See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/344663309

Modeling and Optimizing Method for Axial Flux Induction Motor of Electric
Vehicles

Article  in  IEEE Transactions on Vehicular Technology · October 2020

DOI: 10.1109/TVT.2020.3030280

CITATIONS READS
17 693

4 authors, including:

Jie Mei Yuefei Zuo

Massachusetts Institute of Technology Nanyang Technological University
33 PUBLICATIONS   385 CITATIONS    67 PUBLICATIONS   490 CITATIONS   

SEE PROFILE SEE PROFILE

Christopher Ho Tin Lee

Nanyang Technological University
211 PUBLICATIONS   2,027 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Gearless Motor View project

Direct-Drive Motor Design View project

All content following this page was uploaded by Jie Mei on 23 March 2021.

The user has requested enhancement of the downloaded file.

 Modeling and Optimizing Method for Axial Flux
Induction Motor of Electric Vehicles
Jie Mei, Student Member, IEEE, Yuefei Zuo, Member, IEEE, Christopher H. T. Lee, Senior Member,
IEEE, and James L. Kirtley, Life Fellow, IEEE

Abstract—axial flux induction motors have attracted attention kS Stator winding factor
in electric vehicles due to their advantages over conventional Lslot Rotor bar inductance per unit length
motors, including higher efficiency, compact structure, high
utilization of materials, and good ventilation and cooling. This lend_o Length of end winding per turn at outer radius
paper provides a fast design method for two-stator-one-rotor lend_i Length of end winding per turn at inner radius
axial flux induction motor of electric vehicle applications. The lw Length of armature winding per phase
proposed method consisting of an accurate motor analytical
model and a design variables optimization method based on MSteel Total stator and rotor steel weight
genetic algorithm. Unlike traditional design of axial flux MCu Total winding copper weight
induction motor with single objective, multi-objective is MAl Total rotor bars and rings aluminum weight
considered in this work. Based on the proposed method, the
performance of axial flux induction motor can be simulated and
m Number of stator slots per pole per phase.
optimized in much shorter time compared with finite element NS Number of turns per phase
analysis in ANSYS Maxwell. The comparison results from NR Number of rotor slots
ANSYS Maxwell can prove the effectiveness and accuracy of the
NSIP Number of strands in parallel
proposed method, and the performance of the final designed
motor can meet all the design requirements. NSP Number of short-pitched slots for stator winding
NQ Population size
Index Terms—Axial flux induction motor, optimization, motor p Pole pair number
design, electric vehicles, finite element analysis, genetic algorithm.
Pin Motor input power
NOMENCLATURE Pmax Maximum output mechanical power
Pe Elitist percentage
Aw Winding area
Ps Probability to be selected for crossover
c Gene
Pm Gene mutation probability
c’ Mutated gene
R1 Stator branch equivalent resistance
dw Wire diameter
R2 Rotor branch equivalent resistance
ds Stator slot depression height
Rslot Rotor bar resistance per unit length
fOB1_MAX Maximum motor weight in current population
ro Outer stator radius
fOB1_MIN Minimum motor weight in current population
ri Inner stator radius
fOB2_MAX Highest motor efficiency in current population
s Per-unit slip
fOB2_MIN Lowest motor efficiency in current population
Tring Length of rotor rings
g Effective motor air-gap length
Tiron Thickness of stator back iron
Gmax Maximum generation number
t Current generation number
hr Rotor bar height
us Stator slot depression width
hs Stator slot height
V Magnitude of input voltage
IS Current in stator branch of equivalent circuit
wr Rotor bar width
IR Current in rotor branch of equivalent circuit
ws Stator slot width
W1 Linear weight of motor weight
Copyright (c) 2015 IEEE. Personal use of this material is permitted. W2 Linear weight of motor efficiency
However, permission to use this material for any other purposes must be
obtained from the IEEE by sending a request to [email protected]. Xm Magnetic reactance
J. Mei and J. L. Kirtley are with the Department of Electrical Engineering X2 Rotor leakage reactance
and Computer Science, Massachusetts Institute of Technology, Cambridge,
MA 02139 USA (e-mails: [email protected]; [email protected]). X1 Stator leakage reactance
Y. Zuo and C.H.T. Lee are with the School of Electrical & Electronic Xel Stator end winding leakage
Engineering, Nanyang Technological University, Singapore (e-mails:
[email protected]; [email protected]) Xsl Stator slot leakage
Xbl Stator belt leakage high-power applications had been neglected. Thus, those
Xxl Stator zigzag leakage modeling works are not very accurate when it comes to AFIM
µo Permeability constant of EVs. 3) The mentioned modeling works for AFIMs has not
done any further optimization, so it is difficult to guarantee
σr Rotor bar electrical conductivity
that the generated designs are theoretically optimal, especially
σ Winding electrical conductivity
the design problem is non-linear and has multiple objectives.
ζ Stator slot space factor Considering all those facts, a new AFIM design method
ω Frequency of input voltage consisting of accurate analytical modeling and further
α Short pitch angle optimization for today’s EVs is highly expected.
Λ Chromosome Genetic algorithm (GA), which is a great tool for solving
multi-objective complex and non-linear optimization problems,
I. INTRODUCTION is adopted in this paper. GA imitates the process of species
lectric vehicles (EVs) are growing as a proportion of the evolution in nature. Based on the laws of natural selection and
E commercial market over the past decades with the purpose survival of the fittest, solution candidates evolve over
of reducing pollution and improving the driving experience [1]. generations. Comparing with other classic solving tools for
The core of EV is electric motor, which converts electrical non-linear optimization problems, it shows a fast convergence
energy into mechanical energy to drive the wheels [2]. speed, and has been proven theoretically and empirically to
Comparing with classic internal combustion engines, the provide powerful search functions in complex spaces [8].
electric motors have the advantages of being more powerful, Based on those merits, GA has been extensively applied to
more energy saving, and more compact. To further improve solve many non-linear multi-objective optimization problems,
the energy efficiency of modern EVs, it is desirable to develop for examples, in biology [9]-[11], transportation [12]-[14], and
lighter and more efficient electric motors. Among all the economics [15], [16]. In motor design field, GA has also been
prevalent motors of EV applications, axial flux induction applied to optimize surface-mounted permanent magnet motor
motor (AFIM) has attracted more attention for its special [17], DC motor [18], interior permanent magnet motor [19],
features, which include better power-to-weight ratio, compact switched reluctance motor [20], etc.
structure, higher efficiency, especially in multi-pole motors, This paper provides a fast design method, consisting of an
high utilization of materials, and better ventilation and cooling accurate analytical model and a GA-based design variables
[3], [4]. Designing AFIMs with multiple objectives is a non- optimization method, for AFIM of EV applications. The
linear and complex optimization problem subject to special contributions of the proposed method are threefold: 1) an
design specifications. The design variables are coupled and accurate analytical model of two-stator-one-rotor AFIM is
changing one variable may result in great changes in output proposed, which can provide close results to FEA in ANSYS
values and violating design constraints. Traditionally, AFIMs Maxwell in much shorter simulation time. 2) By applying the
are designed in software such as ANSYS Maxwell through GA, the AFIM design can be further optimized to meet all the
finite element analysis (FEA), where the design constraints design specifications while achieving great motor performance.
and initial specifications are verified based on the given design 3) Different from traditional AFIM design process with single
variables. However, such FEA-based design methods are very design objective, this work considers multiple objectives, thus
inefficient and time-consuming, since it usually takes hours to the design trade-off is compromised and balanced. To the
simply simulate one possible design candidate, especially authors’ knowledge, this is the first detailed paper describing
considering the fact that axial-structure motors are difficult to GA-based fast design method for AFIM of EV applications.
perform 2D simulation thus usually done in 3D FEA. The rest of this paper is organized as follows: in Section II,
To solve the time-consuming problem of FEA simulation, the analytical model and optimization formulation of the
many researchers have proposed equivalent analytical models AFIM are introduced. Section III presents the GA-based
for AFIMs, expecting to generate results close to FEA in much optimization solution method for finding the optimal design
shorter time. Authors in [5] proposed an analytical model for variables. FEA results of magnetic flux density and motor
single-phase AFIMs. For three-phase AFIM, [6] proposed performance in ANSYS Maxwell are presented and compared
modeling method for single-sided AFIMs. For this type of with the results from the proposed method in Section IV.
motor, the unbalanced attractive force in the axial direction Finally, Section V provides conclusions.
between the rotor and stator may cause extra mechanical
losses. Thus in [7], for the purpose of reducing axial forces, II. ANALYTICAL MODEL OF AFIM
researchers have studied a dual-rotor symmetrical AFIM and The AFIM of EV applications in this paper has a structure
proposed corresponding analytical model. In general, there are of distributed-winding and two-stator-one-rotor, as shown in
essentially three main problems of those mentioned modeling Fig.1, where the three-phase coils are drawn in red, blue, and
works: 1) single-phase AFIMs and single-sided AFIMs are yellow, respectively, the rotor are drawn in light grey in the
more suitable for low-power applications, and they could not middle, and the two stators on two sides are in dark grey. The
meet the power requirements for today’s EVs. 2) All those stators and windings are made of steel and copper,
existing modeling works were not designed for EV respectively. The outer and inner rings, and bars of the rotor
applications. Some components that cannot be ignored at such are made of aluminum with steel filled between the bars. The
analytical model of the AFIM and corresponding optimization After considering the rotor ring effect, (5) can be modified as:
formulation are introduced in the remainder of this section.  2 ( ro − ri ) Tring p 2 
12 N S2 k S2 (ro − ri )
R2 = Rslot 1 +  (7)
NR  N r ( ro + ri ) wr 

For the stator winding resistance R1, the length and area of the
armature winding for each phase can be first calculated as:
lw = 2 N S (ro − ri ) + N S lend _ o + N S lend _ i (8)

