1/2, 2007 47
Noel León-Rovira*
Center for Innovation in Design and Technology,
Tecnológico de Monterrey,
Campus Monterrey. Ave. Eugenio Garza Sada # 2501,
Colonia Tecnológico, CP 64841 Monterrey, Mexico
E-mail: noel.leon@itesm.mx
*Corresponding author
Eduardo Uresti
Center for Artifitial Intelligence,
Tecnológico de Monterrey,
Campus Monterrey. Ave. Eugenio Garza Sada #2501,
Sucursal de Correos ‘J’, Colonia Tecnológico,
CP 64841 Monterrey, Mexico
E-mail: euresti@itesm.mx
Waldo Arcos
Center for Innovation in Design and Technology,
Tecnológico de Monterrey,
Campus Monterrey. Ave. Eugenio Garza Sada # 2501,
Colonia Tecnológico, CP 64841 Monterrey, Mexico
E-mail: waldo.arcos@gmail.com
Abstract: The electric motor efficiency represents the effectiveness with which the motor
converts electrical energy into mechanical energy. As the energy losses are converted into heat,
which is dissipated by the motor frame aided by internal and external fans, a better cooling
system adds up to better efficiency. In recent years, improvements in motor efficiency have been
achieved but at higher costs. By using Genetic Algorithms (GAs), changes are introduced to the
fan shape looking for a better aerodynamic performance. The evaluation of the achieved fan
efficiency with the modified shapes is performed with Computational Fluid Dynamics (CFD)
simulation software.
Reference to this paper should be made as follows: León-Rovira, N., Uresti, E. and Arcos, W.
(2007) ‘Fan shape optimisation using CFD and genetic algorithms for increasing the efficiency of
electric motors’, Int. J. Computer Applications in Technology, Vol. 30, Nos. 1/2, pp.47–58.
Biographical notes: Noel León-Rovira is a Professor at the Center for Innovation in Design
and Technology at Mexico’s Tecnológico de Monterrey. He holds a Degree in Mechanical
Engineering, as well as a PhD in Mechanical Engineering (Summa cum Laude), both from the
Dresden University of Technology, Germany. He also made Postdoctoral studies on design
methodology and computer-aided design. He is the Director of the Research Program
Creativity, Inventiveness and Innovation in Engineering. He specialises in product design, design
methodology and computer-aided engineering.
Eduardo Uresti is a Professor at the Center for Intelligent Systems at Mexico’s Tecnológico de
Monterrey. He holds a degree in Mathematics, a master degree in Mathematics and a PhD in
Computer Science. He has served as professor in the Physics and Mathematics School (ESFM) of
the National Politechnical Institute (IPN) and also in the Physics and Mathematics Faculty at
UANL Mexico. His current interests include the use of genetic algorithms for multi-modal and
multi-objective optimisation (new algorithms and their applications) and the application of
Artificial Intelligent strategies in teaching and learning.
Waldo Arcos received the BS Degree in Mechanical with Minor in Electrical Engineering from
the Autonomous University of Nuevo León, Nuevo Leon, Mexico, and the Master’s Degree in
Manufacturing Systems from Tecnológico de Monterrey, Nuevo Leon, Mexico, in 2002 and
2006, respectively. He is currently responsible for the mechanical and electrical design of electric
motors in an electric motors company.
In the second case, the airflow was increased in the The mesh was generated inside of the 3D CAD system
analytical model by 50%, without considering any increase using an interface provided by the CFD software. The grid
of the energy consumption of the fan. The efficiency is set up with periodic boundaries on either side of the
increase calculated was 1.35%. It became evident that the domain. The upstream boundary is defined by velocity and
efficiency of the motor could be more benefited through the downstream boundary by pressure. The single blade
efficiency augmentation of the fan. centred is assumed to be rotating, while the two even
adjacent passages are stationary. The z-axis is used
as a rotation axis for the reference frame. The fan spins
7 Simulation using CFD simulation software clock-wise (looking from the positive end of the z-axis) at
constant rotational speed. The fan rotation speed is defined
integrated into 3D CAD system with meshing
by the motor at full load. The standard k –ε turbulence
capabilities
model (Abbott, 1989) and sea-level conditions were used.
