Optimal Design of An In-Wheel BLDC Motor For A Kick Scooter: October 2010
Optimal Design of An In-Wheel BLDC Motor For A Kick Scooter: October 2010
Optimal Design of An In-Wheel BLDC Motor For A Kick Scooter: October 2010
net/publication/224188231
CITATIONS READS
6 2,443
4 authors, including:
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Yves Perriard on 03 December 2014.
Abstract—The paper presents an optimization design of an and on a horizontal rough surface (asphalt), the measured value
in-wheel BLDC motor for a kick scooter. The optimization is of resistive force (opposing to the movement) is 12.9 N and
performed using a genetic optimization tool combined with a 17.2 N, respectively. It is shown that those two values can be
FEM commercial software. The new contributions of the paper
are: i) introduction of three operating modes for which the motor approximately considered independent of the speed (for the
is optimized with a reduced number of FEM simulations and speed range between 3 and 5 m/s).
ii) new approach to simultaneously maximize the energetic effi-
ciency, minimize the cogging and respect the thermal constraint. A. Operating mode 0
The operating mode 0 is defined as the movement on a
Index Terms—BLDC motor, genetic algorithm, optimization.
horizontal smooth surface. The resistive force in this mode
I. I NTRODUCTION (12.9 N) is sufficiently low so that the user can be left to ride
without the motor assistance.
The electrically assisted vehicles, from the bicycle to the
The measurements show that a typical user generates the
truck, are gaining more and more popularity [1]-[5]. In many
force of 136 N during 0.5 s of acceleration phase. The speed
cases, the electric motor is in-wheel (direct drive) which
increases from the initial 4.17 m/s (15 km/h, 69.4 rad/s) to 4.87
greatly reduces the space occupied by the drive and eliminates
m/s (17.5 km/h, 81.0 rad/s). During the deceleration phase
the need for a mechanical transmission [6],[7]. The problem
which lasts for 4.76 s, the speed drops back to 4.17 m/s,
is that the rotating speed is relatively low, which requires a
completing one period of the movement (Fig. 2 a) and b)).
high torque and current to supply the necessary power.
The ratio between the acceleration time and the period time
This paper presents the optimal design of the motor for
is D=0.095.
an electrically assisted kick scooter. The laboratory [8] has
already built a prototype of the scooter but with a motor off-
the-shelf and mechanical transmission (Fig. 1). The idea now
to design an in-wheel motor for this application.
Battery
Motor
Transmission
293
known in advance. All this means that the motor has to be A set of individuals (forming a generation) is created. Each
simulated for several rotor positions. individual is characterized by a set of parameters (called
In order to have a reasonable simulation time, 5 points over genes). For each individual, the fitness (objective) function is
one electrical quarter-period are simulated. The mean value of calculated. Then, the individuals with the best fitness crossover
the obtained set of torque values is the motor (useful) torque their genes to generate a new generation of individuals, and a
M ; the difference between the maximal and minimal values small part of the genes of the new individuals is modified
is the torque ripple. This ripple divided by 2 is the cogging randomly by mutation. The condition of the end is that a
torque Mc . limited number of generations is achieved.
D. Thermal model B. Application to this case
The most of the motor losses are generated in the stator In the analyzed case, it is chosen that each generation
windings and iron. The generated power is evacuated by the consists of 10 individuals. Each individual is a motor design,
(forced) convection over the kick scooter surface. characterized by the values of 6 free input motor parameters
It is important to point out that the mode 3 is critical for (stator yoke radius r2 , air gap e, rotor yoke thickness d,
the temperature, as the corresponding current is maximal. In magnet thickness md , magnet width mw and tooth width tw ,
addition, note that the copper losses are generated only during as shown in Fig. 5). Additional conditions should be checked
the deceleration phase whereas the iron losses are generated before starting simulations, in order to be sure that for example
during the whole period. It means that the losses generating adjacent magnets do not overlap.
the heat are given by Pf e3 + (1 − D)Pcu3 .
As the stator is interior, the heat transfer is significantly md
reduced by the presence of the rotor: the heat flows from mw
the stator to the rotor either through the air or by conduction
through the bearings, and both flows are weak. The conclusion tw
is that the heat is evacuated from the stator only by the motor
r2 d
shaft and scooter housing (Fig. 4).
r1 r6
294
the efficiency using (1). The iron losses are calculated using cM =1. The coefficient cT should be chosen sufficiently high:
(2) as the FEMM returns the value of B. a suitable value is cT =100.
