Method for Multi-objective Optimized Designs of Surface Mount
Permanent Magnet Motors with Concentrated or Distributed Stator
Windings
Y. Duan, R.G. Harley and T.G. Habetler
Georgia Institute of Technology
Atlanta, GA 30332-0250
Abstract- A fast and efficient multi-objective design method is
developed for comparison of the traditional Distributed Winding
(DW) and the more popular Concentrated Winding (CW)
configuration for Surface Mount Permanent Magnet (SMPM)
machines. First an analytical design model previously developed
for DW machines is adapted for CW machines and the accuracy
is verified by Finite Element Analysis (FEA); then the Particle
Swarm Optimization (PSO) method is applied to optimize the
design. With an objective function defined by the user to
emphasize machine performance indexes, such as weight, volume,
efficiency and so on, the PSO optimized designs of the two
winding type machines provide valuable insight on winding
choices. Two example objective functions and their respective
optimized designs for both winding types are illustrated and
analyzed. Another merit of the proposed method is that it is FEA-
free, fast and efficient, which will save significant time and energy
for machine designers.
I. INTRODUCTION
Recently there has been a growing interest in PM motors
with concentrated windings, mainly due to low cost of
manufacture. Comparisons of the SMPM motors with
distributed winding (Fig 1(b)) and concentrated winding (Fig
1(a)) have been reported either by designing two motors with
exactly the same rotors [2], or by qualitative analysis [3].
However, the design of PM machines, as in the case of any
machine, is an area where suitable results can be obtained with
an almost infinite number of combinations, and depending on
such factors as the intended application, different designs could
be generated. On the other hand, a machine analysis tool, such
as FEA, is time-consuming. A fast, efficient and
comprehensive analysis tool of the two winding types, and
with optimization in terms of the users requirements, is
therefore a more useful tool for a machine designer,
particularly for initial designs. FEA can always be used later
on to check for any violations.
The SMPM machine design optimization method in [1] is
first adapted for the concentrated winding SMPM, and then the
optimization results are compared for motors with both
winding types. The motors to be designed have the same rated
condition: 15 kW, 60 Hz and 1800 rpm. A comprehensive
understanding of the merits and disadvantages of the two
winding types can thus be realized from the quantitative
comparison results.
978-1-4244-4252-2/09/$25.00 2009 IEEE
(a) (b)
Figure 1: Illustration of the (a) concentrated winding and (b) distributed
winding
II. ANALYTICAL DESIGN MODEL
Before any optimization algorithm can be applied, a model
of the machine to predict the machine performance needs to be
developed first. The Finite Element Analysis (FEA) method is
useful to analyze a particular design and obtain detailed
information of the motor, such as cogging torque. However, a
machine must be first fully designed before FEA analysis.
Moreover, FEA takes a significant amount of computation time.
Even optimizing with only 15 variables using FEA takes 273.1
hours [6], but 15 variables are not nearly enough to fully
describe a machine design. Hence, an analytical design model
which is suitable for optimization is developed for both
Distributed and Concentrated Winding motors.
A. Magnet Design
The magnet design plays an essential role in the design of
surface mount permanent magnet machines. On one hand,
using more permanent magnet will increase the power density
of the motor but this effect will diminish as the magnet
increases. On the other hand, the high material cost of the
permanent magnet plays a large part of the total machine
because of the high price of the rare-earth permanent magnet.
Trade-off is made by the designer depending on the machine
performance requirement, the availability of the material and
the price. Hence, the magnet thickness MagThck is chosen as
one of the design variables in the optimization.
Besides the magnet thickness, the magnet pole coverage, the
magnetization and skew of the rotor magnets also affect the
performance of the motor, especially in the cogging torque and
the ripple torque. The magnet pole coverage, which stands for
how much the magnet covers the total pole pitch, is a fractional
323
number between 0 and 1, with 1 to stand for that the magnet torque ripples are also reduced to as low as 15% (peak to peak
covers the total pole area. The commonly used magnetizations value) for both distributed and concentrated winding machines.
are radial magnetization and parallel magnetization.
Assuming infinite permeability of the steel core and
neglecting the slotting effect, the MMF waveform produced by
the permanent magnet for typical SMPM machines can be
calculated according to [13]. The magnitude for the
fundamental, 3rd, 5th and 7th harmonics of is plotted in Fig. 2
for different pole coverages and magnetizations:

(a) Cogging Torque Waveforms

(b) Torque ripples at rated condition

Figure 2: Relative magnitude of MMF harmonics Figure 3: Cogging torque and torque ripple waveforms

