted.

For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
> FOR CONFERENCE-RELATED PAPERS, REPLACE THIS LINE WITH YOUR SESSION NUMBER, E.G., AB-02 (DOUBLE-CLICK HERE) <

1
Analytical Design Framework for Torque and Back-EMF
Optimization, and Inductance calculation in Double-Rotor Radial-Flux
Air-Cored Permanent-Magnet Machines

Sajjad Mohammadi
1
, Mojtaba Mirsalim
1,2


1
Electrical Machines and Transformers Research Laboratory, Department of Electrical Engineering, Amirkabir University of
Technology, Tehran 15916, Iran
2
Electrical Engineering Department, St. Marys University, San Antonio, TX 78228, USA

Analytical models have been demonstrated to be effective tools in the analysis of electromagnetic devices by providing fast, yet
accurate solutions. In this paper, a comprehensive analytical framework for double-rotor radial-flux air-cored permanent-magnet
(PM) machines is developed whereby the average as well as the torque ripples, back-electromotive force waveforms, air-gap flux
density distribution, and a number of other characteristics regarding the design of the machine are predicted. Also, closed-form
expressions for inductances of the stator coils are analytically derived. All machine parameters and material properties including iron
saturation are considered in the model. Moreover, an algorithm in order to optimally design the thicknesses of the rotor yokes based
on the utilized steel is presented. Finally, genetic optimizations regarding the enhancement of the produced torque and back-EMF are
performed. It is also shown that the results obtained from the proposed model match well with those issued from FEM.

I ndex Termsfinite element method, genetic optimization, inductance calculation, magnetic equivalent circuit, nonlinear modeling,
permanent-magnet machine, double-rotor radial-flux machine.

I. INTRODUCTION
IR-CORED PM machines enjoy many advantages such as
slot-less structure, zero cogging torque, no stator core
losses, high torque density, and ease of manufacturing.
However, they suffer from some drawbacks such as large
effective air-gap, excessive use of PM materials, low
developed torque and torque ripples, all of which may be the
reason for relatively few works in the area and restricted
applications in higher power levels. Therefore, satisfactory
optimizations and performance estimations toward enhancing
their overall performance would be necessary.
Electric machines may be analyzed by either numerical or
analytical techniques. Numerical methods such as FEM,
although accurate, are too time-consuming to be used in the
optimizations, while analytical approaches by providing fast,
yet accurate solutions are very good compromise between
simplicity and accuracy, making them very useful and
effective in preliminary design stages. Analytical performance
investigation of PM-assisted machines may be performed
based on the solution of Laplaces and Poisons equations [1]-
[6], or by employing magnetic equivalent circuit (MEC)
techniques [7]-[17]. The former, although powerful, is
relatively complicated, incapable of taking the nonlinearities
such as iron saturation into account, and especially having
essential problems with complex structures, while the latter is
usually simpler and effective in many configurations, even
without any symmetry, and able to account for iron saturation
[11]-[15] and all material properties. However, prediction of
the air-gap flux density distribution which is necessary for
determination of either torque ripples or back-EMF harmonics
is a severe problem that arises in MEC-based approaches. All
the literature [8]-[14] account for only the average torque,
except in [15] where the problem is somewhat solved by
employing a standard normal cumulative distribution function
combined with FEM. But, it is not purely-analytical yet, and
the implemented reluctance network is also oversimplified.
FEM-based and analytical analysis of double-rotor radial-
flux air-cored permanent-magnet (RFAPM) machines has
been well-established in [18] and [19], respectively. The latter
is developed on the basis of Laplaces and Poisons equations
analogous to [1]-[6], wherein the iron saturation is ignored, the
utilized steel is considered to be infinitely permeable, relative
recoil permeability of PMs is assumed unity causing the
inability to exactly account for a PM characteristic, and only
the fundamental harmonic of the back-EMF is regarded, all of
which besides limiting the practical design space, have led to
some inaccuracies as well. Comparative analysis of different
winding types of the machine is investigated in [20], wherein
besides the mentioned problems, only the average torque is
accounted. The same attempts for double-rotor iron-cored
machines with toroidal stator coils are done by the edge
element method (EDM) in [21]. Patent [22] is associated with
the introduction and assemblies of a radial-flux, double-rotor,
double-sided yokeless stator comprising modular lamination
stacks. There also exist symmetrical dual-gap (SDG) dual-
stator machines and frameless motors that provide similar
characteristics e.g. high torque density and direct drive
capability [23], [24].
The main contribution of this paper is developing a
comprehensive MEC-based analytical framework for double-
rotor RFAPM machines. In order to achieve a higher accuracy,
especially in the studied machine wherein a large effective air-
gap exists, elements of the model are precisely calculated in
the cylindrical coordinates based on the original circular
structure. The proposed model accounts for iron saturation as
well as all machine dimensions and material properties, and is
capable of accurately predicting the average as well as torque
ripples, back-EMF waveform including all harmonics, and a
number of other design aspects for the sake of practical
analysis. An iterative algorithm aimed at optimally designing
the thicknesses of the rotor yokes is additionally presented.
A
Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
> FOR CONFERENCE-RELATED PAPERS, REPLACE THIS LINE WITH YOUR SESSION NUMBER, E.G., AB-02 (DOUBLE-CLICK HERE) <

2
Moreover, detailed calculations of the inductances of the stator
are also carried out that provides the designers with the
corresponding closed-form expressions. The proposed
approach is then evaluated and confirmed thanks to FEM, and
is finally employed in torque and back-EMF optimizations
through a genetic algorithm, wherein a number of constraints
are also taken into consideration so that a more practical
design could be achieved.
II. MODELING
A. Motor Topology
Fig. 1 illustrates the schematic view of a RFAPM machine
and detailed dimensions of the utilized concentrated non-
overlapping winding wherein Q=(3/4)p [19]. Therein, six
stator coils as well as eight radially-polarized surface-mounted
PMs alternating in the direction of magnetization and placed
on the air-gap sides of the inner and outer rotors can be
observed. The employed winding introduces advantageous
such as short end-turn, less copper losses, simple maintenance
and easy replacement [20]. Associated parameters, that are
assigned based on the engineering data [18], are reported in
Table I. Referring to Fig. 1, the coil centers can be defined as

,
[(2 1) / 2] ; 1,...,
c i c
i i Q u u = = (1)
where, c =2/Q is the coil-pitch angle. Also, centers of the
left and right coil sides are respectively as in below
, , , ,
/ 2 ; / 2
l i c i cs r i c i cs
u u u u u u = + = (2)
At synchronous speed, centers of PMs are
, 0
( ) [(2 1) / 2] ; 1,...,
pm i p m
t i t i p u u e u = + + = (3)
where, 0 is the initial angle.
B. Field Analysis
Flux paths and MEC of one flux loop of the corresponding
electromagnetic system are shown in Fig. 2. Equivalent MMF
of the inner and outer PMs are respectively given by
,
mi pmi c mo pmo c
F h H F h H = = (4)
Although the circuit elements may be determined by
linearizing the configuration along a specific radius such as
mean air-gap radius [11]-[12], but to achieve a higher
accuracy particularly in the presence of the large effective air-
gap herein, they are all determined in the cylindrical
coordinates based on the original circular configuration [10].
The reluctance of the inner PMs can be calculated from
TABLE I
SPECIFICATIONS OF A TYPICAL RFAPM MACHINE
Parameter Value Parameter Value
axial active length, L 100 mm coil-pitch angle, c 60
inner rotor outer radius Rr 100 mm coil side-width angle, w 22
PM height, hpmi and hpmo 7.9 mm coil side-pitch angle, cs 36
air-gap length, g 1 mm copper fill factor, kf 0.4
coil thickness, hc 12.5 mm No. of PM poles, p 8
inner yoke thickness, Lyi 13.37 mm pole-pitch angle, p 45
outer yoke thickness, Lyo 13.54 mm PM to pole arcs ratio, m
0.75
iron grade M19-29G N42 PM remanence, Br 1.3 T
No. of stator coils, Q 6 N42 PM coercivity, Hc -928 kA/m
No. of phases 3 electrical frequency, f 50 Hz
No. of coils per phase, q 2 mechanical speed, m 750 rpm


Fig. 1. The cross section (top) and the detailed dimensions of stator coils
(bottom).