The cost functions are shown in Fig. 6. The dots present
D. Fitness function
the ideal cases.
The main problem with the genetic algorithm is how to
define the fitness function. A new approach to define it is 2 1.4
used in this paper: the fitness function F (which should be 1.3
minimized in this case) is given in the form of a product of
kM (−)
kη (−)
three cost functions in the form 1.5 1.2
1.1
F = kη kM kT (3)
1 1
The efficiency cost function is given by 0.9 0.95 1 0 0.05 0.1
η (−) Mc (Nm)
kη = 1 + cη (1 − η) (4) 60
kT (−)
gives kη = 1. 20
The torque cost function is given by
Mc 0
10 15 20
k M = 1 + cM (5) T (C)
M1
It is used to impose that the ratio between the cogging torque Fig. 6: The cost functions kη , kM and kT (• ideal cases, N
and useful torque should be minimized; in the ideal case final values).
(which can never be achieved) the cogging is zero, which
gives kM = 1. (The useful torque in the mode 1 is taken
into account as it is lower than in the modes 2 and 3.) V. F INAL DESIGN
Finally, the temperature cost function is given by
( q The optimization procedure lasts for 3 hours on a PC with
cT T −T c
if T > Tc Intel Xeon processor 2.8 GHz. In spite of a partly random
kT = 1 + Tc (6) character of the algorithm, it always converges to one solution.
0 otherwise
The configuration 14/12 is better than 12/9, and is chosen for
It is used to impose the constraint that the overtemperature the final design. The final values of the parameters are: r2 =21.1
T should not exceed its maximally allowed value Tc . If this mm, e=0.54 mm, d=6.3 mm, md =2.9 mm, mw =11.7 mm and
condition is satisfied (which should always be the case), kT = tw =5.6 mm. According to the available DC voltage, 192 turns
1. per phase are wound. The phases are connected in triangle.
The whole system in exploded view is shown in Fig. 7. The
E. Cost functions
energetic efficiency is η=92.7%. The cogging torque is 0.012
Although the three cost functions are simply multiplied to Nm.
make the fitness function, they are obviously different. The
cost functions kη and kM are related to the optimization; they Wheel Rotor
can never take the value of 1, but the goal is to have them as
close to 1 as possible. Cover
The cost function kT is related to the constraint; its value
has to be 1. To do this, its first (right) derivative near the
point T = Tc should be as high as possible to heavily force
T to be less than Tc . The simplest way to do this is to use a
linear function of type c0T (T − Tc ) with a high value of c0T , Stator
but the problem with this function is that as T increases kT Shaft
would increase too much, so that we would lose insight into
the optimization process. Instead, the square root function is Fig. 7: Integration of the motor in the kick scooter wheel
suitable as the derivative at the point T = Tc is infinite, and (exploded view).
as T increases the derivative decreases, keeping numerically
acceptable values of kT . Concerning the fitness function, the final values of cost
The choice of numerical values for the coefficients cη and functions are kη =1.73, kM =1.05 and kT =1 (the corresponding
cM allows to choose importance of the efficiency and of the points are shown by triangles in Fig. 6), giving the fitness
ratio between the cogging and useful torques, respectively. function F =1.81. It is interesting to point out that the air gap
After several trials, suitable values are obtained: cη =10 and e does not necessarily take its minimal value, as the reduction
295
of e means that the field and torque are increased but at the [6] C. Liu, K. Chau, ”A permanent-magnet hybrid in-wheel motor drive for
same time the cogging is increased as well. electric vehicles”, IEEE Vehicle Conference 2008, pp. 1–6
[7] C. Versele et al., ”Analytical design of an axial flux permanent magnet in-
As an illustration, in the mode 3 the necessary current wheel synchronous motor for electric vehicle”, 13th European Conference
density to generate the torque of 1.55 Nm is 5.67 A/mm2 , the on Power Electronics and Applications 2009, pp. 1–9
copper and iron losses are 10.65 W and 0.81 W, respectively. [8] A. Hodder, P. Jaquier, Y. Perriard, ”A new electrically assist scooter”, Int.
Conf. on Electrical Machines ICEM 2008, pp. 1–6
The overtemperature of scooter housing surface is 14◦ C. [9] Thunder Power website [Online]. Available: http://www.thunderpowerrc.
The motor prototype is fabricated according to the obtained com
final design. The photos of the stator and rotor are shown in [10] Finite Element Method Magnetics website [Online]. Available:
http://www.femm.info
Fig. 8.
296