IEEE TRANSACTIONS ON MAGNETICS, VOL. 43, NO.

1, JANUARY 2007 51

Optimal Design and Control of a Wheel Motor

for Electric Passenger Cars
Yee-Pien Yang and Down Su Chuang

Department of Mechanical Engineering, National Taiwan University, Taipei 106, Taiwan, R.O.C.

An optimal design and control technology of a wheel motor is proposed for small electric passenger cars. The axial-flux sandwich-type
disc motor is designed with a rotor embedded with neodymium–iron–boron (NdFeB) magnets and two plates of stators, and is directly
mounted inside the wheel without mechanical transmission and differential gears. Sensitivity analyses are performed to choose critical
design parameters, which are the most influential in design objectives, to maximize the driving torque, efficiency, rated speed, and to
minimize the weight of motor under various constraints of size, materials, and power sources. The optimal driving current waveform is
proven to be the same as the fundamental harmonic of the back electromotive force to produce maximum torque with least ripples. The
finite-element refinement results in the motor prototype with a maximum torque over 38 kg m and a corresponding torque density of
about 1.72 kg m/kg at the maximum allowable phase current of 50.25 A (rms). Two such rear driving wheels are able to drive a 600 kg
passenger car to accelerate from 0 to 40 km/h in 5 s on a 15 degree incline. This dedicated wheel motor is applicable to pure or hybrid
electric vehicles as a promising solution to the direct-driven electric vehicle.
Index Terms—Axial-flux wheel motor, electrical vehicle, optimal driving current waveform, sensitivity analysis.

I. INTRODUCTION of 5 kW to drive a car of gross weight of 600 kg through trans-

mission gears. Chen and Tseng [3] proposed a radial flux di-
rect-driven brushless dc motor with 20 teeth on the stator and 22
E LECTRIC VEHICLES (EVs) have imperatively attracted
quite a few researchers and automobile companies in de-
veloping more efficient and reliable propulsion systems, be-
magnets on the rotor. Compared with conventional three-phase
motors, it had significant improvement on the copper weight,
cause of the increasing concerns about natural environment and ohmic loss, volume, torque ripple, and cogging torque. Wije-
growing shortages of petroleum resources. Traditional power nayake et al. [4] designed, using an optimization procedure,
systems for EVs consist of battery, electric motors with drives, a 216 V disc-type permanent-magnet brushless dc motor for
transmission gears, and differentials to the wheels. Each sub- a direct-driven EV, whose rated power was 100 hp and max-
system converts chemical or electrical energy to mechanical imum speed was 3200 rpm. Patterson and Spèe [5] initiated an
power to drive the vehicle, while consuming energy through axial flux permanent-magnet brushless dc motor for their solar
the dissipation components of windage, friction, magnetic hys- energy vehicle. The curb weight of their electric vehicle was
teresis, and ohmic loss. Various designs for the EVs have been 260 kg, the rated power was between 1–2 kW at 72 km/h, and
proposed, such as those powered by hybrid resources, fuel cells, the output power was 3 kW on the 6 degree slope. A similar
and solar cells, while new concepts of motor designs and their disc-type motor was proposed by Eastham et al. [6], which con-
optimal driving patterns have attracted substantial attention in sisted of three stator plates and four rotors in their simulations.
the industry. Alternatively, Uematsu and Wallace [7] chose the reluctance
In recent years, induction motor drives are preferred for motor for the EV, and found that the motor had to be operated
the EV propulsion system due to their low cost, high relia- at 6000–8000 rpm for enough rated torque at 140–160 Nm and
bility, high-speed properties, and manufacturing facilitation. 100 kW output power. Oh and Emadi [8] investigated a com-
Concurrently, the permanent-magnet brushless dc motors (in- mercial axial flux motor for different drive cycles of hybrid elec-
direct-driven and direct-driven) featuring compactness, low tric vehicles by varying the air gap to extend the range of oper-
weight, and high efficiency, have become an alternative for ating speed and improve the efficiency by the method of hard-
the EV propulsion systems [1]. The latter types of motors are ware in the loop. An interesting double-stator starter generator
also called wheel motors or hub-in motors, which are directly for the hybrid EV was introduced in [9]; it works as a motor at
mounted inside the wheels so that the transmission gears and low speeds when two stator windings are in series, while two
differentials are eliminated with associated energy loss. stators shift relatively to produce various back electromotive
Several successful designs and/or applications of the direct forces in the generator mode. Lately, a different configuration of
and indirect-driven motors and drives on the EV have been pro- an axial-flux motor was presented with a stator sandwiched by
posed. Chan [2] designed a five-phase radius flux brushless dc two permanent-magnet rotors [10], but only the computer-aided
motor with 22 NdFeB magnets on the rotor, which had a power design and finite-element analysis were provided. Each of the
above researches on direct-driven wheel motor has its own spec-
ifications and configurations for their electric vehicles in terms
of gross weight, driving voltage, maximum speed, rated torque,
Digital Object Identifier 10.1109/TMAG.2006.886153 and power source.