Fig. 2. The flux paths (top) and the corresponding MEC (bottom).
Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
> FOR CONFERENCE-RELATED PAPERS, REPLACE THIS LINE WITH YOUR SESSION NUMBER, E.G., AB-02 (DOUBLE-CLICK HERE) <

3
0
1 r pmi
r
R h
mi
R
r m p
dr
R
r L o u
+
=
}
(5)
and similarly for the outer PMs, they respectively yield to
0
1
1
pmi
mi
r m p r
h
R Ln
L R o u

= +
`
)
(6)
0
1
1
pmo
mo
r m p r pmi e
h
R Ln
L R h g o u


= +
`
+ +

)
(7)
where, r=-Br/0 Hc is the relative recoil permeability of PMs
by which, contrary to the subdomain models [1]-[6], both of
PM characteristics, i.e., remanent Br and coercivity Hc, can
successfully be regarded in the proposed model, and ge=2g+hc
is the effective air-gap. According to Fig. 3(a), half circle-
shaped fringing fluxes are assumed at each side of the air-gap,
with the length of
2 2
( ) ( / 2) | / 2|
f e r pmi e
L r g r R h g = (8)
whereby, the effective air-gap reluctance can numerically be
calculated by means of the trapezoidal integral as in below
0
1
2 ( )
r pmi e
r pmi
R h g
ge
R h
m p f
dr
R
L r L r o u
+ +
+
=
+
}
(9)
The flux tube of the leakage permeance between two
adjacent inner PMs is as in Fig. 3(b). Although in reality, flux
tubes originate from the surface of a PM and end at the surface
of the other one, but in order to avoid extra calculations, these
surfaces are approximated by straight lines laid at an angle of
from the radial direction. Leakage permeance of inner PMs
can then be found by the following
/2
0
0
2 ( )(1 )
e
g
mmi
r pmi m p
dx
P L
x R h x

| o u
=
+ + +
}
(10)
where,
/ 4
arccos
e
r pmi
g
R h
| t
| |
=
|
|
+
\ .
(11)
All of which yield to
0
2 (1 )
[ / 2][2 (1 ) ]
1
( )(1 )
mmi
m p
e m p
r pmi m p
L
P
g
Ln
R h

| o u
| o u
o u
=
+
+

+
`
+

)
(12)
According to Fig. 3(c), magnet to iron leakage permeance
correspond to the inner PMs can be calculated from
1
0
0
2
L
mii
pmi
dx
P L
x h

|
=
+
}
(13)
where, thickness of the flux tube L1 is the minimum of the half
of the inter-polar length and half of the effective air-gap, as
1
( ) min{ / 2, ( )(1 ) / 2}
m e r pmi m p
L g R h o o u = + (14)
Finally, we obtain
0 1
2
1
2
mii
pmi
L L
P Ln
h
|
|


= +
`

)
(15)
For outer PMs, according to Fig. 3(d), magnet to magnet
leakage permeance can be determined through the following
/2
0
0
2 ( )(1 )
e
g
mmo
r pmi e m p
Ldx
P
x R h g x

o u
=
+ + +
}
(16)
where,
/ 4
arccos
e
r pmi e
g
R h g

| |
=
|
|
+ +
\ .
(17)

Fig. 3. The flux tubes for calculating (a) the effective air-gap reluctance, (b)
the inner PM to PM reluctance, (c) the outer PM to PM reluctance, (d) the
inner PM to iron reluctance, (e) the outer PM to iron reluctance, and (f) the
saturable yokes reluctances.
Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
> FOR CONFERENCE-RELATED PAPERS, REPLACE THIS LINE WITH YOUR SESSION NUMBER, E.G., AB-02 (DOUBLE-CLICK HERE) <

4
We finally obtain
0
2 (1 )
[ / 2][2 (1 ) ]
1
( )(1 )
mmo
m p
e m p
r pmi e m p
L
P
g
Ln
R h g

o u
o u
o u
=



+
`
+ +

)
(18)
Similarly, according to Fig. 3(e), magnet to iron leakage
permeance of outer PMs can be determined through
2
0
0
2
L
mio
pmo
dx
P L
x h

=
+
}
(19)
where, thickness of the flux tube L2 is defined as follows
2
min{ / 2, ( )(1 ) / 2}
e r pmi e m p
L g R h g o u = + + (20)
We finally obtain
0 2
2
1
2
mio
pmo
L L
P Ln
h

= +
`

)
(21)
As shown in Fig. 3(f), according to the fact that flux density
within rotor yokes are locally higher behind the inter-polar
regions, at which heavier saturations mainly occur, to attain a
higher accuracy, reluctances of either inner or outer rotor
yokes are assumed to be constituted from three separate
saturable parts, respectively defined as in below
1 2 1 2
2 ; 2
yi yi yi yo yo yo
R R R R R R = + = + (22)
Reluctances behind the inter-polar regions are
1
0 1 1
( / 2)(1 )
i yi m p
yi
ii i
R L
R
A
o u

+
= (23)
1
0 1 1
( / 2)(1 )
o yo m p
yo
io o
R L
R
A
o u

+
= (24)
where, Ri and Ro are respectively bore radii of inner and outer
rotors and, Ai1 and Ao1 are the area through which the
corresponding yoke flux flow, respectively given by
1 1
;
i yi o yo
A L L A L L = = (25)
For components behind PMs, by employing the average
area of the surfaces which flux lines start and end, we obtain
2
0 2 1 2
0.5 ( / 2)
( ) / 2
m p i yi
yi
ii i i
R L
R
A A
o u

+
=
+
(26)
2
0 2 1 2
0.5 ( / 2)
( ) / 2
m p o yo
yo
io o o
R L
R
A A
o u

+
=
+
(27)
where,
2 2
0.5 ; 0.5
i m p r o m p o
A R L A R L o u o u = = (28)
where, ii1, io1, ii2 and io2 denote the relative permeability
of the corresponding iron part that vary with the flux density
inside, and are determined according to the B-H curve of the
utilized steel. Finally, by applying the KVL law to the loops 1
to 5, and the KCL law to the nodes 6 to 9 of the circuit
depicted in Fig. 2, we obtain the system of equations (29)
given at the bottom of the page, from which fluxes flowing
into the circuit elements can be determined.
As shown in Fig. 4, an iterative procedure is used to
determine the permeability of saturable permeances. It starts
by assigning an initial value to the relative permeability of
rotor yokes (unsaturated value herein) in order to determine
the reluctance network and calculate the fluxes. Then, the flux
densities within saturable reluctances, i.e., inner and outer
rotor yokes, are calculated as follows
1 2
1 1 2
;
( ) / 2
yi yi
yi yi
i i i
B B
A A A

= =
+
(30)
1 2
1 1 2
;
( ) / 2
yo yo
yo yo
o o o
B B
A A A

= =
+
(31)
Afterwards, by referring to the B-H curve, new
permeabilities can be derived through the following [10]
( ) ( 1) ( 1) ( ) ( ) ( 1) 1
0
/ ; [ ] [ ]
k k k k k d k d
i y y i i i
B H

= = (32)
where, k refers to the iteration number, 0 is the permeability
of free space, and d is a damping constant set to be 0.1 herein.
The process continues until the following criterion is
separately satisfied for all permeabilities.
( ) ( 1) ( 1)
[ ] /
k k k
i i i
c

s (33)
where, is the termination factor (0.01 herein).
One may assume that all the air-gap flux g flows into the
air-gap through the PM surface, resulting in a square-wave
approximation of the radial component of the air-gap flux
density that may even be independent of radius. But, so as to
well-predict the torque ripples and back-EMF waveform
which are only dependent on the air-gap flux density
waveform in the air-cored machines, especially in the presence
0 0 0 0 0 0 0
0 0 0 0 0 0 0
0 0 0 0 2 0 0
0 0 0 2 0 0 0
2 0 0 0 0 0 0
0.5 0 0.5 0 1 0 1 0 0
0.5 0.5 0 1 0 1 0 0 0
0 0 0.5 0 1 0 0 0 1
0 0.5 0 1 0 0 0 1 0
g mo mio
mi mi mii
mo mio mmo yo
mii mii mmi yi
mio ge mmi mmo
mmi
mmo
yi
yo
R R
R R
R R R
R R R
R R R

(
(
(
(
(

(
(

(

(
(

(
(

0
0
0
0
0
0
0
mi
mo
F
F
( (
( (
( (
( (
( (
( (
( (
=
( (
( (
( (
( (
( (
( (

(29)
Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
> FOR CONFERENCE-RELATED PAPERS, REPLACE THIS LINE WITH YOUR SESSION NUMBER, E.G., AB-02 (DOUBLE-CLICK HERE) <