Aw = d w2 N SIP (9)
4
Then R1 can be expressed as:
lw
R1 = (10)
Fig.1. Appearance of the designed AFIM from different angles.  Aw
A. AFIM Analytical Model Stator flux leakage has several components. Each of these
The analytical equivalent circuit model for each phase of components should be added to the calculation of leakage
the two-stator-one-rotor AFIM is shown in Fig. 2 below. impedance X1. The main components of these leakage include
slot, zigzag, belt, and end winding leakages. The stator slot
leakage can be calculated with the same formula for traditional
radial-structure induction motor as in (11) [21]:
1 hs d s N S2  2 5 N SP 
X sl = 2 ( ro − ri ) o ( + )  m − 4 m2  (11)
3 ws us p  
Similar with (2), we can define ns+ and ns-, and corresponding
winding factors. Then, the stator zigzag leakage components
caused by air-gap space harmonics of order ns = Nslots/p ±1 can
be expressed as:
3 4 o N S2 ro2 − ri 2  kns+ kns− 
X zl =   +  (12)
Fig.2. Per-phase analytical equivalent circuit model of the AFIM. 2 g 2  ( N slots + p ) 2
( N slots − p ) 2 
In the analytical model, the rotor reactance component X2 can Likewise, the stator belt leakages are caused by harmonic
be calculated as: orders of 5 and 7. If there is no rotor coupling, the belt
12 N S2 k S2 (ro − ri ) 3o N S2 k S2 ( ro2 − ri 2 ) leakages can be calculated as (13) by replacing n with 5 and 7:
X2 =  Lslot +   ( N R ) (1) 3 4 o N S2 kn2 ro2 − ri 2
NR g X agn =  2 2 (13)
2 n p g 2
where ξ(NR) is a constant function of NR, and given rotor slot
height hr and width wr, the rotor slot inductance per unit length The harmonics of the belt linked to the rotor appear to be in
Lslot can be evaluated as (3): parallel with the components of rotor impedance. Similarly, if
rotor resistance is ignored, these components can be written as:
1 1
 ( NR ) = +
12 N S2 kn2 (ro − ri ) Lslot 3o N S kn ( ro − ri )
(2) 2 2 2 2
( NR + p) ( NR − p)
2 2
X 2, n = +  ( N R ) (14)
NR g
hr
Lslot = o (3)
1 1
3wr  ( NR ) = + (15)
( N R + np ) ( N R − np )
2 2
The magnetic reactance Xm can be calculated as:
3 4 o N S2 k S2 ro2 − ri 2 The belt leakage can then be calculated as Xbln = Xagn//X2,n. In
Xm =  (4) addition, modified from Alger’s method for radial-structure
2  p2 g 2
induction motor in [22], the stator end winding leakage Xel in
The rotor equivalent resistance R2 is expressed as: axial case can be estimated as:
12 N S2 k S2 (ro − ri ) 210 o N S2 ro − ri
R2 = Rslot (5) X el =  ( − 0.3) (16)
NR 16 2 p 2 2
where the rotor slot resistance per unit length Rslot can be Finally, the total stator branch reactance X1 due to the flux
evaluated as (6): leakage components can be generated as:

