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. Citation information: DOI 10.1109/TVT.2016.2631161, IEEE
Transactions on Vehicular Technology
1

Fast Adaptive Evolutionary PM Traction motor

Optimization based on Electric Vehicle Drive
Cycle
Athanasios G. Sarigiannidis, Student Member, IEEE, Minos E. Beniakar, and Antonios G. Kladas,
Senior Member, IEEE
 The technological transition from conventional to more
Abstract—This paper introduces an evolutionary optimization electric transportation means has triggered an extensive
procedure for the design of Permanent Magnet Motors (PMM) research activity in the field of electric motors. The dominant
for Electric Vehicle (EV) applications, considering a specific drive
electric motor technologies for EV applications are the
cycle with multiple operating points. For the purposes of the
analysis, the New European Drive Cycle (NEDC) has been Induction Motor (IM), the Switch Reluctance Motor (SRM)
employed with two alternative PMM configurations. Energy and the Permanent Magnet Motor (PMM) [2-4]. However,
distribution over the NEDC for a small passenger’s EV has been there is a clear tendency over the last years towards the
calculated and the equivalent multiple operating points have integration of rare-earth materials in high power density
been extracted, using appropriate weights, in order to maintain motors. This is especially evident in EV applications [5-11].
an equal energy consumption basis, resulting in reduced Towards that direction, the utilization of PMM configurations
computational cost. The proposed optimization technique is
constituted of an adaptive Differential Evolution (DE) algorithm is inherently advantageous, taking into account both
involving dynamic variation of the mutation factor, combined operational and environmental criteria.
with Finite Element (FE) and circuit models. The procedure is Among PMMs, Surface Mounted (SMPM) and Interior PM
based on the precise calculation of the two axes input current (IPM) configurations are principally considered for EV
components for each candidate solution and operating condition. applications [1-2], [5], [11]. The SMPM topology presents
The methodology introduced presents stable and fast high power density, performance and efficiency in low speeds,
convergence characteristics, and has been applied to optimize the
geometry of both surface mounted and interior PMM
as well as low manufacturing complexity [5]. The utilization
configurations. The proposed motor is based on the thorough of Fractional Slot Concentrated Winding (FSCW)
trade-off among the two alternative optimized geometries and configurations results in SMPM motors with high copper fill
has been validated through measurements on a prototype. factor values, exhibiting sinusoidal back Electro-Motive Force
(EMF) [7], [8]. However, the SMPM topology, due to the
Index Terms—Design optimization, differential evolution, presence of PMs at the air-gap, demands high demagnetization
drive cycle, electric vehicles, finite element method, field current for the Field Weakening (FW) operation and suffers
weakening, permanent magnet motors, wide speed range
operation. from considerable PM eddy losses, especially at high speeds
[5], [11]. Alternatively, the IPM topology, employing Full
I. INTRODUCTION Pitch Distributed Winding (FPDW) configurations, produces
significant reluctance torque additionally to magnet torque,
industry applications are oriented
A
UTOMOTIVE
due to the saliency of the rotor magnetic circuit [10-13],
increasingly towards more energy efficient and
resulting in high performance and efficiency, for a wide speed
environmentally friendly drive-train technologies. This
operating range. Furthermore, such a rotor configuration
trend is mainly attributed to the detrimental effects of the
provides adequate PM shielding from centrifugal and
greenhouse gas emissions, directly related to vehicle
demagnetization forces, rendering it a favorable option for
emissions. Therefore, electric or hybrid vehicles have been
traction applications [14]. Nevertheless, manufacturing
gaining popularity over the last years, comprising a viable and
complexity, rotor iron losses and cross coupling effects should
efficient solution to the abovementioned bottleneck [1], [2].
be taken into consideration [5], [12-15]. The choice of the
more advantageous between the two candidate motor
Copyright (c) 2015 IEEE. Personal use of this material is permitted.
configurations is application specific, as they both benefit
However, permission to use this material for any other purposes must be
obtained from the IEEE by sending a request to pubs-permissions@ieee.org. from their particular complementary operating characteristics
Dr. A. G. Sarigiannidis was supported in his PhD studies by “ΙΚΥ [1-2] [5], [11], [16]. Therefore, the consideration of both
Fellowships of Excellence for Postgraduate Studies in Greece - Siemens SMPM and IPM topologies in an integrative systematized EV
Program”.
A. G. Sarigiannidis and A. G. Kladas are with the Department of Electrical motor design procedure could provide great services in the
and Computer Engineering, National Technical University of Athens, 9 Iroon determination of the most favorable motor design.
Polytechneiou street, 15780 Zografou Athens, Greece (email: Electric vehicle applications require high performance
thsarig@central.ntua.gr; kladasel@central.ntua.gr).
M. E. Beniakar is with ABB Corporate Research, Västerås, Sweden (email: motors with high torque density, high efficiency over a wide
minos.beniakar@se.abb.com). speed operating range, employing constant torque - Maximum

0018-9545 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
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. Citation information: DOI 10.1109/TVT.2016.2631161, IEEE
Transactions on Vehicular Technology
2

Torque per Ampere (MTPA) and constant power - FW driving TABLE I

PROTOTYPE EV DESIGN SPECIFICATIONS
conditions [11], [12]. The strict nature of the specifications,
Quantity Value
both operational and spatial, of the EV application has
Radius of wheels (m) 0.3
highlighted the necessity of the thorough investigation of the Vehicle front area (m2) 2.14
motor’s behavior over the entire drive cycle. More

Electric
Vehicle
Aerodynamic drag coefficient, Cd 0.22
specifically, the main challenge in the design process of the Rolling resistance coeffient, Crr 0.011
EV traction motor is that the maximum torque is required Vehicle mass, kg / Differential gear ratio 720 / 6:1
Maximum vehicle speed, km/h 120
during acceleration, involving low rotational speed, while the Maximum NEDC acceleration, m/s2 1.043
motor operates on a regular basis at lower torque values and Battery pack nominal voltage (V) 230
higher rotational speed. Furthermore, several fluctuations in Active length / Outer diameter (mm) 105 /180

specifications
the required motor torque and speed are encountered, during a Base / maximum speed (rpm) 2500 / 6370

Motor
Continuous / Peak motor torque below and at
typical EV drive cycle [9-11], [17-19]. As a result, a fast and 45 / 77
base speed (Nm)
stable design optimization process, including the whole motor Continuous / Peak power (kW) 11.8 / 20.2
driving condition via a systematized methodology, is Maximum temperature in PMs/ winding (oC) 150 / 180
necessary [7-9], [12], [20], [21]. Iron laminations / PM material M235-35A / N40SH
In this paper, a viable and computationally efficient motor
design optimization method, enabling the minimization of the 6000

Speed (rpm)
total energy losses over the NEDC, rather than at few
4000
operating points, is developed. The EV drive cycle is
systematically taken into account both in the specified torque- 2000
speed envelope and in the design optimization process for the 0
0 200 400 600 800 1000
investigated EV motor, through appropriate representative 60
operating points. After the assignment of the basic operating Torque (Nm)
conditions of the motor, an evolutionary optimization 40
algorithm is introduced. In particular, an adaptive DE 20
optimization routine, coupled with a fixed step nonlinear 2D
FE model and d-q axes motor equations is developed. The 0
0 200 400 600 800 1000
proposed motor electromagnetic analysis provides significant
20
Power (kW)

precision, since it takes into account the magnetic saturation,

cross coupling effects and the employed control strategy for
10
each considered operating condition. The composite cost
function, accounting for motor performance, efficiency, torque 0
quality and armature voltage harmonic distortion for each 0 200 400 600 800 1000
member of the population is composed of the weighted sum of Time (sec)
the respective functions for each operating point of the motor, Fig. 1. Motor speed, absolute load torque and power over the NEDC.
which are extracted by the driving cycle. Two PMM
topologies (SMPM and IPM), employing FSCW and FPDW resistance, the acceleration and the grading resistance forces
configurations, respectively, are optimized by the proposed [22], [23]. The basic specifications and properties for the
technique and are thereafter compared quantitatively to investigated EV are tabulated in Table I. Motor speed, load
evaluate their overall suitability for the specific EV traction torque and power over the NEDC, considering a ground slope
application. For the optimum PMM topology, as highlighted angle equal to 2.5 degrees, are illustrated in Fig. 1. NEDC is
from the preceded analysis, a prototype is manufactured, utilized for the extensive design optimization of the traction
validating experimentally the simulation results. motor since it is the current standard certification for micro-
sized vehicles in Europe [9], [21].
II. PROPOSED PMM DESIGN METHODOLOGY It should be mentioned that NEDC is only used as a
reference drive cycle, against which the traction motor design
A. Vehicle and PM motor design specifications
optimization should be performed. The subsequent analysis
The investigated EV is a small city car with the capability and described techniques can be applied to any given driving
to operate in suburban environment. The drive train employs cycle.
one motor coupled to the front axles via differential. The Taking into account the results during the NEDC operation
differential gear ratio is quite high, in order to reduce the depicted in Fig. 1, the required motor continuous torque for
motor torque demand, resulting in lower motor size. However, the preliminary design procedure is chosen at a value smaller
this selection in differential gear ratio will increase the need than, but adequately close to the maximum required torque at
for a wide operating motor speed range, in order to achieve the acceleration from standstill. Furthermore, the motor should be
maximum vehicle speed. The motor's torque demand and capable of producing a 70% higher torque, for less than ten
operating speed over NEDC are specified via fundamental minutes, in case of an overtaking, or a starting from a high
analytical vehicle loading equations, regarding the rolling ground slope angle. More specifically, the EV propulsion