Fig. 2 shows that as the pole coverage increases, the magnitude

B. Machine Analytical Design Model Flow Chart
of the fundamental increases slowly. When the pole coverage
After the magnet configuration has been determined, the
coefficient is around 0.83, the magnitude of the 5th and 7th
next is to determine the sizing of the machine, the stator slot
harmonics is close to zero. In addition, the parallel and radial
shape, and the winding configuration. The geometry variables
magnetization produces approximately the same magnitude for
are shown as below in Figure 4 with the concentrated winding
the fundamental, while the parallel magnetization has a much
motor as an example.
smaller 3rd harmonics than the radial magnetization when the
DisSYoke
pole coverage coefficient is equal to 0.83. Fewer harmonics in
the MMF waveforms result in fewer harmonics in the airgap
DisSGap
flux density waveform, eventually there will be less torque
ripples with sinusoidal excitation. For the above reasons, a pole
DisSRGap
coverage of 0.83 and parallel magnetization are chosen for the
MagThck DisRYoke
purpose of minimizing the torque ripples. Similar choices are
also reported in the literature from other approaches [8].
For the motor to be designed in this paper, 1800 rpm and 60
Hz indicate that this is a 4 pole machine. Although this pole
number makes it difficult to minimize the cogging torque by Bs0

choosing proper pole/slot number combination [7] [8], the slot Hs0
Hs1
opening and tooth tip is carefully designed to reduce the Tw
Bs1
Hs2
cogging torque. Rs
Bs2

Fig 3 shows the typical cogging torque and torque ripples at

rated condition simulated by FEA after careful design of the
magnet pole coverage, magnetization, slot opening and tooth
tip. The stator slot numbers for the distributed and concentrated
winding motor are 24 and 6, respectively. It can be observed
that the peak-to-peak value of cogging torque has been
effectively reduced to only around 5% of the rated torque. The
Figure 4: Geometry parameter of the motor
The flowchart in Fig. 5 summarizes the procedure developed
for designing the SMPM machines, which is applied to both
distributed and concentrated winding machines. The merit of
this design flow is that the design variables are reduced to only

324
3: the stator diameter in the airgap side (DiaSGap), the motor
axial length (Length) and the magnet thickness (MagThck).
The stator slot fill factor, the stator winding current density, the
output power and the rated voltage are the design constraints.
With the values of the design variables and the constraints
determined, a design of SMPM is generated by running the
flowchart.
DiaSGap
Length
ThichMag
Tooth Width
AirGap Flux
Stator and Rotor
Yoke Thickness
Density
Number of
turns per
Back EMF
Output
Power
Current
Current
Desnity
Slot Fill Factor
Parameters
Loss
Figure 5: Flowchart of the SMPM analytical design model
Details on this flowchart on the distributed winding machine
have been explained in [1]. However, some important steps
will be repeated here and the adaption of this flowchart to the
concentrated winding motor will be highlighted.
The analytical calculation starts from the field solution in the
airgap region. The radial component of the magnetic field in
the airgap produced by the permanent magnet can be solved
and expressed in a Fourier series as follows [4]
np
Bgap mag ( , r ) =
n =1,3,5...
Bn (r ) cos(
2
where Bn is the n-th spatial harmonic component of the airgap
flux density distribution, r is the airgap radius, is the angular
position with reference to the center of a magnet pole and p is
the number of poles.
The back emf per coil turn of the motor can then be
calculated. The stator armature reactance and leakage reactance
with respect to one coil turn is also obtained according to [9].
With the back emf, inductances, output power and phase
voltage specified or calculated, the number of turns can be
determined.
In this design, non-oriented electrical steel with lamination
factor of 0.95 is used for both machines. The lamination gauge
is M19_29G from AK steel [13]. The flux densities in the
stator teeth, stator core and rotor core are specified to be
around 1.4 T, 1.5 T and 1.5 T, respectively. The number of
turns, slot fill factor and current density will then give the tooth
width, tooth length, stator and rotor back iron thickness, by
meeting the flux density specifications at the rated condition. A
typical copper slot fill factor of 40% is chosen for the
Inductance distributed winding machine and a higher slot fill factor of
about 65% [2] is chosen for the concentrated winding machine.
For comparison reasons, the current density for both motors is
chosen to be around 5 A/mm2, which is in the typical range for
15 kW a SMPM motor.
After all the geometry and electrical design parameters are
phase
calculated, the machine performance is calculated, including
the weight, volume and efficiency. Fig. 6 shows how the end
winding is modeled, where the coil is actually the total coils of
one winding layer. The total length of the wire and the stator
winding resistance can be then calculated accordingly after the
end winding is defined.
Design
Weigth Volume
(a) (b)
Figure 6: End winding for (a) Distributed Winding and (b) Concentrated
Winding
The weight of the motor is taken to be the active weight,
which is the sum of the stator core, rotor core, magnets and the
winding. When calculating the volume, the motor is estimated
by the cylinder with the same cross sectional area of the motor
but larger axial length, due to the extra space by the end
) (1) winding.
The loss of the motor is primarily comprised of the copper
loss and the core loss. Due to a relatively low operating
frequency of the motors 60 Hz, the copper loss is simply the
I2R losses. The core loss can also be simply estimated by
looking up the material loss curve from the data sheet given by
the supplier. The efficiency is thus obtained by the output over
the sum the output power and the losses.
C. Verification of the Analytical Design Model by FEA
Before proceeding to optimize the motor design with the
analytical design model, the accuracy of the design needs to be
evaluated by simulating motor design samples in FEA and
325
comparing the results. In the actual design process, the importance is attached to this index. The two objective
optimized design is also verified by FEA to confirm. The functions are defined in (2), where obj1 pays more attention to
accuracy of this design model for the conventional distributed the weight and volume and obj2 gives more importance to the
winding has been shown in [1] and typical results for the efficiency and torque per ampere.
concentrated winding motor are compared below: obj1 output = volume * 20000 + 2 * Weight + (1 Eff ) * 200
+ WtMagnet *5 TperA *5
Phase current waveform obj 2 output = volume *10000 + Weight + (1 Eff ) *1000
+ WtMagnet *5 TperA * 20 (2)
FEA Analytical model
40
30 B. Particle Swarm Optimization
20
The analytical model can now be used to optimize the values
current / A