Rslot =
1
(6) X 1 = X sl + X el + X bl + X zl (17)
wr hr  r
B. Optimization Model for AFIM Design be greater than 250 N-m as in constraint (30). In addition, the
Based on the proposed analytical model and the given range of discrete and continuous design variables can be found
design requirements, the AFIM design can be formulated to an in Table I and II. It should be mentioned that in order to keep
optimization problem as follows: normal motor operation temperature, stator slot space factor ζ
is required to be less than 0.6 so that enough space can be left
OB1: min{M Steel + M Al + M Cu } (18)
for liquid cooling in the final motor product, combining with
OB2 : max{Pmax / Pin } (19) the maximum stator winding current density constraint in (25).
subject to TABLE I. RANGE OF DISCRETE VARIABLES
C1: NR wr < 2πri (20) Design Variable Range
C2: Nslots ws < 2πri (21) p {2, 3, 4}
m {1, 2, 3}
C3: 2⋅(Tiron +hs +ds )+hr +g < 0.15 (22) NR {Nslot, Nslot – 1,…, Nslot/2}
dw {1.15, 1.29, 1.45, 1.63} mm
Vro
C4: 2 (23) NSIP {1, 2,…, 7}
k S N s (ro2 − ri 2 )Tiron NSP {1, 2}
2 ri pV TABLE II. RANGE OF CONTINUOUS VARIABLES
C5: 2 (24)
k S N S  (ro2 − ri 2 ) ( 2 ri − N slot ws ) Design Variable Min Max
−6 Tring 0.90 cm 1.30 cm
4  I S , Pmax 10
C6:  25 (25) Tiron 1.50 cm 3.00 cm
N SIP d w2 V 145 V 163 V
ω 850 rad/s 1200 rad/s
C7: M Steel + M Al + M Cu  26 (26) g 0.80 mm 1.00 mm
ro 8.50 cm 10.00 cm
C8: Pmax / Pin  81% (27) ri 4.50 cm 7.00 cm
C9: Pmax  80000 (28) hr 1.00 cm 2.00 cm
wr 1.00 cm 1.40 cm
C10: Pmax  3100 (29) hs 2.30 cm 3.80 cm
ws 0.35 cm 0.65 cm
p 2 R2 ζ 0.45 0.60
C11: 6 I R , Pmax  250 (30)
 s us 0.15 cm 0.35 cm
ds 0.50 mm 0.80 mm
As in (18) and (19), the formulated optimization has two
objectives (OBs). The first is that the total motor weight is
III. DESIGN OPTIMIZATION BASED ON GA
expected to be as small as possible, which can be calculated
by the sum of the steel stator and rotor weight, MSteel, the This Section presents the GA-based solution method for the
aluminum rotor bars and rings weight, MAl, and the copper design optimization problem in Section II. GAs belong to a
stator winding weight, MCu. Constraint (C) (26) requires that larger class of evolutionary algorithms that use knowledge
the total motor weight to be less than 26 kg. The second inspired by genetic processes (such as selection, crossover,
objective is that the motor efficiency at the maximum output and mutation) that occur in nature to generate solutions to
mechanical power Pmax should be as high as possible, which optimization problems [23], [24]. Fig.3 summarizes the
can be expressed as the ratio of the maximum output general optimization process using a GA. As shown in the
mechanical power to the input power. Constraint (27) requires flowchart, a GA starts off with a population of randomly
that the highest efficiency of each valid design should be generated design candidates (individuals), whose encoded
greater than 81%. Motor dimensions are limited in (20)-(22), string (chromosome) contains information about the variables
which require enough space to be reserved for both the stator to the design problem, and takes place successive iterations
and rotor slots at inner radius, and the motor length should be (generations) that evolves toward better design solutions over
less than 15 cm. The maximum magnetic flux density in the time [25]. Every individual representing a potential design
stator back iron and tooth are required to be less than 2 T as candidate for the problem and is rated through an associated
(23) and (24), respectively. Using circuit theory, the phase fitness function to determine which of them are used to form
current in stator branches (stator windings), the current in new population during the competition, which is called
rotor branch, and the output mechanical power and torque can selection. Besides, in each generation, selected individuals are
be easily calculated based on the proposed analytical model. modified to form a new population by applying genetic
Constraint (25) requires that the stator winding current density operators as crossover and mutation. The new population is
in each phase should be less than 25 A/mm 2. The maximum then used for the next iteration of the algorithm. The algorithm
output mechanical power is required to be greater than 80 kW terminates when the maximum number of generations or a
as in (28) and when it is reached, the corresponding rotor satisfactory average fitness level is reached for the population
spinning frequency, defined as the rated rotation speed, should [26]. In the rest of this section, the operators and element
be less than 3,100 RPM as in (29) and the torque is required to components in the proposed GA-based method for solving the
AFIM optimal design problem are introduced.
 C. Selection
Selection operator helps to preserve the chromosomes with
higher fitness value by assigning them greater chance to pass
their genes to the next generation in crossover, so that the
population can continue to evolve towards optimal solutions.
The selection consists of two parts: probability distribution
and chromosomes sampling for crossover. Elitist is firstly
applied to preserve the Pe percentage chromosomes with the
highest fitness value. The copies of those chromosomes go
directly to the next generation without involving subsequent
steps. Besides, the Pe percentage chromosomes with the
lowest fitness value will be removed from the current
population. By adopting elitist, best design candidates can
always be preserved from the current generation to the next,
while the worst genes are removed. After the elitist step, for
each remaining chromosome Λi, the probability to be selected
for crossover process later, ps(Λi), is evaluated as:
f ( Λi )
ps ( Λi ) = Fit
(32)

NQ
Fig.3. Flowchart of a general genetic algorithm.
j =1
f Fit ( Λj )
A. Chromosome Representation Then based on the probabilities, roulette wheel sampling
The chromosome representation of design candidates in the method [27] is applied to select chromosomes. The population
population is shown in Fig.4. The size of the chromosome is after the elitist step is mapped onto a roulette wheel, where
the same as the total number of design variables, and each each type of chromosome Λi is represented by a space that
gene (slot on the chromosome) represents a design variable. proportional to ps(Λi). By repeatedly rotating the roulette
Each chromosome consists of 20 genes that represent the wheel, chromosomes can be selected for crossover until the
values of 20 design variables randomly selected from their number of all available positions is reached, which is (1-Pe)NQ.
domains. The first 14 gene slots are reserved for continuous By selection, chromosomes with the high fitness value tend to
design variables, and the last 6 gene slots are for discrete produce more offspring than those with below-average fitness.
design variables. During evolution process, the values of these D. Crossover
genes are limited to corresponding design variable domains.
Different design candidates can share their information
through the crossover operator. The selected chromosomes
from the selection operator will be randomly paired as parent
Fig.4. Chromosome representation for the AFIM candidate (1×20 array). chromosomes and exchange their gene parts to generate two
offspring chromosomes. The two-point crossover is applied in
B. Fitness Evaluation the proposed GA here. Suppose Λp1 = (c 1 , c 2 , …, c n) and Λp2
1 1 1