0018-9545 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
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. Citation information: DOI 10.1109/TVT.2016.2631161, IEEE
Transactions on Vehicular Technology
3

80 Motor Energy (J) colormap 5

Torque (NEDC) x 10
60
T continuous 10
Torque, T (Nm)

60 T overload

Torque, T (Nm)
8
40
40 6
4
20
20
2

0 0
0 1000 2000 3000 4000 5000 6000 0 2000 4000 6000
Rotating speed, n (rpm) Rotating speed, n (rpm)
Fig. 2. EV propulsion motor specification for continuous and peak torque Fig. 3. Motor energy distribution and clustering (red) over the NEDC with
versus speed characteristics. torque and rotating speed.

motor with the proposed torque-speed specifications is capable TABLE II

EQUIVALENT OPERATING POINTS OVER THE NEDC
of operating adequately in a high slope road of up to 12
Region (j) ωm (rad/s) Tm,j (Nm) % of total energy wj
degrees, under the specified urban NEDC velocity profile. The
1 176.1 35.2 35.6 0.356
maximum acceleration that can be achieved for the 2 194.3 22.5 11.3 0.113
investigated vehicle, considering zero road incline, is equal to 3 417.7 24.7 17.9 0.179
2.028 m/s2. Furthermore, the least time needed to reach 4 561.1 27.6 35.2 0.352
velocity equal to 50 km/h is 6.8sec, while the time needed to
reach velocity equal to 100 km/h from standstill is 19.2 sec, are selected as equivalent operating points. This action enables
due to the fact that for velocity values larger than 48 km, the a detailed representation of the EV motor behavior over
motor operates in constant power-FW region. These NEDC operation, in order to be incorporated in a systematized
performance rates are typical for a small city vehicle. The EV design optimization methodology. The selected equivalent
propulsion motor specifications for continuous (rated) and points summarize the specifications of the entire NEDC cycle
temporary (peak) torque versus speed are shown in Fig. 2. in a conveniently concise form.
From Fig. 2, it can be seen that all operating points that can The motor’s energy consumption / generation at each time
possibly occur for the NEDC cycle fall within the specified step is calculated by the appropriate integration of power,
torque-speed envelopes of the motor. Therefore, provided that considering the power as constant or linear function of time, as
the continuous and temporary overload torque-speed curves illustrated in Fig. 1. The time step is selected quite low
are satisfied, the whole NEDC operating points are achievable. (dt=0.4sec), in order to achieve a detailed representation of
energy distribution with torque and speed. Particularly, for
B. Energy distribution over NEDC and equivalent operating regions where the power and angular speed are assumed
points extraction constant the motor energy is calculated as follows:
The design of a traction motor, intended to operate in a
wide speed range renders the consideration of multiple E   tk  =constant   Pk  tk (1)
operating points mandatory. This is dictated also by the nature
of the NEDC cycle. The obsolete machine design approach where k∈[1,21] represents the regions of NEDC where the
that optimizes only at the nominal operating point can offer speed remains constant, P is the motor power and t the time
minor services towards that direction. The torque-speed interval of each region. For regions where the power is a linear
characteristics of the motor are specified from the preliminary function with time, the energy in each time step, is calculated
design, based on the specifications of the EV. However, it is by the integration of power and expressed by the following
practically impossible to optimize based on the whole torque - equation:
speed profile and especially when partial loads and overload ti  dt
ai ai
conditions are considered in the problem formulation. The E  ti  dt     a  t dt  E   ti  dt     ti   Eo (2)
2 2

i o

inclusion of the drive cycle in the design procedure through ti 2 2
the extraction of equivalent operating points offers a trade-off
between computational cost and accuracy. where ai is the slope of the power linear function at the ith time
Following the assignment of the specified torque-speed interval and Eo the initial energy at which the power starts to
curve in continuous and temporary overload operation, vary linearly. For the corresponding angular speed ω(ti+dt) for
according to the NEDC torque and speed fluctuations, the each time interval, the mean value is considered.
motor energy distribution for each single NEDC operating The calculated machine energy as a function of speed and
point is calculated. Based on the variation of the motor energy torque is shown in Fig. 3. It should be noted that principally
on the torque-speed plane, a number of representative sub- the machine is operating in motor mode. However,
regions are extracted and their respective energetic centroids regenerative braking is also taken into account in the energy

0018-9545 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
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. Citation information: DOI 10.1109/TVT.2016.2631161, IEEE
Transactions on Vehicular Technology
4

consumption calculation (generator mode), since the absolute

value of power is integrated at each time step. The motor drive
cycle operation is decomposed in four main torque-speed sub-
regions, enable achieving a convenient compromise between
accurate NEDC clustering and computational cost of the
design optimization procedure, as the FE method is utilized
for the maximum precision of the simulated results. Each
region is represented by an equivalent energy weighted
average point [9], while the respective limits of these regions
are equally shown in Fig. 3. The torque and speed of the Fig. 4. Overview of the investigated PMM configurations. (a) IPM. (b)
equivalent operating points extracted via this methodology and SMPM.
the value of the energy of each region as a percentage of the
total motor energy are summarized in Table II. The gravity Layer (SL), but still exhibits very high inductance values
assigned to each equivalent point wj is proportional to the compared to the FPDW configuration. In that sense, it presents
calculated energy associated to it. In that sense, equal much less PM losses than the SL FSCW topology and still
importance is attributed both to MTPA and FW operation, of exhibits good FW capability [26]. It additionally offers the
course via the equivalent point representation. most compact motor design due to shorter end windings and
In particular, the fourth operating point can be considered as axial build.
overload for a city vehicle as it is deep into FW and involves On the other hand, it has been proved that FSCWs reduce
relatively high torque. This kind of operation is encountered the saliency of IPM motors and, thus compromise their
for a small time. The fact that this point falls out of the torque- specific advantages in wide speed range applications, related
speed curve for the continuously rated torque doesn’t render it to the reluctance torque component [13]. Moreover, the
unfeasible. It can be satisfied but for an increased current overload capability is much higher for motors with high
demand and as it can be seen from Fig. 2 it falls inside the saliency [11]. IPM motors exhibit small air-gaps compared to
overload torque-speed curve. SMPM ones and as a result the MMF harmonic content needs
to be mitigated, as it incurs higher iron losses in the rotor. In
C. Preliminary EV motor design
addition, the presence of sub-harmonics in IPM motors,
Regarding the design methodology of both PMMs, on a first employing FSCW, causes unbalanced magnetic loading of the
step classical analytical design techniques are utilized [20], rotor, which incurs saturation, decrease of average torque and
[24], in order to estimate the basic motor characteristics and higher torque ripple [16], [26]. Therefore, for the IPM
specify the appropriate limits of the design variables for the topology, a FPDW configuration, with three slots / pole /
optimization routine. The initial design is focused on the phase is selected, in order to produce MMF with low harmonic
satisfaction of the fundamental spatial limitations and content and sufficient reluctance torque component [10-12].
operational specifications, as tabulated in Table I. The base Furthermore, the DL-VI rotor topology is selected, as shown
speed is selected as the limit of the MTPA motor drive region. in Fig. 4b, due to its higher efficiency, lower produced torque
The 4 pole - 6 slot combination, employing Double Layer ripple and cross saturation effects, compared to other IPM
(DL) FSCW, is adopted for the SMPM topology, as shown in configurations [14], [15]. A 4 pole rotor is also selected for the
Fig. 4a. In order to limit PM eddy current and iron losses to IPM motor, for two main reasons. First, it has been proven
manageable values, due to the high maximum rotational that a 4 pole IPM rotor maximizes saliency, which as
speed, the pole number has been selected as low as possible mentioned before, maximizes reluctance torque and overload
[5], [11], [24]. Thus, the combination of a 4 pole rotor with capability. Second, the 4 pole configuration minimizes the
magnets of three segmentations is a very effective solution respective rotor losses [11]. For both PMMs, the current
towards that direction. The FSCW configuration has been density is assigned to 5.5 A/mm2 for the rated load (45Nm), in
chosen due to its manufacturing simplicity, high copper fill order to achieve adequate thermal robustness [6], taking into
factor and sufficient FW capability [7], [25]. In particular, the account the fact the motor is cooled by an external forced air-
optimal FW capability of SMPM motors is achieved with cooling system.
FSCW configurations, due to the increased values of
inductances [25]. This is attributed to the additional harmonic D. Proposed PMM Design Optimization Methodology
components of the generated Magneto-Motive Force (MMF) After the assignment of the basic operating conditions and
that increase leakage and sub-harmonic inductances. design variables of both PMM topologies, an adaptive single
Additionally, this enhances the fault tolerance capability of the objective DE optimisation routine is implemented. For the
motor [26]. In addition, the FSCW configuration offers shorter implementation of the DE algorithm, the standard DE
end windings that, combined with the higher fill factor, lead to processes are employed [20], [21], regarding donor
superior thermal performance. The 4 pole - 6 slot combination formulation and crossover. However, the mutation factor for
facilitates the manufacturing procedure and has increased fill the differential vector FAdap varies dynamically to a range
factor and efficiency values. The DL winding is chosen as it between 0.5-1.0, as will be explained below. Forced mutation
presents a more sinusoidal MMF than the respective Single and crossover is used for at least one design variable of every