10
0 of the airgap diameter and motor length according to the users
-10 0 0.005 0.01 0.015
defined objectives, such as weight or efficiency, with a given
-20
-30
performance requirement. As indicated in Figure 5, the number
-40 of turns per phase is obtained by solving phasor diagrams with
time / s numerical methods and it is difficult to get a solution in direct
(a) analytical form. Furthermore, this model is actually a hybrid
nonlinear model of continuous and discrete variables because
Torque waveform
the number of turns can only be integer and the wire size needs
86 FEA Analytical model to be one of the standard AWG sizes. The above reasons make
84 it impossible to calculate the gradient, which is essential in
Torque / Nm

82
80
78
76
74
0 0.005 0.01
Time / s
(b)
Figure 7: FEA result of the motor with concentrated winding designed by the
analytical model (a) Phase current waveform at rated condition (b) Output
torque waveform at rated condition
Notice that the analytical design model only considers the
fundamental thus the resulting torque is a constant. However,
the average torque by the FEA and the analytical design model
agrees with each other as shown in Fig. 7(b). The current
waveform in Fig. 7(a) shows the high accuracy of predicting
the phase current by the analytical design model, which
indicates good accuracy of the efficiency calculation.
III. MULTI-OBJECTIVE OPTIMIZATION WITH PARTICLE SWARM
A. Objective Function
Before the optimization algorithm is run, the objective
function needs to be defined. To account for various
requirements in the permanent magnet motor design, the
objective functions used here have five performance indexes,
namely: volume (m3), weight (kg), efficiency, weight of the
magnets (kg) and the torque per ampere at the rated condition
(Nm/Arms). For the machines to be designed, the weight,
volume, efficiency, torque/ampere and magnet weight have
typical range of around 20-50 kg, 0.002-0.005 m3, 0.5-1, 2-5
Nm/Arms and 0.5-2 kg, respectively. Each term in the
objective function is first normalized to be about the same
magnitude, and then the weighting factors or coefficients
before each index are further tuned to represent how much
traditional nonlinear optimizing techniques, such as the
Newton method [10], and the computational techniques are
therefore required. Particle Swarm Optimization (PSO) is
0.015 proposed and applied in this design optimization and shows its
effectiveness in solving such problems as hybrid models.
PSO is an evolutionary computation technique that was
developed in 1995 by Kennedy and Eberhart [11] [12]. The
algorithm can be as summarized as follows:
Define the solution space: select the parameters that need
to be optimized. Here, the parameters are DiaSGap,
Length and MagThck
Define a fitness function, which is the objective function
defined above
Initialize Random Swarm Location and Velocities
Systematically fly the particles through the solution
space: The following steps are performed on each particle
individually:
o Evaluate particle fitness: compare to global best
gbest and personal best pbest
o Update the particles velocity according to the
relative values of pbest and gbest, using the following
expression:
vn = * vn + c1rand () *( pbest ,n xn ) + c2 rand () *( gbest ,n xn )
o Move the particle: The velocity is applied for a
given time-step t, which is usually chosen to be 1,
and the position is updated to
xn = xn + t * vn
Repeat process starting at step 4. In this way the particle
moves for discrete time intervals until the termination
criteria are met.