= (c 1 , c 2 , …, c n) are two paired chromosomes with n genes in

2 2 2
A key issue in GA programming is the calculation of
chromosome fitness value so that design candidates can be the crossover step. Two indexes i ∈ {1, 2, …, n-1} and j ∈
evaluated. Chromosomes with higher fitness value will have {i+1, i+2, …, n} are randomly chosen and two offspring
greater chance to get selected by the selection operator in the chromosomes are generated as (33) and (34). This process is
following process and pass their genes to the next generation also demonstrated with Fig. 5.
Λo1 = ( c11 ,..., ci1−1 , ci2 ,..., c 2j , c1j +1 ,..., c1n )
in the step of crossover. Given the chromosome Λ of a (33)
candidate AFIM design, denote the total motor weight value
of the first objective function (18) as fOB1(Λ) and the motor Λo2 = ( c12 ,..., ci2−1 , ci1 ,..., c1j , c 2j +1 ,..., cn2 ) (34)
efficiency value of the second objective function in (19) as
fOB2(Λ). If any of the constraints from (20)-(30) are violated,
the chromosome fitness value will be assigned zero so that the
gene values that cause such violation will not be passed to the
next generation. If no constraints are violated, the fitness
function is evaluated as (31). In this function, the two
objective functions are weighted based on the selection of W1
and W2, which can be adjusted according to preferences.
f OB1_ MAX − fOB1 ( Λ) fOB 2 ( Λ) − f OB 2 _ MIN
f Fit ( Λ) = W1 + W2 (31)
f OB1_ MAX − f OB1_ MIN f OB 2 _ MAX − f OB 2 _ MIN
Fig.5. Two-point crossover demonstration with random indexes i and j.
E. Mutation efficiency over 93% at the maximum output power, which is
Mutation provides genetic diversity to the population by about 12% higher than the initial design (1st generation).
restoring lost or unexplored genetic material, enabling GA to Compared with the traditional FEA-based motor design
search for a broader space and prevent GA from converging to methods, simulation time and design process are greatly
a sub-optimal solution. Each gene after the crossover step has reduced. The proposed method only takes an average of 30
the probability of Pm to mutate. Suppose ci ∈ [ai, bi] is a gene seconds to terminate and output the optimal AFIM designs,
representing a continuous design variable, its mutated version while FEA 3D simulation usually takes a few hours to test one
ci’ can be generated as (35) below: possible AFIM design at a fixed frequency. The AFIM design
with the highest fitness value in the last generation is selected
c −  ( t , bi − ci ) if  = 0

ci =  i (35) for further FEA analysis in ANSYS Maxwell to verify the
 ci −  ( t , ci − ai ) if  = 1
 accuracy of the proposed analytical model, where the design
where τ is either one or zero with equal probability, and variables can be found in Table IV. Besides, Fig. 7 plots the
motor weight and efficiency of the AFIM designs with the
 t  


1−  highest fitness for different W1 and W2 (W1 = 0.1, 0.2, ..., 0.9,
 ( t , y ) = y  1 −   max  
G
(36)
  W2 = 1 - W1) as red dots and interpolated with blue line to form
  the Pareto front. Based on user’s preference, values of W1 and
where δ is a random number with an interval [0, 1] and β is a W2 can be easily adjusted to find the suitable motor with ideal
user-defined variable that can adjust the dependency on the performance, which fully reflects GA’s flexibility in solving
generation number. If the gene to be mutated represents a this multi-objective optimization problem of AFIM design.
discrete design variable, it is assigned a random value from its TABLE III. GA PARAMETERS
discrete domain. The population that has undergoes the
crossover and mutation process, combined with the copy of Parameter Value
the elitist design candidates, together constitute the new NQ 500
generation with the same population size. When the maximum Pe 0.15
Pm 0.05
generation number Gmax or a satisfactory average fitness level W1 0.6
is reached, the algorithm terminates. W2 0.4
Gmax 50
IV. SIMULATION RESULTS AND ANALYSIS β 5
In this simulation, the proposed fast AFIM design method
is firstly run to show the corresponding optimization process
and the improvement of time efficiency. In the second part of
this section, a design is selected from the last generation as
suggested by the GA and corresponding results from ANSYS
Maxwell and the analytical model are analyzed and compared
to show the accuracy of the proposed model.
A. GA Optimization Path
Table III shows the tuning parameters of the GA. The
corresponding objective weight coefficients are set to be 0.6
and 0.4. The GA starts with a randomly generated design that
meets all design constraints with population size of 500. For
the algorithm termination, either maximum generation number
Gmax = 50 is reached, or the motor efficiency and weight of the Fig.6. Optimization process of the proposed GA-based optimization algorithm.
top 5 designs with the highest fitness value in current
generation has not improved comparing with those in the last
generation, the algorithm is done.
The proposed GA terminated in the 38th generation. In Fig.
6 below, we plotted the motor weight and efficiency at the
maximum output power as hexagrams for the top 5 design
candidates with the highest fitness values in the 1 st (turquoise),
5th (green), 10th (yellow), 15th (orange), 20th (red), 25th (pink),
30th (purple), 35th (light blue), and 38th (dark blue) generation
to show the evolution process of motor optimization. Some
design candidates in a generation are the same, so they overlap
on the figure. In Fig. 6, it shows that the design candidates are
optimize themselves towards lighter and more efficient. In the
last generation, all the top 5 design candidates can achieve
Fig.7. Pareto front of AFIM design optimization based on GA.
 TABLE IV. DESIGN VARIABLES OF THE SELECTED AFIM ANSYS Maxwell results. As can be seen from the Table, the
Design Variable Value most accurate modeling among them has about 8% MAPE,
Tring 1.20 cm which is much higher than the results from the proposed
Tiron 2.08 cm modeling method. This is mainly because all those existing
V 159 V modeling works were not designed for EV applications. Many
ω 1015 rad/s factors that cannot be ignored at such high-power level had
g 0.86 mm been neglected, such that those modeling works are not very
ro 9.95 cm
accurate when it comes to AFIM of EVs.
ri 6.09 cm
hr 1.47 cm
wr 1.18 cm
hs 3.59 cm
ws 0.47 cm
ζ 0.583
us 0.21 mm
ds 0.70 mm
p 3
m 2
NR 20
dw 1.63 mm
NSIP 4
NSP 1