0018-9545 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
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. Citation information: DOI 10.1109/TVT.2016.2631161, IEEE
Transactions on Vehicular Technology
5

2 0
trial vector in order to avoid vector duplication. The DE (-1.7,1.5) (0.43,2) (1.4,1.7)
10 Adaptive DE
strategy employed is the DE/local-to-best/1/bin, as it attempts
(-1.4,1.3)
(0.58,1.1) (1.9,1.3)
Classic DE
1

Best cost value

(-0.41,0.78)
a balance between robustness and fast convergence [8]. For
(1,0.73)
(0.45,0.94)
(-0.44,0.56) (1.5,0.76)

every (ith) mutant vector two difference vectors are utilized: 0 (-1.6,-0.1)
(-0.13,-0.058) (1.1,-0.014)
-5

x2
(-0.76,-0.2)
(-1.3,-0.4) (0.097, -0.14) 10
vi ,G  xi ,G  Fm   xbest ,G  xi ,G   FAdap   xr 2,G  xr 3,G 
(1.3, -0.24)
(-0.77,-0.41) (-0.42,-0.41)
(3) -1 (-0.53, -0.88) (1.2,-1.2)
(-1.2,-1.5)
(1.4,-1.7)

where vi,G is the mutant vector, xbest is the vector of the -2

(-1.4,-2)
-10
-2 -1 0 1 2 10
generation with the lowest cost function value, xr2, xr3 are two x1 0 10 20 30 40 50
randomly chosen vectors from the current population and Fm 0 500 1000 1500 Generation number
(a) (b)
is the constant mutation factor. In order to reduce the
computational cost of the optimization routine, the second Fig. 5. Adaptive DE performance assessment. (a) Initial population
mutation factor (FAdap) is dynamically varied, utilizing a distribution and objective function values (Rosenbrock test function). (b)
Average cost function evolution for 50 optimization runs.
convenient fitness assignment scheme that considers both the
population density and member ranking. Such technique 1.5

Respective fitness values

D(i)=Density
enhances the convergence capability, without compromising 1.1 1.1
1.2 R(i)=Raw fitness
1.1 1.1 1.1
the stability of the algorithm. 1 0.97 0.96 0.96
C(i)=Overall fitness
0.83 0.830.87
The ranking of the population members depends on the 0.72 0.68 0.67
0.75 0.75
0.54 0.55 0.56
number of individuals that dominate them. The single 0.5 0.41
0.49
0.39
0.44 0.41

objective nature of the algorithm enables the straightforward

characterization of the respective domination relationships 0
0 2 3 4 5 6 12 14 16 18 20 22 24 26
between the generation members. A member dominates Generation member number
another if it exhibits a lower cost function value. Therefore, (a)
the raw fitness value of every individual is a penalty value, 1559 0.99 0.99 1
Objective function value

1 0.96 0.96 0.97

1500
lower than unity, directly related to the member’s robustness.

Mutation factor value

1350 0.86 0.84
0.84
0.8 0.74
The raw fitness of every individual is: 1000 0.6
0.7 0.7
0.63
0.62
0.6 0.57 0.57

n d i 
0.52 0.5 0.53
0.51 0.52 0.5 0.5 0.5
766

R i   (4) 0.4
N p 1 500 384
500

287
230 0.2
134 113 151 137
where Np is the number of the population members and nd(i) is 0
12 3 10
63
2 3 7 1 4 2 2 9
6255
0
the number of the generation members that dominate the ith 1 3 6 9
Generation member number
12 15 18 21 24 26 1 3 6 9 12 15 18 21 24 26
Generation member number
generation member. The density value assigned to each (b) (c)
population member, to discriminate between individuals with
Fig. 6. (a) Fitness (b) Cost function and (c) Mutation factor values for each
identical fitness values, is estimated using the kth nearest generation member at 20th iteration.
neighbor method [8], [27]:
1 is illustrated. The modified algorithm with k=5, exhibits faster
D i   (5)
 2k convergence by an average of 14 iterations. Figure 6 shows
i
the respective fitness, cost function and mutation factor
where σik is the distance of the ith population member to the kth values, respectively, for each generation member, as generated
nearest neighbor. The overall fitness value of an individual by the improved adaptive DE scheme.
C(i) is considered as the sum of (4) and (5). The adaptive The optimizer calls an automated parametric surface or
mutation factor for every individual is dictated by the value of interior PM motor design script, generating a nonlinear 2D FE
its overall fitness via a Gaussian distribution tensor: model corresponding to each optimization run. The overall
FAdapt  N (C , max(C (i)),  2 ) (6) procedure is illustrated in Fig. 7. Six and nine design variables
are selected for the SMPM and IPM motor configuration,
where max(C(i)) is the maximum overall fitness value respectively, as shown in Fig. 5, since they play a key-role in
encountered in a generation and σ2 is the standard deviation of terms of motor performance, efficiency and power quality.
the overall fitness values of the generation members. More specifically, special attention should be taken in the
To illustrate the efficiency of the proposed algorithm, it is design of the IPM rotor, as an accurate design of flux barriers
compared to the standard DE algorithm when applied in the can increase the rotor anisotropy, resulting in a higher
standard Rosenbrock test function [28]. For the purposes of reluctance torque component [12], [15]. The selected design
the analysis, every generation comprises 25 members. Figure variable vector for SMPM and IPM is:
5a illustrates the distribution of the initial population over the 
 SPM  mw m Rg Wt Lt wso 
design space, depicting, via a color-map, the respective XG  
G (7)
objective function values. Figure 5b shows the comparison of  IPM   wi 1m L1m w1m L2 m Rg Wt Lt wso 
G
the convergence of the two algorithms. In particular, the
average variation of best cost values for 50 optimization runs where mw, θm are the surface PM width and angle,

0018-9545 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
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. Citation information: DOI 10.1109/TVT.2016.2631161, IEEE
Transactions on Vehicular Technology
6

respectively, Rg is the air-gap diameter, Wt, Lt are the stator

tooth width and length, respectively, wso is the slot opening, wi
is the rotor iron width, θ1m, L1m, w1m are the inner PMs angle,
length and width, respectively, and L2m is the outer PM length.
The design variables limits are shown in Table III. These
limits are fixed at the preliminary design stage and are
selected by combining construction constraints and variables
differences not exceeding 30% of the initial design values,
constituting a convenient compromise between accuracy of the
optimal solution and overall convergence speed. The boundary
geometrical constraints are handled by utilizing the bounce-
back technique [8].
The constraints handling strategy is the “death penalty” one.
For every trial vector generated in each generation, penalty
functions are evaluated and the potential population member is
immediately rejected in the case that a constraint is violated.
The respective penalty function Cj, which is iteratively
calculated for each member and operating point for both PMM
configurations, is expressed as follows:

C j  C1 C2 C3 C4  j   V(1) J rms Tr  (8)

j

where C1 expresses a minimum efficiency value, C2 expresses

a maximum stator voltage for the FW operation in high speed
region, defined by the battery pack voltage level and the
modulation technique, C3 expresses a maximum armature
current density ensuring adequate thermal robustness, C4
expresses a maximum torque ripple, enabling fulfillment of
the industrial standard for the EV traction applications, η is the Fig. 7. Overall optimization procedure block diagram.
motor efficiency, V(1) is the fundamental RMS phase stator
voltage, Jrms is the armature RMS current density and Tr is the energy percentage of the jth operating region in the total motor
torque ripple (% of the rated torque). energy as tabulated in Table II, w1-4 are the cost function
Furthermore, for every considered operating point, the basic weight coefficients and the index 0 expresses the cost function
performance, efficiency and power quality requirements are specifications.
included in the composite cost function. More specifically, the The complexity of the composite cost function proposed
expression involves the minimization of the total losses, the can be justified by taking into account the complexity of the
torque ripple and armature voltage THD and the maximization problem. The consideration of multiple design objectives
of mean torque. The overall cost function is the weighted sum constitutes an adequately accurate discipline for this class of
of the respective functions for each operating point of the applications [7-9], [21], [29]. However, the inclusion of many
motor, extracted by the driving cycle, dictated by the energy objectives, in conjunction with the inevitable consideration of
percentage of each region, as presented in Table II. The cost multiple operating points, imposes additional difficulties in the
function is expressed as follows: cost function formulation. For the given problem, a Pareto
4 4 optimization approach has not been selected due to the high
F  w  F  w   w
j j j 1j 
 F1 j  w2 j  F2 j  w3 j  F3 j  w4 j  F4 j  number of partial objectives involved, and therefore, the
j 1 j 1 (9) weighted sum solution adopted can offer a necessary
4  Tm0, j PL, j Tr , j THDV , j  compromise.
F   wj   w1 j   w2 j   w3 j   w4 j  
 Tm, j PL0, j Tr 0, j THDV 0, j  The feasibility of all the considered operating points is
j 1 
guaranteed during the design stage in a natural way via the
where Tm is the mean produced electromagnetic torque, PL is cost function evaluation mechanism. After the calculation, via
the motor power losses to the input power ratio (%) and THDV non linear FE model, of d-q axis flux maps and inductances as
is the Total Harmonic Distortion of the stator voltage a function of currents for every candidate geometry, the circuit
considering sinusoidal armature input current. A low stator PMM equations in synchronously rotated d-q axis frame are
voltage THD term facilitates the PWM inverter supply of the utilized, in order to calculate the necessary current vector that
PMM, avoiding current harmonics, with beneficial impact on satisfies the given specified torque and speed values for each
the overall motor drive efficiency. Fj is the cost function value of the four equivalent points. In other words, coupled circuit-
of the jth operating region, extracted by the clustering FE models for torque boost and FW modes are utilized to
methodology presented in paragraph IIB, wj expresses the ensure that the operating points will be feasible. For the fourth

0018-9545 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
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. Citation information: DOI 10.1109/TVT.2016.2631161, IEEE
Transactions on Vehicular Technology
7

equivalent point this is guaranteed, but of course at the cost of TABLE III
OPTIMIZATION ROUTINE PARAMETERS
reduced efficiency, since the current amplitude is increased.
Region Tm0,j PL0,j Tr0,j THDV0,j
This reduction is taken into account in the composite cost (j)
w1-4
(Nm) (%) (%) (%)
function through the evaluation of the overall losses term. The 1 [0.2 0.45 0.2 0.15] 35.2 4 5 12
mean torque term in the composite cost function also 2 [0.2 0.45 0.2 0.15] 22.5 4.5 5 12
compensates for potential discrepancies that can be attributed 3 [0.2 0.45 0.2 0.15] 24.7 4.5 7 16
4 [0.2 0.45 0.2 0.15] 27.6 4 7 16
to approximations involved in the circuit-FE model and can C1 , C2 , C3 , C4 93% , 90V , 5.5 A/mm2 , 10%
lead to slight overestimation of the produced torque. The use Fm, FCR, FAdap 0.85, 0.85, [0.5-1]
of the circuit-FE models is also the reason why mean torque is Design variables limits SMPM IPM
not introduced in the constraints vector. mw (mm) [3-5] -
The weight coefficients determination procedure intends to θm / θ1m (degrees) [58-87.3] [108-135]
L1m, L2m (mm) - [15.5-18.5], [29-34]
achieve appropriate balances between construction and wi, w1m (mm) - [7-9.5], [3-5]
operating costs, with particular focus on the minimization of Rg, wso (mm) [49-55], [5-38] [49-55], [1-5.5]
total motor losses, which affect directly the vehicle autonomy, Lt (mm) ,wt (mm) [21-30], [18-25] [20-25], [4-6.5]
as well as mechanical reliability and motor vibrations. The
final weights utilized in the optimization procedure were currents for each PMM operating condition, considering the
determined based on empirical rules, while taking into account saturation of the magnetic circuit, as well as the cross coupling
previous works of the authors on PMM optimization [20], [29] effects between d and q axis, mainly regarding the IPM
and the respective state of the art literature on the field [30]. topology [10], [14]. The drive condition of the PMM (MTPA
The optimization routine parameters are presented in Table III. for regions 1-2 and FW for regions 3-4) is taken into account
The constraints of the problem take into account the main in the two axes calculation technique. The voltage and torque
given specifications, except for the mean torque, for the PMM steady state equations in the d-q rotating reference
reasons already explained. The torque ripple and maximum frame are expressed as follows:
current density are considered as they play an important role Vd  Ra  I d  e  Lq ( I d , I q )  I q (10)
in a traction motor system, in terms of mechanical and thermal
robustness. In particular, the torque ripple in a traction motor Vq  Ra  I q  e  Ld ( I d , I q )  I d  e   mag (11)
is one of the main causes of vibration that leads to premature
wear on the drive-train components and high level of acoustic
noise, reducing the inherent advantages of EVs versus
Te 
3
2

p   mag  I q   Ld ( I d , I q )  Lq ( I d , I q )   I d  I q  (12)
conventional vehicles in terms of less noisy operation. Since where Id, Iq are the d and q-axis currents respectively, Ld(Id,Iq),
there is no standard for the armature voltage THD, no Lq(Id,Iq) are the nonlinear d and q-axis inductances,
constraint is introduced. Additionally, overall efficiency and respectively, Φmag is the induced PMs flux, Ra is the armature
armature voltage in the high speed-FW operating region are resistance, ωe is the stator electrical angular speed and p is the
considered, which is a quite conventional approach. number of pole pairs.
The composite cost function terms are expressed in terms of For the MTPA control strategy, implemented for the
ratios to provide a convenient per unit equivalent equivalent operating points 1 and 2, the corresponding
representation of the results. The utilization of ratios of the relationship between Id and Iq for both PMM topologies is:
respective terms divided by the respective specifications offers
a trade-off between the direct cost function method and the SMPM  Id  0, Iq  I
goal programming method. In the first, the cost function is (13)
mag ma
2
g
expressed in terms of the quantities that are directly to be IPM  Id    Iq2
2   Lq (Id ,q )  Ld (Id ,q ) 4   Lq (Id ,q )  Ld (Id ,q )
2
minimized [20], [27], [29] and the second in terms of the
distances from an a priori defined ideal vector of values [30].
The formulation of the cost terms as ratios enables the Therefore the required d and q-axis currents are calculated by
consideration of normalized values around a specified or ideal solving (12), (13), Ld = f(Id,Iq) and Lq = f(Id,Iq), applying the
value. The use of ratios is quite common in motor specified Tm0,j. For the FW control strategy, applied to the
optimization problems [20], [29], especially when combined equivalent operating points 3 and 4, the following stator
with weighted partial objectives, since for the assigned voltage constraint is implemented, considering Space Vector
weights to be more effective, the objective functions need to Pulse Width Modulation (SVPWM) technique for the
be normalized to factor in their different dynamic ranges. This controller:
formulation by no means alters the nature of the procedure but 1
V d2  V q2  V m2a x  V DC (14)
instead facilitates the solution evaluation process.
3
After the construction of the candidate motor geometry, the
direct (d) and quadratic (q) axis inductances as a function of where Vmax is the maximum acceptable phase voltage, and Vdc
both d-q axis currents, the induced magnet flux and the is the battery voltage level applied to the inverter. Substituting
armature resistance, are calculated from the parametric FE (10), (11) into (14), the voltage constraint can be expressed in
model. This enables the precise calculation of the d and q-axis elliptical form by the following equation:

0018-9545 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
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. Citation information: DOI 10.1109/TVT.2016.2631161, IEEE
Transactions on Vehicular Technology
8