0018-9464/$25.00 © 2006 IEEE

52 IEEE TRANSACTIONS ON MAGNETICS, VOL. 43, NO. 1, JANUARY 2007

TABLE I TABLE II
SPECIFICATIONS OF AN ELECTRIC CAR SPECIFICATIONS OF A SINGLE WHEEL

This paper proposes a systematic optimal design method-

ology for the four-phase, disk-type, permanent-magnet,
axial-flux brushless dc wheel motor for electric passenger cars.
The specifications in Section II are based on the small size,
dual seats, and urban driving requirements. Section III briefly
describes a simplified two-dimensional (2-D) magnetic circuit
model for the preliminary optimal design of wheel motor.
Section IV starts with the sensitivity analysis to determine the
most influential parameters to the motor performance as the
design variables. Then, the multifunctional objective system
tool proceeds to optimize a set of cost functional subject to
the constraints on the design variables and physical properties
to get the preliminary motor geometries. In Section V, further
refinement for the reduction of weight and torque ripples is
performed by the finite-element method, where the thermal
analysis is also investigated for heat dissipation by fins over
motor cover under normal and severe driving conditions.
Finally, Section VI shows that the back electromotive force
(back EMF) provides significant contribution to the optimal
current driving pattern increasing the maximum torque as well
as the torque density. Section VII provides a summary and
conclusions.

II. SPECIFICATIONS
The specifications of a small electric passenger car are il-
lustrated in Table I. The road load on the vehicle consists of
three components—aerodynamic drag force , rolling resis- Fig. 1. Stator and rotor of wheel motor (a) assembly, (b) geometries, and
tance force , and climbing force , which are expressed as (c) series and parallel phase windings between two stator plates.
[11]

(1)
The stator and rotor assembly of the wheel motor and their
(2) geometries are shown in Fig. 1. It is a disk type, axial flux,
(3) permanent-magnet, brushless dc wheel motor, with a rotor disk
sandwiched between two stator plates. The stator toroids are
where is the air density, is the frontal area of the car,
wound and laminated by a continuous sheet of electric steel; its
is the aerodynamic drag coefficient, is the relative vehicle ve-
fan-shaped teeth are factorized by punching slots with varying
locity to the head wind, is the rolling friction coefficient,
pitch for different radius of each layer. The fan-shaped mag-
is the gross mass or vehicle mass with payload, and is the
nets facing the stator windings contribute a main magnetic flux
inclination angle of the road. The force required to reach the
flowing through two air gaps between the stator and rotor along
prescribed acceleration by overcoming the road load is
the axial direction. The windings on each side of the stator can
be connected in series or parallel as shown in Fig. 1(c). The
(4) serial winding connection splits the line voltage into half on ei-
ther side of the stator and is suitable for low-speed, high-torque
Based on the specifications, the required motor torque and the driving cycle, while the parallel winding connection holds the
corresponding speed under various vehicle operation conditions maximum voltage that the battery source can supply and is used
are calculated and summarized in Table II. for high-speed, low-torque driving cycle.
YANG AND CHUANG: OPTIMAL DESIGN AND CONTROL OF A WHEEL MOTOR FOR ELECTRIC PASSENGER CARS 53

Fig. 3. 2-D model and magnetomotive forces.

current , coercivity of the magnet, and thickness . Based

on the assumptions, the magnetic coenergy stored in the air gap
is expressed as

Fig. 2. Configuration of wheel motor on passenger car.

(5)