B. Motor Performance
According to the design variables of the selected AFIM
based on GA, the proposed analytical model is applied to
calculate the output mechanical power, motor efficiency and Fig.8. Output power of the final selected AFIM design.
output torque at different rotor speeds. The results are drawn
with green lines, as shown in Fig. 8, Fig. 9, and Fig. 10. The
selected AFIM is then simulated in ANSYS Maxwell, and the
output mechanical power, motor efficiency and torque is
obtained at selected rotor speeds (every 400 RPM from 400 to
3,200 RPM plus the rated speed 3,051 RPM) and indicated by
red dots in the same figures. The results of the proposed
analytical model and ANSYS Maxwell and their deviations
are also indicated in green, red, and black fonts at these
selected speeds, respectively. The maximum output power
given by the analytical model is 86.4 kW, and the
corresponding efficiency and torque are 93.4% and 274.5 N-m,
respectively. At the same rated rotation speed 3,051 RPM,
ANSYS Maxwell gives an output mechanical power of 83.7
kW, an efficiency of 90.1%, and a torque of 265.2 N-m. All
these values meet the initial design requirements.
Fig.9. Efficiency of the final selected AFIM design.
The accuracy of the proposed analytical model can be
further reflected by mean absolute percentage error (MAPE)
as in formula (37), where xi and zi represent the results from
the proposed analytical model and ANSYS Maxwell,
respectively. The MAPE of these three output values as in Fig.
8, Fig.9 and Fig. 10 are 2.55%, 3.27% and 3.05%, respectively,
indicating a high accuracy of the proposed analytical model.
1 n zi - xi
MAPE = 
n i =1 zi
(37)

In Table V below, the accuracy of the proposed analytical

model is compared with other AFIM modeling methods from
the literature. The expressions in these models are used to
calculate the values of the components in our analytical model
given the design variables in Table IV, and then generate the
output power, efficiency, and torque at the same selected
rotation speeds, and also the MAPE between them and Fig.10. Torque of the final selected AFIM design.
 TABLE V. COMPARISON OF DEFFERENT AFIM MODELING METHODS

Modeling Method Output Power MAPE Efficiency MAPE Torque MAPE Power-to-Weight Ratio (kW/kg)
SP-AFIM [5] 13.47% 15.23% 14.61% 0.54
SS-AFIM [6] 9.53% 10.71% 10.54% 0.56
DS-AFIM [7] 7.74% 8.43% 8.28% 1.09
Proposed Method 2.55% 3.27% 3.05% 4.15