R I e  Lq (I)  Iq   Ra  Iq e  Ld (I)  Id mag   Vmax

The population distribution for the whole of the design
2 2 2
(15)
a d variables for various generations regarding the SMPM and
IPM motor configurations are illustrated in Figs. 9 and 10,
For the FW mode, the operation on the voltage constraint
respectively. From Figs. 8-10, it can be observed that the
curve is chosen for efficient PMM drive [11], [31]. Therefore,
optimization procedure achieves convergence quite quickly,
the required d and q-axis currents are calculated by solving
approximately at the 17th generation, for both PMM
(12), (15) Ld = f(Id,Iq) and Lq = f(Id,Iq), applying the specified
topologies. This is attributed to the adaptive mutation
Tm0,j.
technique, the small domain of the design variables and the
Thereafter, the respective currents for each operating region
appropriate population size. The four constraint penalty
are imported in the parametric fixed step 2D FE model, in
function system (8) constitutes a highly constrained problem
order to determine the steady-state performances. More
but however, quite typical for traction motor optimization
specifically, in the "synchronous rotation analysis" stage, the
applications. No convergence problem has been observed,
introduced instantaneous current and current angle values are
while the rejected candidate solutions due to operational
kept constant and the rotor and stator MMFs are
constraint violation varied between 8%-17% of the total
synchronously rotated for one full electrical period. During
population members. The slightly higher variation of variables
this set of magnetostatic analyses, all the steady state
observed at the 26th generation for the population of the IPM
performance terms are calculated and forwarded to the cost
configuration, particularly concerning the rotor, is mainly due
and penalty function evaluation module. The stator voltage for
to the strict nature of the geometrical constraints, which
every operating point is evaluated through the available
reduce the number of feasible variable combinations.
computed flux linkage data of the magnetostatic analysis by
For the two optimal PMM configurations, obtained from the
introducing the electric frequency as a post processing
adaptive DE optimization methodology, the design parameters
variable. This achieves substantial savings in computational
values as well as the electromagnetic field characteristics and
cost and time. The core losses PFE are computed by relation
performance indexes are tabulated in Table IV. From Fig. 8
(16), taking into consideration space harmonic phenomena
and Table IV, it can be concluded that the DL IPM topology
[32].
exhibits lower overall cost function and torque ripple values,
   as well as higher efficiency, especially at high speeds.
PFE     Ch  Bm ,n  m  e 2  Ce  Bm2 , n  m  e 2   VFE (16)
n  m 1 
1.05 Surface Mounted PMM
Cost Function, F

Interior PMM
where Ch, Ce are the hysteresis and eddy current losses 1
coefficients, respectively, Bn is the flux density of nth element
0.95
of PMM mesh, m is the harmonic order and VFE is the total
volume of the iron core. The coefficients Ch, Ce are equal to 0.9
143 and 0.53 W/(m3*T2*Hz), respectively, for M235-35A iron 0 5 10 15 20 25
Iteration
laminations grade, and are evaluated from experimental data
provided by the manufacturer. The FE model utilized for the Fig. 8. Cost function progress during optimization runs for both PMMs.
proposed design optimization procedure, is based on the open-
source FEMM solver, combined with a post processing G=26
G=13
G=26
Airgap diameter, Rg (mm)

G=13
Slot opening, wso (mm)

toolbox developed by the authors. 55

G=1 40
G=1

Furthermore, for the SMPM topology, the PM eddy losses 53

30

are calculated via analytical techniques, employing a harmonic 51

20

based approach to the problem of PM eddy losses [8]. This 10

49
combination of closed-form analytical tools provides a fast 90 5
0
30
80 27 24
and effective method for predicting losses related to MMF 70 4 24 20
22
60 3 Surface PM 21 18
harmonics, avoiding time-consuming transient analyses [33]. Surface PM angle,
 m (degrees) width, mw (mm)
Tooth length, Tooth width,
tw (mm)
Lt (mm)
This formulation is convenient for optimization problems and
most importantly, can take into account the magnet Fig. 9. Population distribution of the 1st, 13th and 26th generation per set of
three design variables for the SMPM motor topology.
segmentation.

III. RESULTS AND DISCUSSION

For the optimization routine, the initial population contains
24 and 36 members for surface and interior PMMs,
respectively, in order to compromise between computational
cost and convergence speed to the global optimum. It should
be noted that equal PM volume is specified for both PMM
topologies, in order to achieve a common comparison basis.
The cost function progress for 26 iterations is shown in Fig. 8. Fig. 10. Population distribution of the 1st, 13th and 26th generation per set of
three design variables for the IPM motor topology.

0018-9545 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
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. Citation information: DOI 10.1109/TVT.2016.2631161, IEEE
Transactions on Vehicular Technology
9

TABLE IV
OPTIMAL DESIGN CHARACTERISTICS OF BOTH PMMS
Quantity Surface mounted PMM Interior PMM
Operating regions (j) 1 2 3 4 1 2 3 4
Tm (Nm) 34.8 22.7 25.7 28.7 35.1 21.7 24.8 28.6
Electromagnetic field characteristics

η (%) 95.6 96.4 96.9 96.7 96.1 97.1 97.4 97.3

Id (A) 0 0 -37.3 -57.4 -54 -29.5 -59.2 -89.1
Iq(A) 94.8 60.6 66.5 73.1 53.2 39.1 33.5 25.2
Irms (A) 66.8 42.8 53.9 65.8 53.6 34.6 50.1 65.5
Jrms (A/mm2) 4.2 2.6 3.2 4.1 4.5 2.9 4 5.4
(a) (b)
V(1) (V) 35.2 37 66.3 83.9 43.2 41 68.5 84.1
THDV (%) 3.9 6.6 8.9 10.02 10.2 13.4 19.2 21.4 Fig. 11. Magnetic flux density distribution of the optimal SMPM motor
Tr (% of peak topology at (a) j=1 and (b) j=4 equivalent operating points.
6.7 5.2 7.1 9.2 4.9 3.7 6 7.51
torque)
Ld (mH) 0.81 0.81 0.81 0.8 1.52 1.6 1.46 1.05 50 150
j=1, =176.1 rad/sec (MTPA)

Electromagnetic Torque, Te (Nm)

Lq (mH) 0.81 0.81 0.81 0.8 3.7 4.32 3.91 4.12

Phase A Stator Voltage, Va (V)

j=2, =194.3 rad/sec (MTPA)
j=3, =417.7 rad/sec (FW) 100
Φmag (mWb) / Ra (mΩ) 124 / 18 102 / 25 40 j=4, =561.1 rad/sec (FW)
L1m / L2m (mm) - 17 / 32.7 50
Design variables

w1m / w2m / wi (mm) - 4.97 / 1.41 / 7.94 30 0

Lt / wt / Rg (mm) 26.8 / 19.02 / 49.07 23.2 / 4 / 53.43
θm / θ1m (o) 69.6 116.4 20
-50

wso / mw (mm) 30.3 / 3.33 4.1 / - -100

Turns per phase 40 42
10
Copper fill factor 0.55 0.5 0 50 100 150 -150
0 50 100 150 180
Rotor position, m (mechanical degrees) Rotor position, m (mechanical degrees)
(a) (b)
The flux density distributions of the optimal SMPM motor
topology for the first (j=1) and fourth (j=4) equivalent Fig. 12. Simulated SMPM motor (a) electromagnetic torque and (b) phase
voltage with rotor position at four equivalent operating points of NEDC.
operating points are illustrated in Fig. 11. The torque and
stator voltage waveforms for the four equivalent operating
conditions of NEDC for SMPM configuration for one full
electrical rotation are shown in Fig. 12, respectively. The
electromagnetic field characteristics and performances of the
IPM configuration are shown in Figs. 13 and 14.
Figure 15 shows the comparison of the phase voltage and
electromagnetic torque harmonic components with sinusoidal
current supply at the fourth operating point for the two PMM
topologies, where significant distortion effects are created due
to the high demagnetizing current. The IPM topology yields (a) (b)
considerable high order harmonics (11th, 15th and 17th) in Fig. 13. Magnetic flux density distribution of the optimal IPM motor topology
phase voltage, due to the high number of stator slots and rotor at (a) j=1 and (b) j=4 equivalent operating points.
barriers, in comparison with the SMPM topology, where low
50
order harmonics are observed (3rd, 5th and 7th), due to the low j=1, =176.1 rad/sec (MTPA)
150
Electromagnetic Torque, Te (Nm)