The wheel motor is directly mounted on the chassis with the

where is the air-gap length, is the overall
suspension and brake, and the wheel is installed on the shaft ro-
magnetomotive force distribution from the rotor magnets and
tating with the rotor as proposed in Fig. 2. The stator is covered
stator windings, is the permeability of free space, denotes
with the outer case molded with heat dissipation fins to dissi-
the rotor shift, represents the peripheral coordinate along the
pate heat by natural air convection. The final shape of the wheel
circle of the average radius , and and
motor is designed to meet the specifications through a multi-
are, respectively, the inner and outer radii of the motor. The
functional optimization scheme.
torque resulted from the variation of the coenergy with respect
to the rotor shift is then given by
III. MAGNETIC CIRCUIT MODELING OF MOTOR
The optimal design of the wheel motor is to achieve a set of
prescribed objectives in terms of motor torque, torque density, (6)
speed, and efficiency. The excitation current from stator wind-
ings and the magnetic flux from rotor magnets generate magne- All the above functions are expressed explicitly in terms of
tomotive forces, which create the magnetic flux density distribu- motor geometric dimensions, material properties, and electric
tion and magnetic energy in the air gap. The change of magnetic specifications of the wheel motor, as described in Appendix A.
energy with respect to the rotor shift produces torque distribu- The speed of the motor can be obtained by the electrical
tion over the air-gap rings. equation
Based on the assumptions of material linearity and the
collinearity of flux and field densities, the magnetic circuit
model is used to describe the torque produced in the motor. It (7)
is also necessary to make three additional assumptions.
1) The motor is operated in the linear range of the – curve where is the back EMF, , and are, respec-
of the magnetic material. tively, phase inductance, phase current, phase resistance, and
2) The air-gap reluctance of the slotted stator structure is ap- phase voltage. The phasor expression is
proximated by the effective air-gap length with Carter’s
coefficient.
3) The flux flows straight across the air gaps between the (8)
stator and rotor, ignoring the fringing flux for simplified
analysis. where the time differentiation is replaced by introducing the
The three-dimensional (3-D) motor structure can be simpli- electrical speed . For simplification, no power loss from the
fied to a 2-D configuration, and its two-side topology is cut into power source to the mechanical output is assumed that
half for facilitating the magnetic circuit analysis, as shown in , where is the average torque, is the
Fig. 3. The fan-shaped magnets and stator teeth are mapped into number of magnets, and denotes the mechanical rotational
rectangular ones as the arc is transformed to a straight line in the velocity of the motor. Then the maximum speed is obtained by
2-D linear motor mode. The electromotive forces from stator
windings and magnets are expressed, respectively, as
and , in terms of number of turns per tooth, (9)
54 IEEE TRANSACTIONS ON MAGNETICS, VOL. 43, NO. 1, JANUARY 2007

TABLE III TABLE IV

DESIGN VARIABLES FOR SENSITIVITY ANALYSIS PREDETERMINED PARAMETERS

where

IV. MULTIFUNCTIONAL OPTIMIZATION

The performance of the wheel motor is usually evaluated
based on its efficiency, weight, maximum torque, torque ripple,
and time response, which are also known as the design objec- Fig. 4. Sensitivity of torque density versus inner radius.
tive functions or performance indices, describing mechanical
and electrical dynamics in terms of motor geometries, magnetic
materials, and driving conditions. First, the sensitivity analysis
is required to determine the derivatives of the objective func-
tions with respect to the parameters of interest, then a set of
design variables are determined. In this paper, the sensitivity
derivatives are not obtained explicitly from the equation, but are
numerically investigated. Second, the multifunctional optimiza-
tion system tool is used to search for the optimal values of the
design variables that maximize the performance indices, sub-
Fig. 5. Sensitivity of torque density versus outer radius.
ject to various constraints on the limitation of space and mate-
rial properties.

A. Sensitivity Analysis
The purposes of the sensitivity analyses are as follows.
1) The designer may want to discard those design variables
with the least sensitivities of torque, speed, torque density,
torque ripple, and efficiency of the motor.
2) The designer may keep those design variables constant
with sensitivities, which are linear, or monotonic functions. Fig. 6. Sensitivity of torque density versus magnet thickness.
3) Only those design variables that are not included in the
above two cases are retained for the subsequent optimal
design. index, which is defined as the ratio of the variation of motor per-
Table III lists all the variables for the sensitivity analysis, formance and the variation of design variable.
while other motor parameters are predetermined in Table IV 1) Torque Density: Figs. 4–9 show selected illustrations of
according to physical facts and previous design experience. It the sensitivity curves of torque density versus motor dimen-
would be a time and space consuming process to illustrate all sions. It is not surprising that the torque density of the wheel
the sensitivity curves, though it is worth providing sufficient in- motor increases upon decreasing its inner radius and increasing
formation for selecting critical design variables. This section its outer radius, as shown in Figs. 4 and 5. The thicker the
illustrates some sensitivity curves of torque density and effi- magnet, the more the magnetic energy stored and the larger the
ciency with respect to a single variable while others are fixed at torque produced as shown in Fig. 6. The magnet fraction, de-
their prescribed nominal values. The final decision of the design fined as the ratio of the width of magnet and the rotor pitch,
variables for optimization is made by evaluating the sensitivity affects the torque density in the way as shown in Fig. 7. Fig. 8
YANG AND CHUANG: OPTIMAL DESIGN AND CONTROL OF A WHEEL MOTOR FOR ELECTRIC PASSENGER CARS 55

Fig. 7. Sensitivity of torque density versus magnet fraction. Fig. 11. Sensitivity of torque density versus tooth thickness.

Fig. 12. Sensitivity of efficiency versus numbers of winding layers.

Fig. 8. Sensitivity of torque density versus fraction of slot opening.

Fig. 13. Sensitivity index of maximum torque.