In Fig. 11 and Fig. 12, torque capability and corresponding

efficiency of the selected AFIM design based on the proposed
method over wide range of rotor speed are plotted with green
lines. Below the rated speed 3,051 RPM, Volt/Frequency
control is applied to maintain the constant maximum output
torque. Above the rated speed, voltage is fixed at 159 V and
only electric frequency increases. Likewise, the motor is
simulated in ANSYS Maxwell at selected rotor speeds (every
1,000 RPM from 1,000 to 8,000 RPM), and corresponding
results are shown as red dots in the same figures. The results
of the proposed analytical model and ANSYS Maxwell and
deviations between them are also indicated in green, red, and
black fonts at these selected speeds, respectively. For the
maximum torque capability and corresponding efficiency
comparison, the MAPE are 4.03% and 2.29%, suggesting the
high accuracy of the proposed analytical model over a broad
Fig.11. Torque capability of the final selected AFIM.
range of rotation speed. Besides, the ANSYS results suggest
that the final selected AFIM can maintain a torque value of
265 N-m at low speed and can reach approximately 43 N-m at
8,000 RPM with an efficiency of 93.8%. This excellent torque
capability and efficiency at high speed can definitely meet the
requirements of today’s EVs.
Besides, same in Table V, the power-to-weight ratios of the
typical AFIMs in the literature are compared with the selected
AFIM based on the proposed method. With accurate modeling
and subsequent optimization, the power-to-weight ratio of the
selected AFIM is calculated to be 4.15 kW/kg as suggested in
ANSYS Maxwell. On the other hand, the best result among
the other three types of AFIMs in the literature is around 1.09
kW/kg. This comparison fully reflects the importance of
accurate modeling and further optimization, which can greatly
help improve simulation efficiency and motor performance.
Stator tooth and back irons tends to have the highest
magnetic flux density due the narrow cross-sectional areas. Fig.12. Efficiency at max torque capability.
Thus, for the purpose of verifying whether the selected AFIM
design can meet the corresponding magnetic flux density
requirements, the surface magnetic flux density of one stator
core when the maximum output mechanical power is reached
is plotted in Fig. 13 below. It can be seen from the figure that
the magnetic flux densities in back iron and stator tooth can
the meet the design requirements as in constraints (23) and
(24). Parts of the stator slot depressions, which are used to fix
coil windings, may have high magnetic flux density as marked
in yellow. But considering the facts that those places with high
magnetic flux density are very small and only in stator tooth
depression not in tooth and back iron, they will not have much
effects on the motor operation. Therefore, the overall magnetic
flux density of the selected AFIM design can also meet the
corresponding design requirements.
Fig.13. Surface magnetic flux density magnitude of a stator.
 V. CONCLUSION [17] C. Hwang, L. Lyu, C. Liu and P. Li, "Optimal design of an SPM motor
using genetic algorithms and taguchi method," IEEE Trans. Magn., vol.
This paper proposes a fast design method, consisting of an 44, no. 11, pp. 4325-4328, Nov. 2008.
accurate analytical model and a GA-based design variables [18] T. Ishikawa, K. Yonetake and N. Kurita, "An optimal material
distribution design of brushless DC motor by genetic algorithm
optimization method, for AFIM of EV applications. Unlike considering a cluster of material," IEEE Trans. Magn., vol. 47, no. 5, pp.
traditional design of AFIMs with single objective, multi- 1310-1313, May 2011.
objective is considered in this work. Based on the proposed [19] Y. Okamoto, Y. Tominaga, S. Wakao and S. Sato, "Topology
optimization of rotor core combined with identification of current phase
method, the performance of AFIM can be simulated and
angle in IPM motor using multistep genetic algorithm," IEEE Trans.
optimized in much shorter time compared with finite element Magn., vol. 50, no. 2, pp. 725-728, Feb. 2014.
analysis in ANSYS Maxwell. The comparison results from [20] B. Mirzaeian, M. Moallem, V. Tahani and C. Lucas, "multiobjective
ANSYS Maxwell can prove the effectiveness and accuracy of optimization method based on a genetic algorithm for switched
reluctance motor design," IEEE Trans. Magn., vol. 38, no. 3, pp. 1524-
the proposed method, and the performance of the final motor 1527, May 2002.
design can meet all the initial design expectations. [21] A.E. Fitzgerald, C. Kingsley, and S.D. Umans, Electric Machinery, 6th
Edition, McGraw Hill, 2003.
[22] P.L. Alger, Induction Machines, Gordon and Breach, 1969.
REFERENCES [23] K. Deb, A. Pratap, S. Agrawal, and T. Meyarivan, “A fast and elitist