5
of a large effective air-gap, a more accurate model is
employed by considering the fringing effects. However, the
areas under both approximated curves are still the same. By
assuming ramp functions in the fringing regions located on
both sides of the PM arc, i.e., angles of (1-m)p/2 and
(1+m)p/2, the radial component of the flux density in the air-
gap and the stator regions, delimited by Rr+hpmi and
Rr+hpmi+ge, can be formulated as
1
1 1 2
2 3
4 3 4
4
0 ; 0
( ) ;
2
( , ) ;
( ) ;
2
0 ;
m
f
r m
m
f
p
B
B r B
B
u u
u u u u u
u
u u u u
u u u u u
u
u u u
< <

< <

'
= < <

< <

< <

(34)
where, second and fourth lines correspond to the fringing
regions which are increasing ramps from 0 at 1 and 4 located
outside the PM region, to the value of Bm at 2 and 3 located
inside the PM region, where
1 2
(1 ) / 2 ; (1 ) / 2
m p f m p f
u o u u u o u u = = + (35)
3 4
(1 ) / 2 ; (1 ) / 2
m p f m p f
u o u u u o u u = + = + + (36)
wherein, f is the equivalent fringing angle corresponding to
the fringing length Lf, and Bm is the maximum value of the
flux density in the uniform region , respectively as in below
( ) ( ) /
f f
r L r r u = (37)
( )
g
m
m p
B r
r L

o u
= (38)
In terms of Fourier series, the two-dimensional air-gap flux
density distribution Br(r,) can be written as
1,3,...
( , ) ( , ) sin[ / ]
N
r n p
n
B r B r n n u tu u
=
=

(39)
where,
0
2
( , ) ( , )sin( )
p
n r
p p
n
B r n B r d
u
tu
u u
u u
'
=
}
(40)
C. Torque Calculation
Having the radial component of the flux density within the
stator region, the developed torque of each phase coil can be
found on the basis of Lorentz law, as in below
1
1
1
1
/ 2
/ 2
/ 2
/ 2
[ ( ) ( , )]
[ ( ) ( , )]
r w co
r w ci
l w co
l w ci
R
a f a r
R
R
f a r
R
T L r k J t B r r dr d
L r k J t B r r dr d
u u
u u
u u
u u
u u
u u
+

} }
} }
(41)
2
2
2
2
/ 2
/ 2
/ 2
/ 2
[ ( ) ( , )]
[ ( ) ( , )]
r w co
r w ci
l w co
l w ci
R
b f b r
R
R
f b r
R
T L r k J t B r r dr d
L r k J t B r r dr d
u u
u u
u u
u u
u u
u u
+

} }
} }
(42)
3
3
3
3
/ 2
/ 2
/ 2
/ 2
[ ( ) ( , )]
[ ( ) ( , )]
r w co
r w ci
l w co
l w ci
R
c f c r
R
R
f c r
R
T L r k J t B r r dr d
L r k J t B r r dr d
u u
u u
u u
u u
u u
u u
+

} }
} }
(43)
where, Rci and Rco are inner and outer radii of the stator, kf is
the copper filling factor and, Ja, Jb and Jc are current densities
assumed to be uniformly distributed within the coil area, as
( ) sin(2 )
a m
J t J f t t = + (44)
( ) sin(2 2 / 3 )
b m
J t J f t t t = + (45)
( ) sin(2 2 / 3 )
c m
J t J f t t t = + + (46)
where, Jm=Im/Ac is the current density amplitude in which Ac is
the coil area and, Im, f and are amplitude, electrical
frequency and phase angle of the stator currents, respectively.
Accounting for all stator coils, the total developed torque is
( )
t a b c
T q T T T = + + (47)
Given the details in Appendix I, we obtain
1
1 2
2 3
4
sin( ) sin( )
2 2
{ ( ) cos( ) ( ) cos( )
( ) cos( )} ( , )
co
ci
N
f p
w cs
t
n
p p
c c
a b
p p
R
c
c n
R
p
Lk
n n
T q
n
n n
J t J t
n
J t B r n r dr
u
t u t u
t u u
t u t u
u u
t u
u
=

=
+
+

}
(48)
D. Back-EMF Calculation
Per-turn back-EMF of each phase can be determined based
upon the corresponding flux linkage along with the rotor
motion. While a conductor is located at a the radius of r and
the angle of from the centers of the coil side as shown in
Fig. 5, the associated flux linkage is given by
1
1
( , ) ( , )
l
r
a r
r B r Lr d
u o
u o
o u u
+
+
=
}
(49)
2
2
( , ) ( , )
l
r
b r
r B r Lr d
u o
u o
o u u
+
+
=
}
(50)

Fig. 4. Flowchart of the nonlinear magnetic equivalent circuit (NMEC).
Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
> FOR CONFERENCE-RELATED PAPERS, REPLACE THIS LINE WITH YOUR SESSION NUMBER, E.G., AB-02 (DOUBLE-CLICK HERE) <

6
3
3
( , ) ( , )
l
r
c r
r B r Lr d
u o
u o
o u u
+
+
=
}
(51)
For the sake of more precise estimations, the flux linkage
has been averaged with respect to r and over the coil side
area i.e. throughout where all coil turns are spread, as in below
/ 2
,
/ 2
1
( , )
w co
w ci
R
abc ave abc
R
w c
r dr d
h
u
u
o o
u

=
} }
(52)
By knowing that,
, abc ave
abc
d
E
d

u
= (53)
and, by assuming the coordinate system on the rotor, only c,i
are functions of , where c,i/d=m.
Per-turn back-EMF of phase coils are finally obtained as
1,3,...
1
4
sin( )sin( )
2 2
cos( ) ( , )
co
ci
N
p m
w cs
a
n
w c p p
R
c
n
R
p
L
n n
E
n h
n
B r n r dr
u e
t u t u
t u u u
t u
u
=
=

}
(54)
1,3,...
2
4
sin( )sin( )
2 2
cos( ) ( , )
co
ci
N
p m
w cs
b
n
w c p p
R
c
n
R
p
L
n n
E
n h
n
B r n r dr
u e
t u t u
t u u u
t u
u
=
=

}
(55)
1,3,...
3
4
sin( ) sin( )
2 2
cos( ) ( , )
co
ci
N
p m
w cs
c
n
w c p p
R
c
n
R
p
L
n n
E
n h
n
B r n r dr
u e
t u t u
t u u u
t u
u
=
=

}
(56)
E. Design Considerations
As shown in Fig. 6, to avoid saturation and consequent
problems as well as efficiently utilizing the iron material, an
iterative procedure is presented to design the yoke thicknesses
so that they magnetically operate at the knee point of the B-H
curve denoted by Bknee. Hence, in each iteration yoke
thicknesses are updated based on the flux density error as
( ) ( 1) ( 1)
1
m m m
yi yi yi
L L B q

= A (57)
( ) ( 1) ( 1)
1
m m m
yo yo yo
L L B q

= A (58)
where, is the update coefficient selected to be 5. The process
continues until the stop criterion is reached for both Lyi and
Lyo, meaning that yoke flux density behind the inter-polar
regions is Bknee.
It is essential to limit the field intensity inside inner and
outer PMs to avoid irreversible demagnetizations due to the
temperature rise or the reaction fields, respectively defined as
1
mi mi
c r
H B
H B
= (59)
1
mo mo
c r
H B
H B
= (60)
where, Bmi and Bmo are the corresponding flux densities.
To avoid excessive use of PM material, two other
considerations should also be regarded to limit PM-to-PM as
well as PM-to-iron leakage fluxes, respectively as in below
( / 2)(1 )
r pmi e m p e
R h g g o u + + > (61)
2
pm e
h g > (62)
III. MODEL EVALUATION
In this section, characteristics of the studied machine obtained
by the proposed model, are evaluated and compared with
those issued from FEM as well as a model analogous to the
existing model [19] wherein iron saturation is neglected, its
permeability is considered infinity, and relative recoil
permeability of PMs is assumed unity. Number of considered
harmonics N is also set to 101. Fig. 7 shows the B-H
characteristic of the utilized steel. Fig. 8 illustrates the
magnetic flux density distribution and flux lines for both no-

Fig. 5. Dimensions of the coil sides and a single coil-turn inside.

Fig. 6. Flowchart of the optimal design of the rotor yoke thicknesses.