Fig. 9. Sensitivity of torque density versus air-gap length.

per tooth simultaneously increase the maximum torque and

motor weight, and they must reach a compromise by evaluating
the motor efficiency. An example is shown in Fig. 12, where
the efficiency curve reaches a limit as the layers of winding are
increased.
2) Sensitivity Indices: The sensitivities are investigated by
the maximum torque, torque density, torque ripple efficiency,
and maximum speed of the motor with respect to design vari-
ables. For summarizing the results and making a final decision
Fig. 10. Sensitivity of max. torque versus thickness. on the design variables, we define the sensitivity indices as the
ratio of the variation of motor performance and the variation of
design variable. For example, the sensitivity index of maximum
indicates that the torque density is increased by diminishing the torque is denoted by , where symbolizes for design
fraction of slot opening, which is defined as the ratio of slot variables; the sensitivity indices of toque density , torque
opening and pitch of the stator. Fig. 9 shows that the air gap has ripple , efficiency , and maximum speed are expressed,
minor influence on the torque density. respectively, as , and .
Additional sensitivity curves of the torque density were linear Fig. 13 indicates that the motor torque varies significantly
or approximately linear with respect to design variables, such with the air-gap length, number of winding layers, number of
as back iron thickness, shoe depth fraction, tooth fraction, and turns per tooth, and wire diameter. However, it varies relatively
tooth thickness. This is because they are irrelevant to the max- less with the inner radius, outer radius, magnet fraction, and
imum torque, and their variations only affect the motor weight, thickness as well as other variables in Table III where each vari-
as shown in Figs. 10 and 11, for the sensitivities of maximum able has a number used to locate on the -axis of the sensitivity
torque and torque density with respect to tooth thickness. It is index plot. Fig. 14 shows that the torque density is heavily
obvious that the number of winding layers and number of turns affected by magnet fraction, magnet thickness, stator tooth
56 IEEE TRANSACTIONS ON MAGNETICS, VOL. 43, NO. 1, JANUARY 2007

Fig. 14. Sensitivity index of torque density. Fig. 17. Sensitivity index of maximum speed.

3) Stator tooth fraction has less influence on the maximum

torque, and is of little effect on efficiency, ripple, and speed.
However, the tooth fraction is an indication of the crossover
area that allows stacking windings in terms of the number
of turns and the number of layers; those determine the max-
imum allowable magnetomotive force to produce torque. It
is therefore selected as a design variable.
4) The tooth thickness, slot opening fraction, shoe depth frac-
tion, and back iron thickness are not influential for torque,
Fig. 15. Sensitivity index of efficiency. torque density, efficiency, and speed, and are excluded
from the design variables.
5) The air-gap length is a compromising factor between
torque and speed. It is usually made as small as possible
to obtain a maximum torque under the constraint of speed
limit. This design variable is set at 1 mm, which is reason-
able for manufacturing as the tradeoff between precision
and cost is considered.
6) The number of winding layers, number of turns per tooth,
and copper wire diameter determine the value of magne-
tomotive force, which produces torque but restricts speed.
Fig. 16. Sensitivity index of torque ripple. They are selected as important design variables to compro-
mise between the maximum torque and speed.
Eventually, eight variables are chosen (denoted by stars in
fraction, air-gap length, number of winding layers, number of Table III) for the following optimal motor design. It is also worth
turns per tooth, and wire diameter. In Fig. 15, the outer radius noting that the torque ripple was included in the sensitivity anal-
and copper wire diameter have indispensable influence on ysis but is excluded from the optimal design. The fact is that
efficiency, where larger outer radius produces more torque and the torque ripple is very sensitive only to the magnet fraction
larger wire diameter results in less copper loss. Other variables, and slot opening fraction of the wheel motor. In the optimiza-
such as inner radius, magnet fraction and thickness, air-gap tion process, the 2-D model used in the optimal design does
length, number of winding layers, and turns per tooth, also not provide accurate values of torque ripple. Therefore, it has
provide more influence on efficiency than the rest of the design not chosen an objective function in the following optimization
variables. Fig. 16 shows that the variations of magnet and slot process, but remained for further design refinement by the fi-
opening fractions are very crucial to the torque ripple. The nite-element method in ways of adjusting its corresponding sen-
variation of the air-gap length is an apparent index of motor sitive variables.
speed as shown in Fig. 17; the air-gap length is sometimes made
adjustable to tune the motor speed under various operations [8]. B. Design Optimization
From the above sensitivity analyses, we find the following The compromise programming method in the multifunctional
facts. optimization system tool (MOST) [12] is applied to search the
1) The inner and outer radii determine the size of the air-gap optimal values of design variables that maximize the following
surface between the stator and rotor of the wheel motor, performance indices:
and therefore, very sensitive to torque and torque density.
Both are important design variables. Motor torque: (10)
2) The magnet fraction and thickness have much influence on
Motor weight: (11)
all motor performances, especially the maximum torque,
torque ripple, and speed of motor; they are selected as de- Motor efficiency: % (12)
sign variables. Motor speed: (13)
YANG AND CHUANG: OPTIMAL DESIGN AND CONTROL OF A WHEEL MOTOR FOR ELECTRIC PASSENGER CARS 57