Phase A Stator Voltage, Va (V)

j=2, =194.3 rad/sec (MTPA)

number of stator slots employed. From Figs. 12 and 14, it j=3, =417.7 rad/sec
j=4, =561.1 rad/sec
(FW)
(FW) 100
40
should be noted that in the SMPM topology, the FW capability 50
is based on flux leakage paths in the stator yoke, due to the 30 0
FSCW topology, while in the IPM topology flux leakage paths -50
are observed within the rotor, between the inner PMs and 20 -100
through the iron bridges, enhancing its FW capability.
-150
Furthermore, the IPM topology presents lower torque ripple 10
0 50 100 150 0 50 100 150
over the whole NEDC, compared to the SMPM topology, Rotor position, m (mechanical degrees) Rotor position, m (mechanical degrees)
mainly due to the reduced cogging torque and magnetic (a) (b)
saturation in stator teeth [34]. Furthermore, from the Fig.14. Simulated IPM motor (a) electromagnetic torque and (b) phase voltage
electromagnetic characteristics of the two PMMs, as tabulated with rotor position at four equivalent operating points of NEDC.
in Table IV, it can be observed that the IPM presents higher
inductances values, as the PMs are placed inside the rotor, Furthermore, for the optimal IPM topology, the
increasing the equivalent permeability of the air-gap, demagnetization withstand of the outer PMs has been
compared to the respective for the SMPM case. The higher examined, due to their low thickness. The FE analysis of the
inductance value favors the IPM topology in case of a fault, fourth operating point illustrated in Fig. 13b, results a flux
reducing the fault current, as well as facilitating the efficient density value at the surface of the outer PMs of 0.47T, due to
current filtering. the high d-axis demagnetizing current imposed. Furthermore,

0018-9545 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
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. Citation information: DOI 10.1109/TVT.2016.2631161, IEEE
Transactions on Vehicular Technology
10

Normalized Torque component (pu)

Normalized Voltage Amplitude (pu)
1 Surface Mounted PMM 1
Interior PMM
0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0
0 5 10 15 20 0 5 10 15 20
Harmonic order Harmonic order
(a) (b) Fig. 16. Power loss distribution for both PMM topologies at the equivalent
Fig. 15. Harmonic content of (a) stator phase voltage and (b) electromagnetic NEDC operating points (j=1-4).
torque at 4th operating point (FW operation) for both PMM topologies.
250 Analytical model j=1 j=2 j=3 j=4

PM eddy losses (W)

PM eddy losses (W)

in case of a three phase fault, the outer PMs flux density is Transient FEM 250
200
equal to 0.7T. These flux density levels do not involve 200
150
considerable demagnetization risks for the N40SH grade PM 150
100
material implemented. 100
50 50
The power loss distribution for both PMM topologies at the
0 0
four equivalent operating points is illustrated in Fig.16. 1 2 3 4 0 2 4 6
Operating point (j) Electrical angle (rad)
Copper losses are similar for both PMM topologies, as the
(a) (b)
IPM topology presents higher winding factor and produces
reluctance torque, while the SMPM employs lower armature
resistance, due to its higher fill factor and shorter end
windings. In addition, it should be noted that for the SMPM
topology, three PM radial segmentations per pole are utilized,
for the necessary reduction in the PM eddy losses.
Nevertheless, the SMPM topology exhibits significant PM
eddy losses values, especially at high speeds, due to the high
conductivity of the PM material and the high variability of the
magnetic field in the air-gap. On the other hand, the IPM (c) (d)
topology presents higher rotor core losses, especially in the
Fig. 17. PM eddy losses validation for the SMPM topology by a transient FE
FW operating conditions (j=3&4), due to the fluctuation of the
model. (a) comparison between analytical and FE model and (b) PM eddy
magnetic flux in the rotor iron, in order to suppress the PM losses versus electrical angle at the equivalent NEDC operating points (j=1-4).
field. However, the PM eddy losses of SMPM topology are PMs (c) eddy current density and (d) loss distribution at j=1 operating point.
much higher than IPM iron losses, rendering the IPM topology
more efficient for a wide-speed range PMM drive. ripple and armature voltage THD is not easy to be estimated in
In order to account for the precision of the analytical a straightforward manner. Since both constitute important
model, the respective PM eddy losses for the 4 equivalent components of the composite cost function, a sensitivity
NEDC points are calculated via a transient FE model of the analysis is conducted, in order to validate the optimal nature of
optimum SMPM motor geometry. Maxwell software was the final geometry. In particular, four slot opening widths are
employed for the analysis. Figure 17 shows the comparison of examined, namely 33 mm that practically corresponds to open
the respective values, the waveforms of PM eddy losses and slots, 30 mm that corresponds to the optimum value, 27 mm
the distributions of eddy current density and losses in the and 21 mm. For the purposes of the analysis, the specified
segmented magnets. It can be pointed out that the results of torque is considered for equivalent NEDC points 1 and 4. The
the analytical model are in good agreement with the respective PM eddy losses are calculated via a transient FE model. The
ones generated by the transient FE model. The deviation is results of the sensitivity analysis are shown in Fig. 18.
less than 8% for all considered operating points. From the results illustrated in Fig. 18, it can be concluded
The optimum SMPM topology exhibits a high value of slot that as the slot opening width values decrease, the copper and
opening width, equal to 30 mm, which leads to a geometry PM eddy losses decrease. This is attributed to the wider tooth
with practically open slots. Stator configurations with high slot tip, which increases the produced mean electromagnetic
openings are often employed in FSCW because they increase torque. However, the iron losses slightly increase. Moreover,
the saliency and as a result the average torque. Additionally, it can be seen that smaller slot opening widths exhibit very
they provide higher fill factor and easier manufacturing, high stator voltage THD values and increased torque ripple,
especially for DL windings, when the slot opening has to be due to the local saturation observed in the tooth tips. It has to
adequately large to allow the placing of the coils [26], [29]. be noted that the formulation of the cost function emphasizes
The increase of the slot opening incurs higher PM eddy in the reduction of torque ripple and stator voltage THD along
losses. However, the influence of the slot opening in torque with the maximization of efficiency. We can see that the slot

0018-9545 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
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. Citation information: DOI 10.1109/TVT.2016.2631161, IEEE
Transactions on Vehicular Technology
11

NEDC point 4 15
Copper losses
500 PM losses NEDC point 1 NEDC point 4
12

Tripple / THDV (%)

Iron losses
400
Losses (W)

NEDC point 1 9
300
6
200
T ripple
100 3 Stator voltage THD
T ripple constraint
0 0 (a) (b) (c)
33 30 27 21 33 30 27 21 33 30 27 21 33 30 27 21
Slot opening width, wso (mm) Slot opening width, wso (mm)
Fig. 19. Prototype. (a) Stator and housing. (b) rotor part. (c) stator and rotor
(a) (b) M235-35A iron laminations.
Fig. 18. Slot opening sensitivity analysis for SMPM configuration. (a) Total
losses and (b) Harmonic distortion effects in torque and armature voltage.

opening width variation incurs violent changes in their values,

increasing the associated cost terms. Additionally, it can be
seen that the torque ripple limits are violated for the cases of
33 mm and 27 mm, therefore such solutions would be
automatically rejected during the optimization procedure.
The magnet-to-pole ratio plays an equally important role
with the slot opening width for SMPMs in the armature
voltage THD, torque ripple and average torque values.
Through systematic optimization, stator geometries with
nearly open slots can be extracted, exhibiting low harmonic
content in stator voltage and produced electromagnetic torque
[8], [29].
Finally, from the results presented in this section, it can be
concluded that the IPM topology over-performs the SMPM
one over the whole NEDC, in terms of efficiency, torque Fig. 20. Experimental setup. (a) DC supply (b) Power electronics and control
quality, fault tolerance and mechanical robustness. unit (c) IPM motor test rig and programmable DC loads (d) Data acquisition.

IV. EXPERIMENTAL VALIDATION

In a next step, for the optimum IPM topology, a prototype
motor has been manufactured, in order to experimentally
validate the preceded simulation results. The basic
components of the manufactured prototype are illustrated in
Fig. 19. The test rig for measuring the performance of the
prototype IPM is shown in Fig. 20. It includes the prototype (a)
motor and the load machine, which is a separately excited 20-
No load phase Voltage, Va (V)

60 Measured Measured
40
kW Direct Current (DC) machine, with variable Simulated Simulated
Voltage amplitude (V)

40
programmable loads connected in the armature winding, the 20 30
torque transducer and the rotary encoder. The prototype motor 0
20
is driven by an Insulated Gate Bipolar Transistor (IGBT) -20
inverter (1200V/300A). For proper motor control, a -40 10
TMS320F2812 Digital Signal Processor (DSP) is employed, -60
0
supplying the prototype with appropriate SVPWM voltage 0 3 6 9 12 15 0 5 10 15 20
Time, t (msec) Harmonic order
with switching frequency equal to 7 kHz, due to its (b) (c)
advantageous characteristics in terms of drive system
efficiency and torque quality [35]. Fig. 21. No load test. (a) measured and simulated IPM motor phase voltage.
(b) comparison of simulated and experimental phase voltage waveforms at
Initially, the no-load test on the prototype motor is
2000 rpm. (c) Back-EMF harmonic contents comparison at 2000 rpm.
performed in order to validate the voltage constant and the
sinusoidality of the induced back-EMF. The prototype is In a second step, the load test on the IPM motor is
rotated at different speeds by means of the load motor and the performed in order to validate the torque capability of the
back EMFs are measured at the terminals of the winding. The machine and the saturation of the magnetic circuit. The
motor exhibits satisfactory behavior even under an operating machine is rotated at base speed (2500rpm) and the full range
speed of 6300 rpm. From Fig. 21, it can be noted that the EMF of current up to the rated is loaded into the winding structure.
measurements are in good agreement with the simulated ones. The comparison of simulated, from FE model, and measured