Fig. 7. B-H characteristic of the utilized steel with grade M19-29G.
Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
> FOR CONFERENCE-RELATED PAPERS, REPLACE THIS LINE WITH YOUR SESSION NUMBER, E.G., AB-02 (DOUBLE-CLICK HERE) <

7
load and full load conditions. It is seen that due to the reaction
fluxes developed by the stator currents under the latter,
although not remarkable because of a very large effective air-
gap herein, some distortions in the flux distribution has been
appeared. It is also worth noting that as a result of the optimal
design of the rotor yoke thicknesses, flux densities within the
portions behind inter-polar regions corresponding to Ryi1 and
Ryo1 equal to the knee point of the B-H curve, i.e. 1.7 T, and
the adjacent iron portion corresponding to Ryi2 and Ryo2 operate
at lower flux densities, as expected. A close agreement
between analytically calculated radial component of the flux
density distribution in the air-gap as well as the stator region,
and that obtained from FEM, can be observed in Fig. 9.
Justified by the half circle-shaped fringing fluxes defined by
(8), it can be seen that the fringing effects are zero above PM
surfaces, i.e., r= Rr+hpmi and r= Rr+hpmi+ge, and reaches to
the maximum at the middle of the stator region, i.e., r=
Rr+hpmi+ge/2. The developed torque (for a load angle of 4
mechanical degrees) and per-turn back-EMF of the machine
versus rotor angle are shown in Fig. 10. Once again, the
results closely agree with FEM, with a maximum discrepancy
less than 2 %. It is seen that the angular period of torque
ripples is 360/LCM(p,Q) =15, where LCM(p,Q) is the least
common multiple of p and Q. Torque versus load-angle
characteristic, and the maximum torque versus current density
within stator coils are shown in Fig. 11, which again correlate
well with FEM by an error less than 3 %. All these
overestimated results are mainly due to approximations in
establishing the proposed approach. The ratio of Hmi/Hc and
Hmo/Hc are also acceptable values of 0.64 and 0.7. Compared
to the existing approach, it can be observed that the proposed
model more accurately predicts the performance of the
machine, as well as incorporates all machine parameters and
material properties.

Fig. 8. Magnetic flux density distribution and flux lines at no-load condition
(top) and full-load condition i.e. load angle of 22.5 and Jm=5 A/mm
2

(bottom).

Fig. 11. Torque versus load-angle characteristic (left), and maximum torque
versus stator current density curve (right).

Fig. 10. The torque (top) and per-turn back-EMF (bottom) of the machine.

Fig. 9. Radial component of the flux density distribution in the air-gap and
the stator region calculated by FEM (top), and analytical model (bottom).

Fig. 12. Variations of the average torque versus yoke thickness (left) and
PM coercivity (right).
Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
> FOR CONFERENCE-RELATED PAPERS, REPLACE THIS LINE WITH YOUR SESSION NUMBER, E.G., AB-02 (DOUBLE-CLICK HERE) <

8
As shown in Fig. 12, it is evident that inaccuracy problems
of the existing model become even more severe, e.g., under
heavier iron saturations, in the cases of utilizing smaller yoke
thicknesses for purposes such as active material minimization,
and adoption of PMs with lower coercivities as well.
IV. OPTIMIZATION
A genetic algorithm, whose parameters are reported in
Table II, is employed to find the optimal dimensions of PMs
and stator coils which are chosen as design variables herein. A
number of design constraints should also be placed on the
procedure so as to achieve a more practical design, as
addressed in Table III. A new restriction is the bound of the
average torque required in the application, and the other is the
equality of field intensity within inner and outer PMs.
Compared to the case of equal PM heights, the latter
restriction guarantees an equal demagnetization withstand
capability for both inner and outer PMs that avoid a magnetic
stress on one side, while the same torque can be produced.
Leakage fluxes are limited by (61) and (62), and moreover, the
thicknesses of the rotor yokes are optimally selected based on
the presented algorithm.
Table IV reports the design parameters, corresponding
search domains, optimal values and obtained characteristics
for two independent optimizations: first, maximization of the
ratio of the average torque Tave to torque ripple, and second,
minimization of the back-EMF THD. Fig. 13 presents the
developed torques and half cycle of back-EMF waveforms for
both cases as well as the initial machine. It is clear that torque
ripples are efficiently reduced in the former from 0.92 N.m of
the initial machine to a very small value of 0.05 N.m.
However, a more sinusoidal back-EMF waveform is obtained
through the latter, whereby its THD is reduced from 0.0305 to
the appreciable value of 0.0026. Harmonic contents of the
mentioned characteristics are also shown in Fig. 14. As
expected, multiples of the third harmonic do not appear in the
electromagnetic torque of a three-phase driven machine, and
others are more efficiently reduced in the first optimization,
while back-EMF harmonics associated with the second
scheme have been reduced more efficiently, wherein it can
also be observed that more PM material should be utilized,
which results in thicker rotor yokes as well. Equal field
intensity within inner and outer PMs is also obtained that
provides a better demagnetization withstand capability.
However, in order to achieve a satisfactory performance for
both characteristics as well as other indices such as consumed
active materials and production cost, a tradeoff may be
performed through multi-objective optimizations.
V. CONCLUSION
In this paper, an analytical model of RFAPM machines in
terms of all design parameters, i.e. geometrical dimensions
and material properties, was developed, wherein accurate
calculations in the cylindrical coordinates were performed,
resulting in a rather high accuracy. The proposed approach is

Fig. 14. Harmonic content of the developed torque (top) and per-turn back-
EMF waveform (bottom).

Fig. 13. The developed torque (left) and per-turn back-EMF waveform
(right) of the optimized machines.
TABLE III
DESIGN CONSTRAINTS
Quantity Limit Quantity Limit
Taverage 13< Tave <14 cs and th th cs
max Hmi= Hmo 0.75 Hc cs and th cs+ th c
Lyi , Lyo optimal, Fig. 6 leakage fluxes (61) and (62)

TABLE IV
OPTIMIZATION RESULTS FOR DIFFERENT SCHEMES
Parameter Min Max
Optimized values
T-ripple E-THD
pm
0.4 1 0.854 0.858
cs 1 60 36.183 30.002
th 1 30 15.975 17.92
hpmi , hpmo 4 15 11 , 7.8 12.3 , 9
hc 5 15 14.4 12.8
Obtained Characteristics
torque ripple (N.m) 0.05 0.361
back-EMF THD 0.0973 0.0026
Lyi , Lyo (mm) 15.46 , 17.72 17.03 , 19.35
Hmi/Hc , Hmo/Hc 0.61 , 0.61 0.58 , 0.58

TABLE II
GENETIC ALGORITHM PARAMETERS
Quantity value Quantity value
number of genes 5 population size 50
number of generations 100 elite individuals 2
probability of crossover 0.75 probability of mutations 0.25


Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
> FOR CONFERENCE-RELATED PAPERS, REPLACE THIS LINE WITH YOUR SESSION NUMBER, E.G., AB-02 (DOUBLE-CLICK HERE) <

9
also capable of accounting for iron saturation, as well as well-
estimation of the air-gap flux density distribution, both
average and torque ripples, back-EMF waveforms, and a
number of other design-oriented characteristics. Moreover, a
number of considerations toward more practical designing of
the device have been analytically presented. An iterative
algorithm aimed at the optimal design of the rotor yokes based
upon the utilized steel was additionally presented. Also, self-
inductances of the stator coils are analytically calculated and
the corresponding closed-form expression are derived. The
results obtained from the proposed model were in a close
agreement with those issued from FEM, and were much more
accurate and effective compared to the existing approach. It
was shown that the framework can advantageously be
employed in the preliminary design and analysis of the
machine throughout a complete search space. Although a
concentrated non-overlapping winding was utilized in this
study, the model is sufficiently general to be used in the
studies of the other winding types. Finally, the genetic
optimizations were carried out whereby the developed torque
and back-EMF waveforms of the machine were effectively
enhanced.
The authors are now working on multi-objective design
optimization of the device by incorporating all machine
parameters and other performance indices, which will be
presented in another occasion.
APPENDIX I
CALCULATIONS OF THE DEVELOPED TORQUE
Herein, details for derivation of the developed torque are
presented. By substituting (39) into (41)-(43), we achieve
1
1
1
1
/ 2
1,3,...
/ 2
2
/ 2
1,3,...
/ 2
2
( ) ( , ) sin( )
( ) ( , ) sin( )
r w co
r w ci
l w co
l w ci
R
N
a f a n
n
p R
R
N
f a n
n
p R
n
T Lk J t B r n
r dr d
n
Lk J t B r n
r dr d
u u
u u
u u
u u
t u
u
u
t u
u
u
+
=

+
=

=
`

)

`

)

} }

} }

(63)
2
2
2
2
/ 2
1,3,...
/ 2
2
/ 2
1,3,...
/ 2
2
( ) ( , ) sin( )
( ) ( , ) sin( )
r w co
r w ci
l w co
l w ci
R
N
b f b n
n
p R
R
N
f b n
n
p R
n
T Lk J t B r n
r dr d
n
Lk J t B r n
r dr d
u u
u u
u u
u u
t u
u
u
t u
u
u
+
=