TABLE V
MOTOR CONSTRAINTS IN OPTIMIZATION

TABLE VI
OPTIMIZED MOTOR PARAMETERS AND PERFORMANCE

Fig. 18. 3-D model of magnetic flux density distribution.

due to linear assumptions and simplified 2-D motor configura-

tion. Therefore, further investigation on the motor performance
and refinement of geometries must be made by finite-element
analysis for the improvement of dynamics and the reduction of
weight of the dedicated wheel motor, thus providing designers
advanced information for making a final decision.
From Table VI, the maximum torque of the weighting ratio
The rated torque is an implicit function of design vari- 1:1:1:1 is the best among the three. The motor weight obtained
ables, and is calculated from the magnetic circuit model by from the case of the weighing ratio 1:6:1:1 has the least value
equally spaced points of rotor shift over an electric period. Like- due to the heaviest weighting on , representing the impor-
wise, the weight of the motor , its rated speed , the core tance of reducing weight. The best motor speed is also obtained
loss , and the stray loss composed of windage, friction, from the 1:6:1:1 weighting. However, the best torque density
noise, and other less dominant loss components are all func- happens for the 1:1:1:1 weighting. Both cases deserve further
tions of design variables. The optimizer weighs these perfor- investigation by the following finite-element analysis, which re-
mance indices to reach a satisfactory compromise among the de- fines the motor shape to reduce its weight and torque ripple.
sign variables under the prescribed constraints, as summarized The magnetic analyzer Maxwell 3D of ANSOFT1 is used to
in Table V. verify the motor prototype performance and to refine the motor
The optimizer MOST can deal with real, integer, and dis- geometries by numerical calculation on the 3-D motor config-
crete design variables simultaneously. In this design, the per- uration. A quarter section of the motor comprising a half elec-
formance indices, design functions, and prescribed constraints tric period is sufficient to model and simulate the motor per-
are expressed in terms of design variables, in which the number formance. The finite-element mesh is automatically generated
of winding layers and the number of turns per layer are inte- for the calculation of magnetic flux, flux density, and torque
gers; the wire diameter provided by the manufacturers is dis- distributions. The boundary of the finite-element model is sur-
crete, while others are real. The computation flow of the gra- rounded by air of enough thickness that its magnetic perme-
dient-based optimization algorithm in MOST is composed of: ability is much smaller than other magnetic materials in the
1) initial guess of design variables; 2) calculation of gradients of magnetic loop. Fig. 18 shows the magnetic flux density distri-
objective and constraint functions; 3) determination of the max- bution over a quarter section of the motor, where the flux flows
imum descent direction and the next set of design parameters; through the air gap, stator teeth, and back iron through a closed
and 4) convergence test until reaching the final solution. Dif- magnetic loop. On the cylinder of radius ,
ferent weightings were assigned to the four performance indices the flux density distribution curves under the maximum torque
for which relative importance is addressed for the optimization, output are illustrated in Fig. 19.
but listed in Table VI are the three best results in terms of motor A. Reduction of Motor Weight
performances.
According to the sensitivity analysis, the torque and speed of
V. FINITE-ELEMENT ANALYSIS the wheel motor are insensitive to the back iron thickness and
tooth thickness. However, the adjustment of these variables re-
The above results are obtained by the optimization scheme
duces the weight before the magnetic flux is saturated. Fig. 20
based on the 2-D magnetic model. This conventional magnetic
model does not usually produce precise results of optimization 1ANSOFT is a registered trademark of Ansoft Corporation.
58 IEEE TRANSACTIONS ON MAGNETICS, VOL. 43, NO. 1, JANUARY 2007

TABLE VII
PERFORMANCE BEFORE AND AFTER FE REFINEMENT

Fig. 19. Magnetic flux density distribution on R.

Fig. 22. Torque versus shoe depth fraction.

Fig. 20. Torque versus back iron thickness curve for weighting 1:1:1:1 ( ) and
1:6:1:1 ().