[1] J. Zhu, K. Cheng, X. Xue, and Y. Zou, “Design of a new enhanced multiobjective genetic algorithm: NSGA-II,” IEEE Trans. Evol. Comput.,
torque in-wheel switched reluctance motor with divided teeth for electric vol. 6, no. 2, pp. 182–197, Apr. 2002.
vehicles,” IEEE Trans. Magn., vol. 53, no. 11, Nov. 2017. [24] D. Panda and M. Ramteke, “Preventive crude oil scheduling under
[2] Y. Zhang, W. Cao, S. Mcloone, and J. Morrow, “Design and flux- demand uncertainty using structure adapted genetic algorithm,” Applied
weaking control of an interior permanent magnet synchronous motor for Energy, vol. 235, pp. 68-82, Feb. 2019.
electric vehicles,” IEEE Trans. Appl. Supercond., vol. 26, no. 7, Oct. [25] C. M. Fonseca and P. J. Fleming, “Multiobjective optimization and
2016. multiple constraint handling with evolutionary algorithms—Part I: A
[3] Z. Nasiri-Gheidari and H. Lesani, “A survey on axial flux induction unified formulation,” IEEE Trans. System, Man, Cybern. Part A:
motors,” PRZEGLD ELEKTROTECHNICZNY (Electrical Review), vol. Systems and Humans, vol. 28, no. 1, pp. 26–37, Jan. 1998.
88, no. 2, pp. 300–305, 2012. [26] M. Wieczorek and M. Lewandowski, “A mathematical representation of
[4] M. Mirzaei, M. Mirsalim, and S. E. Abdollahi, “Analytical modeling of an energy management strategy for hybrid energy storage system in
axial air gap solid rotor induction machines using a quasi-three- electric vehicle and real time optimization using a genetic algorithm,”
dimensional method,” IEEE Trans. Magn., vol. 43, no. 7, pp. 3237–3242, Applied Energy, vol. 192, pp. 222-233, Apr. 2017.
Jul. 2007. [27] N. Gupta, R. Shekhar, and P. K. Kalra, “Congestion management-based
[5] Z. Nasiri-Gheidari and H. Lesani, “Optimal design of adjustable air-gap, roulette wheel simulation for optimal capacity selection: Probabilistic
two-speed, capacitor-run, single-phase axial flux induction motors,” transmission expansion planning,” Int. J. Elect. Power Energy Syst., vol.
IEEE Trans. Energy Convers., vol. 28, no. 3, pp. 543-552, Sept. 2013. 43, no. 1, pp. 1259–1266, 2012.
[6] Z. Nasiri-Gheidari and H. Lesani, “New design solution for static
eccentricity in single stator–single rotor axial flux induction motors,”
IET Electr. Power Appl., vol. 7, no. 6, pp. 523-534, Jul. 2013.
[7] D. K. Banchhor and A. Dhabale, “Design, modeling, and analysis of
dual rotor axial flux induction motor,” 2018 IEEE International
Conference on Power Electronics, Drives and Energy Systems (PEDES),
Chennai, India, 2018, pp. 1-6.
[8] D. Golberg, Getrelic Algorithms, Addison-Wesley, 1989.
[9] S. Y. Shin, I. H. Lee, D. Kim, and B. T. Zhang, “Multiobjective
evolutionary optimization of DNA sequences for reliable DNA
computing,” IEEE Trans. Evol. Comput., vol. 9, pp. 143–158, Apr. 2005.
[10] Z. Hu, C. W. Chan, and G. H. Huang, “Multi-objective optimization for
process control of the in-situ bioremediation system under uncertainty,”
Eng. Appl. Artif. Intell., vol. 20, no. 2, pp. 225–237, 2007.
[11] C. Bates Congdon, J. C. Aman, G. M. Nava, H. R. Gaskins, and C. J.
Mattingly, “An evaluation of information content as a metric for the
inference of putative conserved non-coding regions in DNA sequences
using a genetic algorithms approach,” IEEE/ACM Trans. Comput. Biol.
Bioinformat., vol. 5, no. 1, pp. 1–14, Feb. 2008.
[12] C. W. Ahn and R. S. Ramakrishna, “A genetic algorithm for shortest
path routing problem and the sizing of populations,” IEEE Trans. Evol.
Comput., vol. 6, no. 6, pp. 566–579, 2002.
[13] K. Wang and Z. Shen, “A GPU-based parallel genetic algorithm for
generating daily activity plans,” IEEE Trans. Intell. Transp. Syst., vol.
13, no. 3, pp. 1474–1480, Sep. 2012.
[14] M. K. Jiau and S. C. Huang, “Services-oriented computing using the
compact genetic algorithm for solving the carpool services problem,”
IEEE Trans. Intell. Transp. Syst., vol. 16, no. 5, pp. 2711–2722, Oct.
2015.
[15] D. Zou, S. Li, X. Kong, H. Ouyang, and Z. Li, “Solving the combined
heat and power economic dispatch problems by an improved genetic
algorithm and a new constraint handling strategy,” Applied Energy, vol.
237, pp. 646-670, Mar. 2019.
[16] M. Nemati, M. Braun, and S. Tenbohlen, “Optimization of unit
commitment and economic dispatch in microgrids based on genetic
algorithm and mixed integer linear programming,” Applied Energy, vol.
210, pp. 944-963, Jan. 2018.
 Jie Mei (S’13) received his B.S. degree in electrical
engineering from the School of Electrical and
Computer Engineering, Georgia Institute of
Technology, Atlanta, GA, USA, in 2015. He is
currently pursuing his Ph.D. degree at the
Department of Electrical Engineering and Computer
Science, Massachusetts Institute of Technology,
Cambridge, MA, USA.
His research interests include Multi-Energy
System and Electric Vehicle Technologies.