+
=

=
`

)

`

)

} }

} }

(64)
3
3
3
3
/ 2
1,3,...
/ 2
2
/ 2
1,3,...
/ 2
2
( ) ( , ) sin( )
( ) ( , ) sin( )
r w co
r w ci
l w co
l w ci
R
N
c f c n
n
p R
R
N
f c n
n
p R
n
T Lk J t B r n
r dr d
n
Lk J t B r n
r dr d
u u
u u
u u
u u
t u
u
u
t u
u
u
+
=

+
=

=
`

)

`

)

} }

} }

(65)
Substituting the place of the integrals and the summations,
executing the integrations and simplifying, yields
{ }
2
1,3,...
1 1
2 ( )
( , )
sin( ) sin( ) sin( )
2
co
ci
N
R
f p a
a n
R
n
w l r
p p p
Lk J t
T B r n r dr
n
n n n
u
t
t u t u t u
u u u
=
=

`

)

}

(66)
{ }
2
1,3,...
2 2
2 ( )
( , )
sin( ) sin( ) sin( )
2
co
ci
N
R
f p b
b n
R
n
w l r
p p p
Lk J t
T B r n r dr
n
n n n
u
t
t u t u t u
u u u
=
=

`

)

}

(67)
{ }
2
1,3,...
3 3
2 ( )
( , )
sin( ) sin( ) sin( )
2
co
ci
N
R
f p c
c n
R
n
w l r
p p p
Lk J t
T B r n r dr
n
n n n
u
t
t u t u t u
u u u
=
=

`

)

}

(68)
By substituting l,i and r,i in terms of cs and c,i as in (2),
and perform simple mathematical manipulations, we obtain
the total developed torque given in (48).
APPENDIX II
CALCULATIONS OF THE BACK-EMF
Details for derivation of the back-EMF expressions are
provided herein. Substituting (39) into (49)-(51) yields
1
1
1,3,...
( , ) ( , ) sin( )
l
r
N
a n
n
p
n
r B r n L r d
u o
u o
t u
o u
u
+
=
+


=
`

)

}

(69)
2
2
1,3,...
( , ) ( , ) sin( )
l
r
N
b n
n
p
n
r B r n L r d
u o
u o
t u
o u
u
+
=
+


=
`

)

}

(70)
3
3
1,3,...
( , ) ( , ) sin( )
l
r
N
c n
n
p
n
r B r n L r d
u o
u o
t u
o u
u
+
=
+


=
`

)

}

(71)
Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
> FOR CONFERENCE-RELATED PAPERS, REPLACE THIS LINE WITH YOUR SESSION NUMBER, E.G., AB-02 (DOUBLE-CLICK HERE) <

10
Following the replacement of the integrals and the
summations, as well as execution of the integrals, we reach
1,3,...
1 1
( , )
( , )
cos[ ( )] cos[ ( )]
N
p n
a
n
l r
p p
r L B r n
r
n
n n
u
o
t
t t
u o u o
u u
=

=


+ +
`

)

(72)
1,3,...
2 2
( , )
( , )
cos[ ( )] cos[ ( )]
N
p n
b
n
l r
p p
r L B r n
r
n
n n
u
o
t
t t
u o u o
u u
=

=


+ +
`

)

(73)
1,3,...
3 3
( , )
( , )
cos[ ( )] cos[ ( )]
N
p n
c
n
l r
p p
r L B r n
r
n
n n
u
o
t
t t
u o u o
u u
=

=


+ +
`

)

(74)
By substituting l,i and r,i in terms of cs and c,i as in (2),
and performing simple mathematical manipulations, we obtain
the following relationships for a coil turn placed at positions r
and as depicted in Fig. 5.

1,3,...
1
2 ( , )
( , ) sin( )
2
( )
sin[ ]
N
p n
cs
a
n
p
c
p
r L B r n
n
r
n
n
u
tu
o
t u
t u o
u
=

(75)
1,3,...
2
2 ( , )
( , ) sin( )
2
( )
sin[ ]
N
p n
cs
b
n
p
c
p
r L B r n
n
r
n
n
u
tu
o
t u
t u o
u
=

(76)
1,3,...
3
2 ( , )
( , ) sin( )
2
( )
sin[ ]
N
p n
cs
c
n
p
c
p
r L B r n
n
r
n
n
u
tu
o
t u
t u o
u
=

(77)
Now, in order to achieve more accurate estimations, the
above equations are averaged with respect to r and over the
coil side area through (52), all of which results in
,
1,3,...
/ 2
1
/ 2
/ 2
1
/ 2
2
sin( ) ( , )
2
{ sin( ) cos( )
2
cos( ) sin( ) }
2
co
ci
w
w
w
w
N
R
p
cs
a ave n
R
n
w c p
c
p p
c
p p
L
n
B r n r dr
n h
n n
d
n n
d
u
u
u
u
u
t u

t u u
t u t
o o
u u
t u t
o o
u u
=

}
}
}

(78)
,
1,3,...
/ 2
2
/ 2
/ 2
2
/ 2
2
sin( ) ( , )
2
{ sin( ) cos( )
2
cos( ) sin( ) }
2
co
ci
w
w
w
w
N
R
p
cs
b ave n
R
n
w c p
c
p p
c
p p
L
n
B r n r dr
n h
n n
d
n n
d
u
u
u
u
u
t u

t u u
t u t
o o
u u
t u t
o o
u u
=

}
}
}

(79)
,
1,3,...
/ 2
3
/ 2
/ 2
3
/ 2
2
sin( ) ( , )
2
{ sin( ) cos( )
2
cos( ) sin( ) }
2
co
ci
w
w
w
w
N
R
p
cs
c ave n
R
n
w c p
c
p p
c
p p
L
n
B r n r dr
n h
n n
d
n n
d
u
u
u
u
u
t u

t u u
t u t
o o
u u
t u t
o o
u u
=

}
}
}

(80)
In each relationship, executing those integrations which are
respect to , and knowing that the second integral terms are
zero, respectively yields
,
1,3,...
1
2 2
sin( )sin( )
2 2
sin( )sin( ) ( , )
2
co
ci
N
p p
w cs
a ave
n
w c p p
R
cs c
n
R
p p
L
n n
n h n
n n
B r n r dr
u u
t u t u

t u t u u
t u t u
u u
=

}
(81)
,
1,3,...
2
2 2
sin( )sin( )
2 2
sin( )sin( ) ( , )
2
co
ci
N
p p
w cs
b ave
n
w c p p
R
cs c
n
R
p p
L
n n
n h n
n n
B r n r dr
u u
t u t u

t u t u u
t u t u
u u
=

}

(82)
,
1,3,...
3
2 2
sin( )sin( )
2 2
sin( )sin( ) ( , )
2
co
ci
N
p p
w cs
c ave
n
w c p p
R
cs c
n
R
p p
L
n n
n h n
n n
B r n r dr
u u
t u t u

t u t u u
t u t u
u u
=

}

(83)
Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
> FOR CONFERENCE-RELATED PAPERS, REPLACE THIS LINE WITH YOUR SESSION NUMBER, E.G., AB-02 (DOUBLE-CLICK HERE) <

11
Finally, by differentiating the above flux linkage equations
with respect to as in (53), we obtain the back EMF closed-
form expressions given in (54)-(56).
APPENDIX III
CALCULATION OF THE STATOR INDUCTANCES
Because of symmetry, self-inductances of all phases are
equal and hence, determination of the inductance of only one
phase would be sufficient. Rotor PMs do not contribute to the
coil self-inductances and thus, are not regarded herein. Fig. 15
shows the distribution of flux lines due to the current within
the coils of one phase of the device. The distributions of the
flux lines and magnetic flux density, magnetic field intensity,
and coenergy in the linearized configuration are provided in
Fig. 16, wherein only the region associated with one flux loop
is considered, i.e. one coil side and the area in which the
corresponding flux lines flow. As seen in Fig. 16(b) and (c),
due to the high relative permeability of the iron parts, the
associated magnetic flux intensity and coenergy are almost
vanishing and can be ignored in the inductance calculations.
Based on Fig. 16, three flux tubes can finally be considered as
depicted in Fig. 17, forming the inductance components Ls1,
Ls2 and Ls3. It can be observed that the flux lines
corresponding to the first path, link all coil turns and close
their path through the rotor back-irons. Those of the second
one, close their path through the rotor irons again, but do not
link all coil turns due to crossing of the coil area. Also, the
third flux tube illustrates the flux paths of those that close their
path in the air-gap. As depicted, two horizontal axes x1 and x2
are defined to simplify the calculation of the left and right
sides of the flux tubes, respectively.
The total inductance is then defined as below
1 2 3
2 ( )
s s s s
L q L L L = + + (84)
where, factors 2 and q are respectively to account for the other
coil-side and all series coils in one stator phase. The coil side-
pitch, coil side-width, and the phase coil-pitch are respectively
as in the following
cs av cs
R t u = (85)
w av w
R t u = (86)
[2 / ]
q av
R q t t = (87)
where, Rav is the mean radius given below
/ 2
av r pmi c
R R h g h = + + + (88)
Since the relative recoil permeability of PMs is almost
unity, it is clear that the total air-gap seen by the armature
fluxes includes the mechanical clearances, the PM areas and
the stator region, with the length given below

Fig. 16. Magnetic domain associated with one coil side in the linearized
configuration: (a) magnetic flux lines and flux density distribution, (b)
magnetic field intensity distribution, and (c) coenergy distribution.