B. Reduction of Torque Ripple

The torque ripple is caused by uneven distribution of flux den-
sity in the air gap between rotor and stators, where the air-gap
length and reluctance are functions of rotor shift and contain
many high harmonics. From the motor design point of view, the
torque ripple can be reduced by properly shaping the geometry
of the motor so that the variation of air-gap reluctance becomes
smooth by evaluating the produced torque ripple. From the sen-
sitivity analysis, the torque ripple is sensitive to the thickness
and width of the magnet, air-gap length, stator shoe depth, and
Fig. 21. Torque versus tooth thickness curve for weighting 1:1:1:1 ( ) and slot opening. Either increasing the thickness and width of the
1:6:1:1 (). magnet or the reduction of air-gap length will produce more
magnetic energy in the air gap under the stator teeth and mag-
nets, thereby intensifying the variation of the flux density distri-
presents the change of torque with respect to back iron thick- bution. Since the optimal design has satisfied the torque spec-
ness. For the weighting 1:1:1:1 on the objective functions, the ification, the values of thickness and width of the magnet and
optimal design of back iron thickness of 10 mm is over designed, air-gap length are preserved in the design.
therefore its reduction to 8 mm is good enough to satisfy the The shoe depth fraction was not a design variable but prede-
torque specification. The tooth thickness is also irrelevant to the termined at 0.11 mm. From Figs. 22 and 23, the average torque
torque of the motor if it provides sufficient area for the mag- excited by a square wave is not much affected by the shoe depth
netic flux without saturation under an allowable maximum cur- fraction, but the torque ripple is. Furthermore, the torque ripple
rent excitation. The space for stack coil windings, however, is is sensitive to the slot opening fraction as shown in Fig. 24. The
limited by the tooth thickness. By fixing the number of turns optimal design determines the slot opening fraction at 0.28 and
of the winding, the stacking factor, defined as the ratio of the the corresponding slot opening is about 9 mm; the size of the
area occupied by coils and the slot area, varies with the tooth slot is therefore large enough for winding coils. Manufacturers
thickness. Fig. 21 reveals that for the weighting of 1:1:1:1, the suggest a permissible slot opening value of 5 mm that the corre-
tooth thickness can be reduced from the optimal design of 40 to sponding slot opening fraction can be chosen at 0.14 according
36 mm without diminishing torque, but concurrently preserving to Fig. 24 for less torque ripples.
a permissible stacking factor and increasing the torque density
by reducing the weight. Similar evaluation refinement for the C. Thermal Analysis and Decision Making
case of weighting 1:6:1:1 is also performed as summarized in After the finite-element refinement, the case of the weighting
Table VII. ratio 1:1:1:1 prevails in the performance of motor torque, torque
YANG AND CHUANG: OPTIMAL DESIGN AND CONTROL OF A WHEEL MOTOR FOR ELECTRIC PASSENGER CARS 59

Fig. 23. Torque ripple versus shoe depth fraction.

Fig. 25. Heat dissipation fins on motor cover.

Fig. 24. Torque ripple versus slot opening fraction.

Fig. 26. Steady-state temperature distribution of wheel motor at rated cruising

speed of 40 km/h.
density, and torque ripple over the weighting 1:6:1:1, although
the latter comes up with less weight. The comparison of effi-
ciency is not presented here because it was similar among the coolant or fans, but such an improvement will consume extra
three cases in Table VI, and it depends more on the factorization power on the vehicle.
and motor drive design. The case of the weighting ratio 1:1:1:1
is then chosen as the final design. VI. OPTIMAL CONTROL WAVEFORM
Finally, the thermal analysis has been carried out by de- In the process of optimal design, the motor was excited by a
signing heat dissipation fins distributed as concentric circles square wave for simplicity. For better driving torque and effi-
over both plates of aluminum covers, as shown in Fig. 25. The ciency, an optimal control current pattern for the wheel motor
height, width, and the gap of the fin are 12, 3, and 12 mm, was proved to be the same as the flux variation in the air gap be-
respectively. By the simulation of a passenger car moving at a tween the stator and rotor [13], [14], where the optimal driving
cruising speed of 40 km/h, the convection coefficient is chosen waveform was obtained by maximizing the average torque of
at 25 W/m K and the ambient temperature is 20 C. The the motor under a constraint on the average ohmic loss. In other
major heat source coming from the stator copper windings is words, the stator current to produce a maximum torque must
estimated to be 640 W for the rated and cruising speed, and have a phase lag of 90 electrical degrees from the permanent-
results in a steady-state temperature distribution as shown in magnetic flux of the rotor. Since the back EMF occurs as the
Fig. 26. The maximum temperature is 88 C around the stator flux linkage through the coil changes, the flux variation is pro-
teeth made of alloy steel and the temperature on the cover portional to the back EMF and the optimal current waveform
is about 82 C. The heat dissipation on the normal driving turns out to be proportional to the back EMF.
condition is satisfactory. Without the heat dissipation fins, It is well known that the torque constant of a dc motor is de-
the maximum temperature on the stator teeth increases up to fined as the ratio of the produced torque and the corresponding
130 C at the steady state, which may cause demagnetization of phase current, and is theoretically the same as the electric con-
magnets and deteriorate motor performance. In a severe driving stant defined as the proportional coefficient of back EMF and
condition when the car is accelerating on the 15 slope and rotational speed of the motor. The back EMF waveform can thus
a 2560 W power is to be dissipated from the stator windings be obtained by calculating the torque constant of the motor via
through fins, the maximum temperature goes up to 100 C the finite-element tool ANSOFT with 3-D motor model. Fig. 27
after 11 min; the motor will be soon heated up over 130 C illustrates the back EMF of phase A and its corresponding spec-
after 11 min without the dissipation fins. The results suggest trum, where the back EMF wave presents a bunch of high fre-
that severe driving must be restrained or protection circuits of quency harmonics causing undesirable torque ripples. Since the
the motor drive have to be properly designed for better motor amplitude of the fundamental harmonic dominates, it is then
performance. Further improvement on the heat dissipation can extracted as the proposed driving current waveform with less
be achieved by introducing forced cooling systems by liquid torque ripple.
60 IEEE TRANSACTIONS ON MAGNETICS, VOL. 43, NO. 1, JANUARY 2007