0018-9545 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
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. Citation information: DOI 10.1109/TVT.2016.2631161, IEEE
Transactions on Vehicular Technology
12

Output Torque, T (Nm)

100

Output Torque, T (Nm)

80
60
90

Efficiency,  (%)
60 80
40
70
40 Simulated T continuous
20 prosomeiwmeni 60 Simulated T overload
Measured T continuous
peiramatiki 20 50 Measured T overload
0 40
4 equivalent NEDC points
0 20 40 60 80 100 0 2000 4000 6000 0 2000 4000 6000
Input phase current rms, I (A) Rotating speed, n (rpm) Rotating speed, n (rpm)
(a) (a) (b)
200 60 1
THD Ia=3.9%

Normalized Amplitude
40 Fig. 23. Simulated and measured (a) torque-speed and (b) efficiency-speed
Phase Voltage, Va (V)

Phase current, I (A) 0.8

100 curves for nominal and overload operating conditions. Measured (a) Torque
20
0.6 and (b) Speed of the four NEDC points.
0 0
0.4
-20
-100 the simulated, due to the quite longer end windings and the
-40 0.2
resistance of the external cables. Furthermore, the iron losses
-200
0 0.005 0.01 0.015 0.02
-60
0.025
0
0 2.5 5 7.5 10 12.5 15 are increased approximately 12% due to the manufacturing
Time, t (sec) Frequency, fs(kHz) procedure, according to no-load loss measurement. In
(b) (c)
addition, possible errors in torque and speed measurements
Fig. 22. Load test at constant torque operation and at 2500 rpm. (a) Simulated may also contribute to the difference. From Figs. 2, 22 and 23,
(FE) and measured torque with current. (b) PMM experimental phase voltage it can be concluded that the proposed PMM design provides
and current with 20Nm load torque. (c) current harmonic content. the investigated EV with the necessary torque for the whole
NEDC operation, employing high efficiency values (simulated
torque are illustrated in Fig. 22a. The non linearity between above 93%-measured above 90.7%) over wide speed range.
the produced electromagnetic torque and armature current can
be noticed in the region above the rated continuous torque V. CONCLUSION
(45Nm) and is attributed to the saturation of the iron core. It
should also be mentioned that the flux density distributions at The paper introduces a particular DE optimization
equivalent NEDC points illustrated in Fig. 13 represent non methodology adapted to EV drive applications, considering
saturated operating conditions, below the rated continuous overall driving cycle of the motor. More specifically, the
torque. The slight discrepancies between the simulated and proposed technique undertakes the multidisciplinary problem
measured torque can be attributed to mechanical losses and of EV traction motor design optimization in a computationally
torque transducer accuracy. Experimental SVPWM voltage efficient way, utilizing equivalent energy weighted operating
and current waveforms for load torque equal to 20Nm at base points in the composite cost function, combined with two axis
speed are illustrated in Fig. 22b, while the harmonic content of circuit equations, that takes into account the motor control
the measured current is shown in Fig. 22c. It is evident, from strategy. The optimization algorithm offers fast and stable
Fig. 22, that the motor is driven quite efficiently at the MTPA convergence characteristics, enhanced by the dynamic
operation with the proposed drive system, presenting currents variation of the mutation factor. Two different PMM
with low harmonic content (THD=3.9%). This occurs mainly topologies, namely IPM and SMPM configurations, have been
due to the quite sinusoidal induced stator voltage, as shown in optimized for implementation in a small EV. The comparative
Fig. 14b, the appropriate switching frequency employed and analysis, considering the combined contributions of the four
the high filtering action by winding inductance. equivalent operating NEDC points, illustrated that the IPM
In a next step, the torque and efficiency versus speed topology presents 15% lower power losses and 21% lower
characteristics under constant torque (MTPA) and constant torque ripple, while the SMPM one involves 51% less THD in
power (FW) control for the continuous torque and overloading the armature stator voltage. Consequently, as the efficiency
are calculated. Furthermore, the experimental torque and constitutes a key feature affecting directly the vehicle
efficiency values for the four equivalent NEDC operating autonomy and taking into consideration the superior fault
points are depicted in Figs. 23a and 23b, respectively. The tolerant capability and mechanical robustness in high speed
comparison between the simulated (FE) and measured results ranges, the IPM topology has been adopted as favorable
are illustrated in Fig. 23. It should be noted that mechanical option for the specific EV application. The optimized IPM
losses have been subtracted from both the simulation and motor has been manufactured and tested under various
experimental results. Moreover, the temperature variation is operating conditions, validating the simulation results over the
taken into account in the FE model, for the calculation of IPM entire NEDC.
motor efficiency. The measured efficiency, for rated load,
equivalent NEDC points and overloading, compared with the REFERENCES
simulated one is approximately 0.7-2.3% lower. This small [1] J. de Santiago, H. Bernhoff, B. Ekergård, S. Eriksson, S. Ferhatovic, R.
Waters, and M. Leijon, “Electrical Motor Drivelines in Commercial All-
discrepancy is owing to several reasons. Primarily, the Electric Vehicles: A Review,” IEEE Trans. Veh. Technol., vol.61, no.2,
armature resistance has approximately 15% higher value than pp.475-484, Feb. 2012.

0018-9545 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
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. Citation information: DOI 10.1109/TVT.2016.2631161, IEEE
Transactions on Vehicular Technology
13