Fig. 17. The flux tubes corresponding to the self-inductance components
associated with one coil side.

Fig. 15. Magnetic flux lines due to excitation of one phase.
Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
> FOR CONFERENCE-RELATED PAPERS, REPLACE THIS LINE WITH YOUR SESSION NUMBER, E.G., AB-02 (DOUBLE-CLICK HERE) <

12
2
t pmi pmo c
g h h g h = + + + (89)
Each of the mentioned inductance components can be
calculated through the coenergy approach given below
2 2
0
/
i i
volume
L H dV i =
}}}
(90)
where, Hi is the field intensity within the corresponding flux
tube. The other way aimed at determining each inductance
term is through the associated flux linkage , as in below
/
i i
L i = (91)
A. Calculations of Ls1
The flux tube associated with Ls1 is depicted in Fig. 17.
Magnetic field intensity and stored energy in the iron parts are
very small and can be ignored due to the associated high
relative permeability, as revealed earlier. The Amperes
circuital law is now applied to the corresponding closed path
C shown in Fig. 18(a), as in below
.
enclosed
C
H dl I =
}
(92)
or
I I
l t r t
H g H g Ni + = (93)
where, Hl
I
and Hr
I
are respectively field intensities in the left
and right air-gap parts of the associated flux tube. Based upon
the fact that the total flux flowing within the flux tube is
constant, the other required equation is obtaining by putting
the total fluxes of the left and right sides equal, as in below
0 0 1
( ) / 2
I I I I
l r l cs w r
H L H L t t t = = (94)
Solving the system of equations of (93) and (94), yields
1
1
[ ( ) / 2]
I
l
t cs w
N i
H
g
t
t t t
=
+
(95)
1
[ ] / 2
[ ( ) / 2]
I cs w
r
t cs w
N i
H
g
t t
t t t

=
+
(96)
Finally, Ls1 is calculated by substituting (95) and (96) into
(90) as in the following
{
}
2 1
2
/ 2 ( )/ 2
2
1 0 1
0 / 2
/ 2
2 2
0 2
0 / 2
[ ]
[ ] /
t w cs
t w
t
t
L g
I
s l
g
L g
I
r
g
L H dx dy dz
H dx dy dz i
t t
t
t t
t

=
+
} } }
} } }
(97)
By executing the integrations and simplifying, we achieve
2
0 1
1
1
( ) / 2
( ) / 2
cs w
s
t cs w
N L
L
g
t t t
t t t

=
+
(98)
The second approach to determine Ls1 is through the
associated flux linkage
I
. The differential flux within the
element dx1 at position x1 shown in Fig. 19(a) is as follows
1
I I
l
d B Ldx = (99)
where, Bl
I
is the magnetic flux density within the left side,
given below
0 1
0
1
[ / 2 / 2]
I I
l l
t cs w
N i
B H
g
t

t t t
= =
+
(100)
Therefore, the associated flux linkage would be
I I
d N d = (101)
Finally, the total flux linkage can be obtained by
( )/ 2
2
0 1
1
1
[ ( ) / 2]
w cs
w
I
t cs w
L N i
dx
g
t t
t
t

t t t
+
=
+
}
(102)
whereby, Ls1 can again be achieved through (91). It is also
worth noting that the same calculations could be done based
on the right portion of the flux tube.
An alternative approach to determine Ls1 is to employ the
following well-known relationship
2
1
/
I
s
L N R = (103)
where, R1 is the total reluctance seen by the associated MMF
source. Ignoring the iron reluctances, R1 is apparently divided
into the following components corresponding to the left and
right air-gap parts, respectively.
0
( ) / 2
t
l
cs w
g
R
L t t
=

(104)

Fig. 18. The closed paths of the Apmeres law for calculations of the self-
inductance components (a) Ls1, (b) Ls2, and (c) Ls3.
Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
> FOR CONFERENCE-RELATED PAPERS, REPLACE THIS LINE WITH YOUR SESSION NUMBER, E.G., AB-02 (DOUBLE-CLICK HERE) <

13
0 1
t
r
g
R
L t
= (105)
Knowing that,
I
l r
R R R = + (106)
The inductance expression (98) can similarly be achieved
by substituting (106) into (103).
B. Calculations of Ls2
The flux tube associated with Ls2 is depicted in Fig. 17.
Applying the Amperes circuital law given in (92) to the
corresponding closed path C depicted in Fig. 18(b), yields
1
1 2
( ) ( )
II II
l t r t
w
x
H x g H x g N i
t
+ = (107)
where, Hl
II
and Hr
II
are respectively field intensities in the
left and right air-gap parts of the associated flux tube. It is
seen that the total current enclosed in the mentioned path is
not constant, resulting in the field intensities to be functions of
positions x1 and x2. Something like (94), fluxes within the
corresponding flux elements located at positions x1 and x2 are
equal, as in below
1
| |
x
II II
l x r x
d d = (108)
Simplifying the above relationship results in
0 1 1 0 2 2
( ) ( )
II II
l r
H x Ldx H x Ldx = (109)
It is evident from Fig. 18(b) that if the left side of a flux line
is at the position x1=20, the right side will apparently be at the
position x2=0, and similarly for the pair x1=w and x2=2. The
following relationship generally exists
2
2 1 0
0
( 2 )
2
w
x x
t
t
t t
=

(110)
Thus,
2
2 1
0
2
w
dx dx
t
t t
=

(111)
Substitution of the above equation into (109) yields
2
1 2
0
( ) ( )
2
II II
l r
w
H x H x
t
t t
=

(112)
Solving the system of equations of (107) and (112), and
using (110), the field intensity functions within the two
mentioned regions are obtained as in the following
2
1 1
2 0
( )
[ 2 ]
II
l
t w w
N i
H x x
g
t
t t t t
=
+
(113)
0 0
2 2 0
2 0 2
( 2 ) 2
( ) 2
( 2 )
II w w
r
t w w
Ni
H x x
g
t t t t
t
t t t t t

= +
`
+
)

(114)
Finally, Ls2 can be calculated by substituting (113) and
(114) into (90) as in the following
{
}
0
2
/ 2
2
2 0 1
0 / 2 2
/ 2
2 2
0 2
0 / 2 0
[ ]
[ ] /
t w
t
t
t
L g
II
s l
g
L g
II
r
g
L H dx dy dz
H dx dy dz i
t
t
t

=
+
} } }
} } }
(115)
Executing the above integrations and simplifying, yields
2 3 3
0 2 0
2 2
2 0
( 8 )
3 ( 2 )
w
s
t w w
N L
L
g
t t t
t t t t

=
+
(116)
Alternatively, Ls2 can be determined through the
corresponding flux linkage. It is evident that the magnetic flux
density within the left side is as in below
0 2
1 0 1 1
2 0
( ) ( )
[ 2 ]
II II
l l
t w w
N i
B x H x x
g
t

t t t t
= =
+
(117)
Thus, the differential flux within the element dx1 at the
position x1 depicted in Fig. 19(b) is as follows
1 1
( )
II II
l
d B x Ldx = (118)
Therefore, the flux linkage of the element would be
1
{ }
II II
w
x
d N d
t
= (119)
wherein, substitution of (118) yields the total flux linkage as in
below

Fig. 19. The differential fluxes associated with the assumed flux tubes: (a)
L1, (b) L2, and (c) L3.
Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
> FOR CONFERENCE-RELATED PAPERS, REPLACE THIS LINE WITH YOUR SESSION NUMBER, E.G., AB-02 (DOUBLE-CLICK HERE) <