Fig. 27. Back EMF of phase A and its spectrum. Fig. 28. Torque distributions for various current waveforms.

TABLE VIII
MOTOR PERFORMANCE WITH VARIOUS CURRENT PATTERNS motor under prescribed constraints. A systematic procedure
from the magnetic circuit analyses to the finite-element mod-
ification and verification constitutes a complete design of the
wheel motor. The back electromotive force of the motor pro-
totype provides information about the optimal driving current
pattern, and its fundamental harmonic is verified to yield a least
torque ripple. The resulting prototype has a torque density of
1.72 kg m/kg that provides a maximum torque up to 38.65 kg m
torque for an optimal current input at its maximum allowable
phase current of 50.25 A (rms) with a least ratio of torque
Table VIII illustrates the maximum torque and torque ripple ripple; this enables a 600 kg passenger car to accelerate from
for different kinds of driving current patterns at the maximum 0 to 40 km/h in 5 s up a 15 degree incline. The thermal anal-
allowable phase current of 50.25 A rms. The original back EMF ysis provides satisfactory information on the heat dissipation
pattern produces the largest torque, but its high order harmonics through fins over the cover. After the prototype wheel motor
yields the worst torque ripple up to 31.1%. The fundamental is fabricated, its performance will be tested on a dynamometer
harmonic of the back EMF, which has a phase lag of less than and an optimal driving strategy will be determined by the
1 degree from a pure sine wave, produces a little less torque but design of power drive and power management system.
the resulting torque ripple is the least. The current input of a
pure sine waveform also produces small torque ripples as the APPENDIX A
fundamental harmonic of the back EMF does, but the latter pro- The air-gap length , magnetomotive force ,
vides 1.69 kg m more torque than the pure sine current input. coenergy , and torque are formulated explicitly as
Fig. 28 shows the torque distributions produced by the input functions of motor geometric and electric design variables.
current waveforms of square, pure sine, original back EMF and
the fundamental harmonic of the back EMF. The current exci- A. Air-Gap Length
tation of the pattern of the fundamental harmonic of back EMF
increases the torque density up to 1.72 kg m/kg due to the in-
crease of the average torque, and therefore is chosen as the op- (A-1)
timal control waveform for the dedicated wheel motor.
The effective air-gap length on the stator side is defined as [15]
VII. SUMMARY AND CONCLUSION
This paper presents a multifunctional optimization design (A-2)
of an axial-flux permanent-magnet brushless dc wheel motor (A-3)
for a small electric passenger car. The sensitivity analysis with (A-4)
the magnetic circuit model provides an effective way to select
(A-5)
the design parameters, which are iteratively tuned through the
multiobjective optimal design process to maximize the output (A-6)
torque, speed, efficiency, and torque density of the dedicated (A-7)
YANG AND CHUANG: OPTIMAL DESIGN AND CONTROL OF A WHEEL MOTOR FOR ELECTRIC PASSENGER CARS 61