[2] A. El-Refaie, “Motors/generators for traction/propulsion applications: A [21] S. Gunther, S. Ulbrich, and W. Hofmann, “Driving Cycle-Based Design
review,” IEEE Veh. Technol. Mag., vol. 8, no. 1, pp. 90–99, Mar. 2013. Optimization of Interior Permanent Magnet Synchronous Motor Drives
[3] V. T. Buyukdegirmenci, A. M. Bazzi, and P. T. Krein, “Evaluation of for Electric Vehicle Application,” in International Symposium on Power
Induction and Permanent-Magnet Synchronous Machines Using Drive- Electronics, Electrical Drives, Automation and Motion (SPEEDAM),
Cycle Energy and Loss Minimization in Traction Applications,” IEEE pp.25-30, 18-20 June 2014.
Trans. Ind. Appl., vol.50, no.1, pp.395-403, Jan.-Feb. 2014. [22] J. M. Terras, D.M. Sousa, A. Roque, A. Neves, "Simulation of a
[4] C. S. Edrington, M. Krishnamurthy and B. Fahimi, "Bipolar Switched commercial electric vehicle: Dynamic aspects and performance", in
Reluctance Machines: A Novel Solution for Automotive Applications," Proc. European Conf. on Power Electronics and Applications (EPE),
IEEE Trans. Veh. Technol., vol. 54, no. 3, pp. 795-808, May 2005. Birmingham, UK, pp.1-10, Aug. 2011.
[5] K. I. Laskaris, and A. G. Kladas, “Optimal Power Utilization by [23] C. Chan, A. Bouscayrol, and K. Chen, “Electric, hybrid, fuel-cell
Adjusting Torque Boost and Field Weakening Operation in Permanent vehicles: Architectures and modeling,” IEEE Trans. Veh. Technol., vol.
Magnet Traction Motors,” IEEE Trans. Energy Convers., vol.27, no.3, 59, no. 2, pp. 589–598, Feb. 2010.
pp.615-623, Sep. 2012. [24] N. Bernard, L. Dang, J. C. Olivier, N. Bracikowski, G. Wasselynck and
[6] G. Li, J. Ojeda, E. Hoang, M. Gabsi, and M. Lecrivain, “Thermal- G. Berthiau, "Design Optimization of High-Speed PMSM for Electric
electromagnetic analysis for driving cycles of embedded flux-switching Vehicles," in IEEE Vehicle Power and Propulsion Conference (VPPC),
permanent-magnet motors,” IEEE Trans. Veh. Technol., vol. 61, no. 1, pp. 1-6, Montreal, QC, 2015.
pp. 140–151, Jan. 2012. [25] A. M. EL-Refaie and T. M. Jahns, "Optimal flux weakening in surface
[7] J. Wang, X. Yuan, and K. Atallah, “Design optimization of a surface PM machines using fractional-slot concentrated windings," IEEE Trans.
mounted permanent-magnet motor with concentrated windings for Ind. Appl., vol. 41, no. 3, pp. 790-800, May-June 2005.
electric vehicle applications,” IEEE Trans. Veh. Technol., vol. 62, no. 3, [26] E. Fornasiero, L. Alberti, N. Bianchi and S. Bolognani, "Considerations
pp. 1053–1064, Mar. 2013. on Selecting Fractional-Slot Nonoverlapped Coil Windings," IEEE
[8] M. E. Beniakar, P. E. Kakosimos, and A. G. Kladas, “Strength Pareto Trans. Ind. Appl., vol. 49, no. 3, pp. 1316-1324, May-June 2013.
Evolutionary Optimization of an In-Wheel PM Motor with Unequal [27] B. Y. Qu, P. N. Suganthan and J. J. Liang, "Differential Evolution With
Teeth for Electric Traction,” IEEE Trans. Magn., vol.51, no.3, March Neighborhood Mutation for Multimodal Optimization," IEEE Trans.
2015, Art. ID 8102804. Evol. Comput., vol. 16, no. 5, pp. 601-614, Oct. 2012.
[9] P. Lazari, J. Wang, and L. Chen, “A Computationally Efficient Design [28] W. Ahmed, M. F. Shirazi, O. M. Jamil, M. H. Abbasi, “PSO with
Technique for Electric Vehicle Traction Machines,” IEEE Trans. Ind. Gompertz increasing inertia weight,” 8th IEEE Conference on Industrial
Appl., vol.50, no.5, pp.3203-3213, Sept-Oct. 2014. Electronics and Applications (ICIEA), pp.923-928, June 19-21, 2013.
[10] J. Nerg, M. Rilla, V. Ruuskanen, J. Pyrhönen, and S. Ruotsalainen, [29] A. Sarigiannidis, M. Beniakar, P. Kakosimos, A. Kladas, L. Papini, and
“Direct-Driven Interior Magnet Permanent-Magnet Synchronous Motors C. Gerada, "Fault Tolerant Design of Fractional Slot Winding Permanent
for a Full Electric Sports Car,” IEEE Trans. Ind. Electron., vol. 61, no.8, Magnet Aerospace Actuator," IEEE Trans. Transp. Electrif., in press,
Aug. 2014. doi: 10.1109/TTE.2016.2574947.
[11] G. Pellegrino, A. Vagati, P. Guglielmi, and B. Boazzo, “Performance [30] P. Ponomarev, I. Petrov, N. Bianchi and J. Pyrhönen, "Selection of
Comparison Between Surface-Mounted and Interior PM Motor Drives Geometric Design Variables for Fine Numerical Optimizations of
for Electric Vehicle Application,” IEEE Trans. Ind. Electron., vol.59, Electrical Machines," IEEE Trans. Magn., vol. 51, no. 12, pp. 1-8, Dec.
no.2, pp.803-811, Feb.2012. 2015.
[12] F. Parasiliti, M. Villani, S. Lucidi, and F. Rinaldi, “Finite-Element- [31] S.-M. Sue, and C.-T. Pan, “Voltage-Constraint-Tracking-Based Field-
Based Multiobjective Design Optimization Procedure of Interior Weakening Control of IPM Synchronous Motor Drives,” IEEE Trans.
Permanent Magnet Synchronous Motors for Wide Constant-Power Ind. Electron, vol.55, no.1, pp.340-347, Jan. 2008.
Region Operation,” IEEE Trans. Ind. Electron., vol.59, no.6, pp.2503- [32] E. Dlala, "Comparison of Models for Estimating Magnetic Core Losses
2514, June 2012. in Electrical Machines Using the Finite-Element Method," IEEE Trans.
[13] F. Hain, N. Domann, and M. Henke, "Interaction of winding topologies Magn., vol.45, no.2, pp.716-725, Feb. 2009.
and rotor structure in interior permanent magnet machines," in [33] J. Pyrhonen, H. Jussila, Y. Alexandrova, P. Rafajdus, and J. Nerg,
International Conference on Electrical Machines (ICEM), pp. 2031- "Harmonic Loss Calculation in Rotor Surface Permanent Magnets—
2037, Berlin, 2014. New Analytic Approach," IEEE Trans. Magn., vol.48, no.8, pp.2358-
[14] K. Yamazaki and M. Kumagai, "Torque Analysis of Interior Permanent- 2366, Aug. 2012.
Magnet Synchronous Motors by Considering Cross-Magnetization: [34] Z. Azar, Z. Q. Zhu, and G. Ombach, “Influence of Electric Loading and
Variation in Torque Components With Permanent-Magnet Magnetic Saturation on Cogging Torque, Back-EMF and Torque Ripple
Configurations,", IEEE Trans. Ind. Electron., vol. 61, no. 7, pp. 3192- of PM Machines,” IEEE Trans. Magn., vol.48, no.10, pp.2650-2658,
3201, July 2014. Oct. 2012.
[15] K. Yamazaki, M. Kumagai, T. Ikemi, and S. Ohki, “A Novel Rotor [35] A. G. Sarigiannidis, and A. G. Kladas, “Switching Frequency Impact on
Design of Interior Permanent-Magnet Synchronous Motors to Cope with Permanent Magnet Motors Drive System for Electric Actuation
Both Maximum Torque and Iron-Loss Reduction,” IEEE Trans. Ind. Applications,” IEEE Trans. Magn., vol.51, no.3, pp.1-4, Mar. 2015, Art.
Appl., vol.49, no.6, pp.2478-2486, Nov.-Dec. 2013. ID 8202204.
[16] P. B. Reddy, A. M. El-Refaie, K. K. Huh, J. K. Tangudu and T. M.
Jahns, "Comparison of Interior and Surface PM Machines Equipped
With Fractional-Slot Concentrated Windings for Hybrid Traction Athanasios G. Sarigiannidis (S’13) received the Diploma
Applications," IEEE Trans. Energy Convers., vol. 27, no. 3, pp. 593- degree in electrical and computer engineering from the
602, Sept. 2012. University of Patras, Greece, in 2010 and the Ph.D. degree
[17] V. Ruuskanen, J. Nerg, J. Pyrhonen, S. Ruotsalainen, R. Kennel, "Drive from the school of electrical and computer engineering,
Cycle Analysis of a Permanent-Magnet Traction Motor Based on National Technical University of Athens, Greece, in 2016.
Magnetostatic Finite-Element Analysis", IEEE Trans. Veh. Technol., Now he is a Postdoctoral researcher in the Laboratory of
vol. 64, no. 3, pp. 1249-1254, Mar. 2015. Electrical Machines and Power Electronics, National
[18] P. H. Nguyen, E. Hoang, and M. Gabsi, “Performance synthesis of Technical University of Athens, Greece. His current research involves permanent
permanent-magnet synchronous machines during the driving cycle of a magnet motor analysis, design optimization and control for traction applications.
hybrid electric vehicle,” IEEE Trans. Veh. Technol., vol. 60, no. 5, pp. Dr. Sarigiannidis is a member of the Technical Chamber of Greece (TCG).
1991–1998, Jun. 2011.
[19] V. Schwarzer and R. Ghorbani, “Drive cycle generation for design Minos E. Beniakar received the M. Eng. and Ph.D. degree
optimization of electric vehicles,” IEEE Trans. Veh. Technol., vol. 62, in electrical and computer engineering from the National
no. 1, pp. 89–97, Jan. 2013. Technical University of Athens, Greece, in 2008 and 2014
[20] M. E. Beniakar, A. G. Sarigiannidis, P. E. Kakosimos, and A. G. Kladas, respectively. His research involves design optimization of
“Multi-objective Evolutionary Optimization of a Surface Mounted PM electric motors for aerospace and electric traction
Actuator with Fractional Slot Winding for Aerospace Applications,” applications and industrial drives.
IEEE Trans. Magn., vol.50, no.2, pp.665-668, Feb. 2014, Art. ID Dr. Beniakar is currently with ABB Corporate
7016404. Research, Västerås, Sweden.

0018-9545 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
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. Citation information: DOI 10.1109/TVT.2016.2631161, IEEE
Transactions on Vehicular Technology
14

Antonios G. Kladas (SM’10) was born in Greece in 1959.

He received the Diploma in Electrical Engineering from the
Aristotle University of Thessaloniki, Thessaloniki, Greece,
in 1982 and the D.E.A. and Ph.D. degrees from the
University of Pierre and Marie Curie (Paris 6), Paris,
France, in 1983 and 1987, respectively.
From 1984 to 1989, he was an Associate Assistant with
the University of Pierre and Marie Curie. During the period 1991 to 1996, he
joined the Public Power Corporation of Greece. Since 1996, he has been with
the Department of Electrical and Computer Engineering, National Technical
University of Athens, Athens, Greece, where he is currently a Professor. His
research interests include transformer and electric machine modeling and
design, analysis of generating units by renewable energy sources and
industrial drives.

0018-9545 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.