326
 Mtk

Mtk
Mtk

(a) (b) (c)

Lst DiaG Lst DiaG Lst DiaG
Figure 8: Particle positions as iteration N increases (a) N=0, obj1=138.5 (b) N=40, obj1=129.7(c)N=100, obj1=123.9

TABLE 1. MULTI-OBJECTIVE OPTIMIZATION RESULT

Objective Function 1 Objective Function 2

CW DW CW DW
Design1 Design2 Design1 Design2 Design1 Design2 Design1 Design2

Weight / kg 28.5 27.9 30.0 29.4 32.12 32.39 32.02 33.23

3
Volume / m 0.0031 0.0032 0.0038 0.0037 0.0043 0.0041 0.0048 0.0047

Efficiency 93.31% 93.34% 94.69% 93.68% 95.08% 94.89% 95.88% 95.85%

Torque/Ampere 2.79 2.79 3.54 2.79 3.79 3.74 3.73 3.75

(Nm/Arms)
Magnet Weight / kg 0.685 0.780 0.95 0.600 1.48 1.26 1.12 1.04
Objective Function 122.5 123.2 134.3 134.4 56.38 56.42 52.39 52.17

One typical result of running PSO with obj1 for the overall performance than the distributed winding one. However,
concentrated winding machine appears in Fig. 8. As the the opposite is true where a higher efficiency and torque per
number of iteration increases, the 6 particles start randomly in ampere are desired. These results are a helpful guide for the
solution space and travels in the solution space. After 100 designer to understand and make decisions on winding choices
iterations all 6 particles converge to the correct optimal and trade offs.
solution. Table 2 shows the detailed machine design information of all
the Design 1 for each winding type and objective function.
IV. RESULTS AND COMPARISONS TABLE 2 CW AND DW OPTIMIZED DESIGN
Both distributed and concentrated winding type machines
have the same rated condition to be 15 kW, 60 Hz, 1800 rpm
Obj1, CW Obj1, DW Obj2, CW Obj2, DW
and an excitation voltage of 480 V (line to line). The
DiaSGap (mm) 82.7 83.3 84.8 77.1
distributed winding motors have 24 slots and double layer, full
Length (mm) 82.3 88.1 139.9 148.1
pitch winding. The concentrated winding motors have 6 slots
MagThck (mm) 5.76 7.56 7.63 5.62
and double layer winding with 120 electrical degree coil pitch.
Stator outer
For comparison reasons, the airgap length for both winding 220.1 235.0 198.9 203.3
diameter (mm)
types is 1 mm. NdFeB with a residual flux density of 1.23 T
and relative permeability of 1.01 is used as the rotor magnet Number of turns
208 218 158 154
material. The magnet pole coverage are both 0.83 and parallel per phase
magnetization is used. No skew is applied to the magnet. Winding Factor 0.866 0.9659 0.866 0.9659
PSO is run for both winding type machines with obj1 and Tooth width (mm) 27.0 7.19 28.8 6.25
obj2, respectively. The optimized solution for each type lies in Slot opening(mm) 3 1 3 1
a region and the particles converge to one of the solution points Phase
0.33 0.42 0.39 0.31
in the region. Columns under Design 1 and Design 2 in Table 1 resistance()
are two typical PSO results for each winding type and Synchronous
7.31 6.82 6.23 6.41
objective function. Table 1 shows that while values of each of Reactance()
the objective functions fall within a close range, the differences Armature Leakage
3.37 2.30 3.18 1.92
between the concentrated and distributed winding are clear. Reactance()
When more importance is given to weight and volume as in
obj1, the concentrated winding motor has a significantly better