in which is the minimum air-gap length between the stator and [6] J. F. Eastham, F. Profumo, A. Tenconi, R. Hill-Cottingham, P. Coles,
the rotor, and is the slot width of the stator. and G. Gianolio, “Novel axial flux machine for aircraft drive: Design
and modeling,” IEEE Trans. Magn., vol. 38, no. 5, pp. 3003–3005, Sep.
Referring to Figs. 1 and 2, the air-gap length on the rotor 2002.
side is a function of magnet fraction ; it is over the [7] T. Uematsu and R. S. Wallace, “Design of a 100 KW switched reluc-
nonmagnet material part, and is approximated by over tance motor for electric vehicle propulsion,” in Applied Power Elec-
tronics Conf. and Expo., 1995, Conference Proceedings, Mar. 5–9,
the width of the magnet, where its recoil permeability is . 1995, vol. 1, pp. 411–415.
Apparently, the distribution of the air-gap length is an explicit [8] S. C. Oh and A. Emadi, “Test and simulation of axial flux-motor char-
function of geometric design variables. acteristics for hybrid electric vehicles,” IEEE Trans. Veh. Technol., vol.
53, no. 3, pp. 912–919, May 2004.
[9] F. Chia, S. Cui, and S. Cheng, “Performance analysis of double-stator
B. Coenergy and Torque starter generator for the hybrid electric vehicle,” IEEE Trans. Magn.,
vol. 41, no. 1, pp. 484–487, Jan. 2005.
In the optimization program, both the coenergy and torque [10] P. R. Upadhyay and K. R. Rajagopal, “FE analysis and computer-
aided design of a sandwiched axial-flux permanent magnet brushless
are coded in discrete forms: dc motor,” IEEE Trans. Magn., vol. 42, no. 10, pp. 3401–3403, Oct.
2006.
[11] G. G. Lucas, Road Vehicle Performance: Methods of Measurement and
Calculation. New York: Gordon and Breach, 1996.
[12] C.T. Tseng, W. C. Liao, and T. C. Tang, MOST User’s Manual Na-
tional Chiao-Tung University, Taiwan, R.O.C., Version 1.2, Tech. Rep.
(A-8) AODL-93-0, 1993.
[13] Y. P. Yang, Y. P. Luh, and C. H. Cheung, “Design and control of
axial-flux brushless dc wheel motors for electric vehicles—Part I:
multi-objective optimal design and analysis,” IEEE Trans. Magn., vol.
40, no. 4, pp. 1873–1882, Jul. 2004.
(A-9) [14] Y. P. Yang, J. P. Wang, S. W. Wu, and Y. P. Luh, “Design and control
of axial-flux brushless dc wheel motors for electric vehicles—Part II:
in which an electric period is divided by equally spaced optimal current waveforms and performance test,” IEEE Trans. Magn.,
vol. 40, no. 4, pp. 1883–1891, Jul. 2004.
points, each with mechanical angle apart while [15] D. C. Hanselman, Brushless Permanent-Magnet Motor Design. New
and . York: McGraw-Hill, 1994.

ACKNOWLEDGMENT Manuscript received May 27, 2005; revised September 25, 2006. Corre-
sponding author: Y.-P. Yang (e-mail: ypyang@ntu.edu.tw).
This work was supported by the National Science Council of
Taiwan, R.O.C., under Contract NSC92-2218- E-002-020, and
in part by NSC90-2218-E-002-053. Yee-Pien Yang (M’91) was born in Taiwan, R.O.C., in 1957. He received the
B.S. and M.S. degrees in mechanical engineering from the National Cheng-
Kung University, Taiwan, R.O.C., in 1979 and 1981, respectively, and the Ph.D.
degree in mechanical, aerospace, and nuclear engineering from the University
REFERENCES of California, Los Angeles, in 1988.
[1] L. Chang, “Comparison of AC drives for electric vehicles—A report on He was an Associate Professor in the Department of Mechanical Engineering
experts’ opinion survey,” IEEE Aerosp. Electron. Syst. Mag., pp. 7–10, of National Taiwan University from 1988 through 1996, and was promoted to the
Aug. 1994. rank of Full Professor and led the Electromechanical System Research Group
[2] C. C. Chan and K. T. Chau, “An advanced permanent magnet motor since 1996. He was a visiting scholar at the University of California, Los An-
drive system for battery-powered electric vehicles,” IEEE Trans. Veh. geles, from 2004 to 2005, conducting research on the design and control of
Technol., vol. 45, no. 1, pp. 180–188, Feb. 1996. electromechanical systems, biomedical engineering, and assistive tool design
[3] G. H. Chen and K. J. Tseng, “Design of a permanent-magnet direct- for the disabled.
driven wheel motor drive for electric vehicle,” in 27th Annu. IEEE
Power Electronics Specialists Conf., Jun. 1996, vol. 2, pp. 1933–1939.
[4] A. H. Wijenayake, J. M. Bailey, and P.J. McCleer, “Design optimiza-
tion of an axial gap permanent magnet brushless dc motor for electric Down-Su Chuang was born in Taiwan, R.O.C., in 1979. He received the B.S.
vehicle application,” in 13th IAS Annu. Meeting, Industry Applications degree in mechanical engineering from the National Cheng Kung University and
Conf. Proc., Oct. 1995, vol. 1, pp. 685–692. the M.S. degree in mechanical engineering from the National Taiwan University,
[5] D. Patterson and R. Speée, “The design and development of an Taiwan, in 2001 and 2004, respectively.
axial flux permanent magnet brushless dc motor for wheel drive in He is currently a System Engineer at the Asia Pacific Fuel Cell Technology
a solar powered vehicle,” IEEE Trans. Ind. Appl., vol. 31, no. 5, pp. Corporation, responsible for fuel cell system integration. His expertise is on the
1054–1061, Sep.–Oct. 1995. design and control of electromechanical systems, and fuel cell system controls.