14
0
2
2 0 2
1 1 2
2 0 2
[ 2 ]
w
II
t w w
L N i
x dx
g
t
t
t

t t t t
=
+
}
(120)
Now, executing the above integration, (116) can again be
obtained through (91).
C. Calculations of Ls3
The flux tube associated with Ls3 is depicted in Fig. 17,
based on which applying the Amperes circuital law given in
(92) to the path C depicted in Fig. 18(c), yields
1 0
1
2( )
( ) (2 2 )
III
c
w
x
H x h r N i
t
t
t

+ = (121)
where, H
III
is the field intensity in the flux tube, and r=x10 is
the radius of the top and bottom half-circles. It is seen that the
magnetic field intensity is constant along the closed path C. It
is finally obtained as follows
1 1 0
1 0
( ) ( )
[ ( )]
III
w c
N i
H x x
h x
t
t t t
=
+
(122)
Finally, Ls3 could be determined by substituting (112) into
(90) as in the following
{
}
0
0
0
0
/ 2 2
2
3 0 1
0 / 2
2
2 2
0 1 0 1
0 0
2 [ ]
2 [ ] ( ) /
c
c
L h
III
s
h
L
III
L H dx dy dz
H x dx d dz i
t
t
t t
t

t u

=
' +
} } }
} } }

(123)
wherein, the first term corresponds to both the left and right
rectangles, and the second accounts for both the top and down
half-circles in the cylindrical coordinates (r=x1). Finally, Ls3 is
obtained as
2 2
0 0 0
3 0 2 2
2
ln(1 )
2
c c
s
w c c
N h L h
L
h h
tt t t
t
t t t

= + + +
`
)

(124)
The associated flux linkage is another way to calculate Ls3.
It is clear that the magnetic flux density within the tube is
defined as in below
0
1 0 1 1 0
1 0
( ) ( ) ( )
[ ( )]
III III
w c
N i
B x H x x
h x

t
t t t
= =
+

(125)
whereby, the differential flux in the element dx1 at the position
x1 depicted in Fig. 19(c) is as in below
1 1
( )
III III
d B x Ldx = (126)
Therefore, the flux linkage of the element would be
1 0
2( )
{ }
III III
w
x
d N d
t

t

= (127)
wherein, substitution of (126) yields the total flux linkage as
0
0
2
2 0
1 0 1 2
1 0
2
( )
[ ( )]
III
w c
N i
x dx
h x
t
t

t
t t t
=
+
}
(128)
Now, executing the above integration, Ls3 given in (124)
can similarly be obtained through (91). It is also worth noting
that Ls3 is very small and can be ignored since the
corresponding fluxes close their path through the air-gap, and
the linked number of turns is small.
D. Calculations of associated parameters
The regions of the discussed flux tubes, i.e. the parameters
0, 1 and 2 must be determined herein. It is clear that the flux
lines tend to flow through a path with the minimum reluctance
and therefore, 0 illustrates where the flux lines tend to close
their path in the air-gap instead of the back-irons, and can thus
be determined by assuming equal lengths of the flux paths for
the two regions at the boundary (x1=20). Since flux paths in
region III consist of two half-circles, plus two straight lines
each having length hc, we have
0
2 2 2
II III t c
l l g h tt = = + (129)
From which we obtain
0
( ) /
t c
g h t t = (130)
From Fig. 16(a), it is clear that the flux lines of the return
path on the right side (those of the first and the second
regions) are uniformly distributed, i.e. there exists the same
magnetic flux density in the both regions. Thus, the ratio of 1
to 2 is found to be proportional to the ratio of the net fluxes
flowing into these areas, as in below
1 2
/ /
I II I II
r r
B B t t = = (131)
where,
I
and
II
are respectively total fluxes within the first
and the second flux tubes that can be obtained through (99)
and (118), respectively. For the former we have
( )/ 2
0 1 1
1
[ ( ) / 2]
w cs
w
I
t cs w
Ni Ldx
g
t t
t
t

t t t
+
=
+
}
(132)
Executing the integration yields

Fig. 20. (a) The assumed dimensions of the end-turn of a coil and (b) the
differential fluxes associated with the corresponding flux tube.
Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
> FOR CONFERENCE-RELATED PAPERS, REPLACE THIS LINE WITH YOUR SESSION NUMBER, E.G., AB-02 (DOUBLE-CLICK HERE) <

15
0 1
1
( ) / 2
[ ( ) / 2]
I cs w
t cs w
N i L
g
t t t

t t t

=
+
(133)
For the latter, we similarly obtain
0
0 2
1 1
2 0 2
[ 2 ]
w
II
t w w
Ni L
x dx
g
t
t
t

t t t t
=
+
}
(134)
which simplifies as in below
2 2
0 2 0
2 0
( 4 )
2 [ 2 ]
II w
t w w
Ni L
g
t t t

t t t t

=
+
(135)
It is also evident that
1 2 r
t t t + = (136)
where,
/ 2 ( ) / 2
r q cs w
t t t t = + (137)
Finally, by solving the system of equations (131) and (136),
1 and 2 are calculated as in the following
2 2
0 0
1 2
0
( )[0.5 2 ( 2 )]
4
cs w w w r
w cs
t t t t t t t
t
t t t
+ +
=
+
(138)
{
}
{ }
2
2 0
2 2
0 0
2
0
4 ( )
[0.5 2 ( 2 )]
/ 4
r r w cs cs w
w w r
w cs
t t t t t t t t
t t t t t
t t t
= +
+ +
+
(139)
E. Calculations of the End-Turn Inductances
The end-turn leakage inductances could be approximated by
considering half circle-shaped end-turns [16] with circular
cross sections that contains the same number of turns as the
original coil, as shown in Fig. 20. The radius of the assumed
circular coil R1 is then defined such that the coil areas are still
the same, as in below
1
/
c w
R h t t = (140)
For the sake of simplified calculations, the end turns can be
considered as infinitely long cylinders with length D, about
which the magnetic fields exist out to a radius of R2,
respectively given as in the following
2
/ 2 ; / 2
cs cs
D R tt t = = (141)
It is evident that the magnetic field intensity distribution
about the mentioned cylinder can be calculated through the
Amperes law as in the following
1 2
( ) ;
2
e
N i
H r R r R
r t
= < < (142)
Finally, the end-turn leakage of a coil can be determined
through (90), as in below
{ }
2
1
2 2
0
0 0
2 [ ] /
D R
e e
R
L H r dr d dz i
t
u =
} } }
(143)
wherein, a factor of 2 is employed to account for both the top
and bottom end-turns. By integrating, we obtain
2
0 2
1
[ ]
2
e
N D R
L Ln
R

t
= (144)
The other approach to achieve the above equation is through
the associated flux linkage. The magnetic flux density within
the associated flux tube is as in below
0
0 1 2
( ) ( ) ;
2
e e
N i
B r H r R r R
r

t
= = < < (145)
Thus, the differential flux within the element dr at position r
is defined as in below
( )
e e
d B r Ddr = (146)
Therefore, the flux linkage of the element would be
e e
d N d = (147)
in which, substitution of (146) yields the total flux linkage as
2
1
2
0
2
R
e
R
N i D
dr
r

t
=
}
(148)
Following the calculation of e, Le given in (144) can again
be obtained by (91).
End-turn leakage inductance can also be determined
through the following relationship
2
e e
L N P = (149)
where, Pe is the total permeance seen by the associated fluxes,
calculated by
2
1
0 0 2
1
[ ]
2 2
R
e
R
Ddr D R
P Ln
r R

t t
= =
}
(150)
F. Evaluations of the Inductance Calculations
Finally, a FEM whose meshed regions are shown in Fig. 21,
is utilized to evaluate the obtained closed-form expressions.
Assuming that the number of turns per coil is 100, it is seen
that the analytical calculations well-agree with those extracted
from FEM, respectively 3.5 mH and 3.6 mH, i.e. with an error
less than 3%.
REFERENCES
[1] T. Lubin, S. Mezani, and A. Rezzoug, Two-Dimensional Analytical
Calculation of Magnetic Field and Electromagnetic Torque for Surface-

Fig. 21. Meshed regions of the FEM employed for calculating the self-
inductance of the stator coils.
Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
> FOR CONFERENCE-RELATED PAPERS, REPLACE THIS LINE WITH YOUR SESSION NUMBER, E.G., AB-02 (DOUBLE-CLICK HERE) <