327
 Comparing the CW and DW design in Table 2, it is observed
that the CW motor have much smaller outer diameter for both
designs under obj1 and obj2, which leads to less use of stator
steel and thus less weight and volume. This is due to the higher
slot fill factor. The short end winding, another advantage of the
CW, helps to reduce the phase resistance. This effect is
observed in Table 2 under obj2 when the efficiency is put high
attention. Though the CW has a lower winding factor, the CW
design is only slightly worse (less than 1% of efficiency) than
the DW.
The comparison results show that the CW designs have
superior performance in terms of weight and volume. In
addition, the lower phase resistance helps to diminish the
negative impact of the relatively low winding factor. These
PSO optimization results agree with the physics reasoning
behind the analytical equations. The result also shows that the
particle swarm intelligence is able to correctly understand the
underlying physical behavior of the objective system in the
searching process and find the best solution.
V. CONCLUSION
An analytical model is developed for a SMPM motor with
either distributed or concentrated stator windings and PSO is
applied to optimize the design of each motor. The optimized
designs with the two winding types are compared
quantitatively for certain objective functions, which can be
defined by the designer according to the application. This tool
provides optimized designs of a SMPM motor with both types
of windings in minutes on an ordinary desktop computer,
which saves significant time and energy for the machine
designer, compared to using FEA for every choice of
parameter.
ACKNOWLEDGMENT
Financial support for this work from the Office of Naval
Research and Grainger Center for Electric Machinery and
328
Electromechanics, at the University of Illinois, Urbana
Champaign, is gratefully acknowledged.
REFERENCES
[1] Y. Duan, R. G. Harley and T. G. Habetler, "Multi-objective Design
Optimization of Surface Mount Permanent Magnet Machine with Particle
Swarm Intelligence", IEEE SIS 2008, St. Louis, pp. 1-5, September 2008.
[2] F. Magnussen, P. Thelin and C. Sadarangani, Performance evaluation of
permanent magnet synchronous machines with concentrated and
distributed windings including the effect of field-weakening, 2nd IEE
International Conference on Power Electronics, Machines and Drives,
Edinburgh, UK, Vol.2. pp. 679-85, 2004
[3] A. M. El-Refaie, T. M. Jahns and D. W. Novotny, Analysis of Surface
Permanent Magnet Machines with Fractional-Slot Concentrated
Windings, IEEE Tran. Energy Conversion, Vol. 21, No. 1, pp. 34-43,
Mar. 2006
[4] Z. Q. Zhu, Improved analytical model for predicting the magnetic field
distribution in brushless permanent-magnet machines, IEEE Tran.
Magn., Vol. 38, No. 1, pp. 229-238, Jan. 2002.
[5] J. Robinson and Y. Rahmat-Samii, Particle Swarm Optimization in
Electromagnetics, IEEE Transactions on Antennas and Propagation, Vol.
52, No. 2, pp. 397 407, February 2004G. Eason, B. Noble, and I.N.
Sneddon, On certain integrals of Lipschitz-Hankel type involving
products of Bessel functions, Phil. Trans. Roy. Soc. London, vol. A247,
pp. 529-551, April 1955.
[6] W. Ouyang, D. Zarko and T.A. Lipo, Permanent Magnet Machine
Design Practice and Optimization, 2006 IEEE conference on Industrial
Applications, Glasgow, UK, pp. 1905 1911, 2006.
[7] Zhu, Z.Q., Howe, D., Influence of design parameters on cogging torque
in permanent magnet machines, IEEE Transactions on Energy
Conversion, Volume 15, Issue 4, pp. 920-925, Dec. 2000
[8] Jahns, T.M., Soong, W.L., Pulsating torque minimization techniques for
permanent magnet AC motor drives-a review, IEEE Transactions on
Industrial Electronics, Vol. 43, Iss. 2, pp. 791-794, April 1996
[9] Z.Q. Zhu, D. Howe, Instantaneous Magnetic Field Distribution in
Brushless Permanent Magnet dc Motors, Part II: Armature-Reaction
Field, IEEE Transactions on Magnetics, Vol. 29, No. 1, pp. 136 142,
January 1993
[10] Bazaraa, Mokhtar S. and Shetty, C. M., Nonlinear programming: Theory
and algorithms, John Wiley & Sons. ISBN 0-471-78610-1.
[11] J. Robinson and Y. Rahmat-Samii, Particle Swarm Optimization in
Electromagnetics, IEEE Transactions on Antennas and Propagation, Vol.
52, No. 2, pp. 397 407, February 2004.
[12] Y. del Valle, G.K. Venayagamoorthy, S. Mohagheghi, J.-C Hernandez
and R.G Harley, Particle Swarm Optimization: Basic Concepts, Variants
and Applications in Power Systems, IEEE Transactions on Evolutionary
Computation, Vol. 12, Issue 2, Apr. 2008.
[13] AK Steel, Product Data Bulletin, Selection of Electrical Steels for
Magnetic Cores, 2000