Yuefei Zuo (M’18) received the B.Sc. and the Ph. D.

degrees in electrical engineering and automation from
Nanjing University of Aeronautics and Astronautics,
Nanjing, China, in 2010 and 2016, respectively.
He has been working as a Lecturer at Jiangsu
University since July 2016. He is currently a
Postdoctoral Research Fellow with the School of
Electrical and Electronic Engineering, Nanyang
Technological University, Singapore. His research
interests include advanced control strategies for high
performance electric drives.

Christopher H. T. Lee (M’12-SM’18) received his

B.Eng. (First Class Honors) degree, and Ph.D. degree
both in electrical engineering from Department of
Electrical and Electronic Engineering, The University
of Hong Kong, Hong Kong.
He currently serves as an Assistant Professor at
Nanyang Technological University, Singapore, a
Visiting Assistant Professor at Massachusetts Institute
of Technology, United States, and an Honorary
Assistant Professor at The University of Hong Kong,
Hong Kong. He is an Editor for IEEE Transactions on Energy Conversion,
and Associate Editor for IEEE Access and IET Renewable Power Generation.
He is a Chartered Engineer in Hong Kong. His research interests include
Electric Machines and Drives, Renewable Energies, and Electromechanical
Propulsion Technologies. In these areas, he has published 1 book, 3 books
chapters, and over 80 referred papers.
Dr. Lee has received many awards, including NRF Fellowship, Nanyang
Assistant Professorship, Li Ka Shing Prize (the best Ph.D. thesis prize) and
Croucher Foundation Fellowship.

James L. Kirtley (S’69–M’71–SM’80–F’91– LF’11)

received the Ph.D. degree from the Massachusetts
Institute of Technology (MIT), Cambridge, MA, USA,
in 1971. He is the Professor of electrical engineering
at MIT.
He was with the Department of Large Steam
Turbine Generator, General Electric, as an Electrical
Engineer, with Satcon Technology Corporation as the
Vice President, and the General Manager of the Tech
Center and the Chief Scientist and the Director. He was the Gastdozent at the
Swiss Federal Institute of Technology.
Dr. Kirtley is a specialist in electric machinery and electric power systems.
He served as an Editor-in-Chief of the IEEE TRANSACTIONS ON
ENERGY CONVERSION from 1998 to 2006 and continues to serve as an
Editor for that journal and as a Member of the Editorial Board of the Journal
Electric Power Components and Systems. He was awarded the IEEE Third
Millennium Medal in 2000 and the Nikola Tesla Prize in 2002. He was elected
to the United States National Academy of Engineering in 2007. He is a
Registered Professional Engineer in Massachusetts.

View publication stats