16
Inset Permanent-Magnet Motors, IEEE Trans. Magn., vol. 48, no. 6,
pp. 20802091, Jun. 2012.
[2] T. Lubin, S. Mezani, and A. Rezzoug, 2-D exact analytical model for
surface-mounted permanent-magnet motors with semi-closed slots,
IEEE Trans. Magn., vol. 47, no. 2, pp. 4794929, Feb. 2011.
[3] A. H. Isfahani, Analytical Framework for Thrust Enhancement in
Permanent-Magnet (PM) Linear Synchronous Motors With Segmented
PM Poles, IEEE Trans. Magn., vol. 46, no. 4, pp. 11161122, Jan.
2010.
[4] S. Y. Sung, J. H. Jeong, Y. S. Park, J. Y. Choi, S. M. Jang, and A.
Rezzoug, Improved Analytical Modeling of Axial Flux Machine With a
Double-Sided Permanent Magnet Rotor and Slotless Stator Based on an
Analytical Method, IEEE Trans. Magn., vol. 48, no. 11, pp. 2945
2948, Nov. 2012.
[5] L. J. Wu, Z. Q. Zhu, D. Staton, M. Popescu, and D. Hawkins, An
Improved Subdomain Model for Predicting Magnetic Field of Surface-
Mounted Permanent Magnet Machines Accounting for Tooth-Tips,
IEEE Trans. Magn., vol. 47, no. 6, pp. 16931704, Jun. 2011.
[6] O. de la Barrire, H. Ben Ahmed, M. Gabsi, and M. LoBue, Two-
Dimensional Analytical Airgap Field Model of an Inset Permanent
Magnet Synchronous Machine, Taking Into Account the Slotting
Effect, IEEE Trans. Magn., vol. 49, no. 4, pp. 14231435, Apr. 2013.
[7] Mohammadi S., Mirsalim M., Double-sided permanent-magnet radial-
flux eddy-current couplers: three-dimensional analytical modeling, static
and transient study, and sensitivity analysis, IET Electr. Power Appl., to
be published.
[8] M. Bash and S. Pekarek, Modeling of salient-pole wound-rotor
synchronous machines for population-based design, IEEE Trans.
Energy Convers., vol. 26, no. 2, pp. 381392, June 2011.
[9] B. N. Cassimere, S. D. Sudhoff, and D. H. Sudhoff, Analytical design
model for surface-mounted permanent-magnet synchronous machines,
IEEE Trans. Energy Convers., vol. 24, no. 2, pp. 338346, Jun. 2009.
[10] J.A. Krizan, S.D. Sudhoff, A Design Model for Salient Permanent-
Magnet Machines With Investigation of Saliency and Wide-Speed-
Range Performance, IEEE Trans. Energy Convers., vol. 28, no. 1, pp.
95105, Mar. 2013.
[11] Y. Kano, T. Kosaka, and N. Matsui, Simple nonlinear magnetic
analysis for permanent-magnet motors, IEEE Trans. Ind. Appl., vol. 41,
no. 5, pp. 12051214, 2005.
[12] M. Hsieh and Y. Hsu, A generalized magnetic circuit modeling
approach for design of surface-permanent magnet machines, IEEE
Trans. Ind. Electron., vol. 59, no. 2, pp. 779792, Aug. 2012.
[13] B. Sheikh-Ghalavand, S. Vaez-Zadeh and A. H. Isfahani, An Improved
Magnetic Equivalent Circuit Model for Iron-Core Linear Permanent-
Magnet Synchronous Motors, IEEE Trans. Magn., vol. 46, no. 1, pp.
112120, Jan. 2010.
[14] Y. Liu, Y. Zhu, J. Yang, and B. Chen, Optimization of Voice Coil
Motor to Enhance Dynamic Response Based on an Improved Magnetic
Equivalent Circuit Model, IEEE Trans. Magn., vol. 47, no. 9, pp.
22472251, Sep. 2011.
[15] E. Kazan, and E. Onat, Modeling of Air Core Permanent-Magnet
Linear Motors With a Simplified Nonlinear Magnetic Analysis, IEEE
Trans. Magn., vol. 47, no. 6, pp. 17531762, Jun. 2011.
[16] J.F. Gieras, Permanent Magnet Motor Technology. 3rd ed., New York:
Taylor & Francis, 2010.
[17] D. C. Hanselman, Brushless Permanent Magnet Motor Design. 2nd ed.,
USA: Magna Physics Publishing, 2006.
[18] J. A. Stegmann, and M. J. Kamper, Design Aspects of Double-Sided
Rotor Radial Flux Air-Cored Permanent-Magnet Wind Generator,
IEEE Trans. Ind. Appl., vol. 47, no. 2, pp. 767778, Mar./Apr. 2011.
[19] P. J. Randewijk, and M. J. Kamper, Analytical Analysis of a Radial
Flux Air-Cored Permanent Magnet Machine with a Double-Sided Rotor
and Non-overlapping Double-layer Windings, in Proc. ICEM,
Marseille, France, pp. 11781184, 2012.
[20] M. J. Kamper, J. H. J. Potgieter, J. A. Stegmann, and P. Bouwer,
Comparison of Air-Cored and Iron-Cored Non-Overlap Winding
Radial Flux Permanent Magnet Direct Drive Wind Generators, in Proc.
ECCE, AZ, USA, pp. 16201627, 2011.
[21] R. Wojciechowski, C. Jedryczka, A. Demenko, and E. A. Mendrela, 3D
Edge Element Calculations of Electrical Motor with Double Cylindrical
Rotor, Stud. in Comput. Intell., vol. 119, pp. 147153, 2008.
[22] Y. Qu, P. L. Jansen, B. S. Bagepalli, J. R. J. Carl, A. D. Gadre, and F. J.
Lopez, Electrical Machines and Assemblies Including a Yokeless
Stator with Modular Lamination Stacks, U.S. Patent 0 131 985, Jan. 22,
2006.
[23] Symmetrical Dual Gap (SDG) Construction of Synchronous Electric
Machines [Online]. Available: http://www.sineton.com
[24] High Torque Density Motors [Online]. Available: http://www.phase.eu




Sajjad Mohammadi was born in Kermanshah, Iran, in 1989. He received the
B.S. degree (First Honor) in electrical engineering from the Kermanshah
University of Technology, Kermanshah, Iran, in 2011. He is currently
pursuing the M.S. degree at the Electrical Machines and Transformers
Research Laboratory (EMTRL), electrical engineering department, Amirkabir
University of Technology (Tehran Polytechnic), Tehran.
He has attained a number of ranks and awards as: the 2
nd
place in National
Chem-E-Car Competitions 2009 held by Sharif University of Technology,
Tehran, the 5
th
place in 2
nd
International Chem-E-Car Competitions 2009 held
by McGill University, Montreal, Canada, the 2
nd
place for the Humanoid
Robot League together with the 1
st
place from Technical Challenge both in
International Iran-Open 2010 Robotics Competitions & Symposium, Tehran.
Also, the 3
rd
place in National Chem-E-Car Competitions 2010 held by Razi
University, Kermanshah, and the 1
st
place for the Humanoid Robot League
together with the 1
st
place from Technical Challenge both in International
AUTCUP 2010 Robotics Competitions and the 12
th
Khwarizmi Youth Award
(KYA) Robotics held by Amirkabir University of Technology, Tehran. He is
also the recipient of several Distinguished Student Awards from Kermanshah
University of Technology. His current research interests include design,
modeling, FE analysis, optimization, prototyping, real-time monitoring, DSP-
based control and drive of rotary and linear electric machines, power
electronics, and renewable energies http://ele.aut.ac.ir/~emtrl .




Mojtaba Mirsalim (SM04) was born in Tehran, Iran, on February 14, 1956.
He received the B.S. degree in EECS/NE and the M.S. degree in nuclear
engineering from the University of California, Berkley, in 1978 and 1980,
respectively, and the Ph.D. degree in electrical engineering from Oregon State
University, Corvallis, in 1986.
Since 1987, he has been at Amirkabir University of Technology, where he
has served five years as the Vice Chairman and more than seven years as the
General Director in Charge of Academic Assessments, and currently is a Full
Professor in the Department of Electrical Engineering where he teaches
courses and conducts research in energy conversion, electrical machine
design, and hybrid vehicles, among others. His special fields of interest
include the design, analysis, and optimization of electric machines, FEM,
renewable energy, and hybrid vehicles. He is the author of more than 160
international journal and conference papers and four books on electric
machinery and FEM. He is the founder and director of the Electrical Machines
and Transformers Research Laboratory http://ele.aut.ac.ir/emtrl . Currently, he
is a professor in the engineering department at St. Marys University.