Two commercial CFD simulation programs were compared. The 3D CAD model and the mesh generated are shown in
In the first case, a 3D CAD system for modelling Figure 1.
the geometry and CFD software with dedicated mesh The simulations were performed based on the
generation software were used. The analysis time was defined boundary conditions. The simulation stops after
approximately 4 h. Then the same case study was performed convergence criteria are met. The convergence criteria are
using CFD simulation software integrated into a 3D CAD shown in the Figure 2. Each curve should become horizontal
system with meshing capabilities. The analysis time was as criterion that the convergence was reached. A guideline
about 40 min. The second package was selected, as it followed is that the quantities in the calculation progresses
performed the analysis up to ten times faster than the first change <5% over the last 20% of the total iterations.
with similar results. Figure 3 shows the velocity vectors as found at the
Taking advantage of the periodicity of the geometry, midspan of the rotational area. The velocity vector plot
only one-eighth of the fan geometry was modelled. provides insight into the behaviour of the flow around the
This way the total computational time is reduced fan blade. This analysis is the basis for finding the
accordingly. variations of blade profiles that increase the airflow.
The geometry was constructed as a single blade centred The pressure distribution on the blade is shown in
within the region divided by two even adjacent passages. Figure 4.
Figure 1 (a) 3D CAD geometry and (b) mesh model in CFD program
(a)
(b)
Fan shape optimisation using CFD and genetic algorithms 51
The contour plots illustrate the pressure rise on the be modified during the optimisation run. For a first trial, the
fluid caused by the fan blade. Positive pressure values straight profile of the fan blade was substituted by a curved
occur on the surface, where the fan blade pushes the profile as shown in Figure 5.
air and negative pressures (suction) on the other This analysis shows an increment in the velocity
surface. vector’s magnitude. The results are shown in Figure 6.
The velocity vector’s magnitude was increased 4.7%
7.1 Profile modification with respect to the current design. The increment of the
parameters measured shows that it is possible to get higher
As the purpose of this paper is to look for new designs that airflow provided by the fan over the motor. By increasing
increase the efficiency of the electric motor using shape the airflow the electric motor efficiency can be improved,
optimisation through GAs, the blade profiles are required to as was discussed in Section 5.
Figure 6 (a) Velocity vectors (profile curve-simulation) and (b) pressure distribution on the blade (profile curve simulation)
(a)
Fan shape optimisation using CFD and genetic algorithms 53
Figure 6 (a) Velocity vectors (profile curve-simulation) and (b) pressure distribution on the blade (profile curve simulation) (continued)
(b)
8 Representation of fan blade for optimisation evaluated and represents a possible solution. Following
a strategy adopted by several researches (Lamping, 2003;
Shape optimisation deals with variations of the form: find
Leon et al., 2004, 2005; Marco et al., 2004; Michalewicz,
a shape (in two or three dimensions) that is optimal in a
1992; Obayashi et al., 2000a) and the concepts studied in
certain sense, while satisfying certain requirements and
the preceding sections, the shape parameterisation is
constraints. In the present case study, as was already
based on B-spline curves. B-splines are commonly used
mentioned, advantage was taken of the periodicity
to define curved bodies and profiles. In some cases
of the geometry and only one fan blade was simulated.
(Marco et al., 2004), Bézier curves are used, and the
In order to simplify the shape optimisation problem,
only the profile half way up the fan blade will be modified, parameters constituting the chromosome are coded as binary
as shown in Figure 7. The fan blade profile analysed string. The smoothness properties of Bézier curve permits to
in Section 7 will be codified below as chromosome in avoid non-feasible shapes by the crossover operator.
the GA. In this case study, non-feasible shapes are controlled
by using constraints added in the GA, which are discussed
Figure 7 Fan blade profile briefly in the last paragraph of this section. Six control
points are used in this case to control the whole fan profile
shape modification by the crossover operator, as shown in
Figure 8.
The control points shown above (P1–P6) are defined done yet automatically (because actual CFD software lacks
by their x- and y-coordinates, respectively; thus, scripting capabilities), it was necessary to use an evaluation
P1 = (x1, y1), …, P6 = (x6, y6). For simplifying the shape team. The evaluation team was a group of specialists
parameterisation, the x values are fixed, and only the y- (volunteer Master degree students in the Computer-Aided
coordinates are allowed to vary. This way, P1 = y1, P2 = y2, Engineering course), where the simulation results of each
…, P6 = y6 represent the search space, while Pi, Pi′, Pf and population member was assigned for evaluation to one
Pf′ are fixed. Several approaches have been proposed person of this team.
for controlling the geometry through control points; for
example, in Obayashi et al. (2000a), the airfoil geometry is Figure 9 Process flow diagram from Genetic Algorithms
split into a mean camber line and thickness distribution.
Some points are used for the camber line and other points
are used for the thickness distributions. In this case, the
extreme points have to be fixed to avoid a new fan blade
profile being thinner than the current profile. Therefore, the
chromosome will be represented only by the y-coordinates:
Chromosome = y1 , y2 , y3 , y4 , y5 , y6 .
Upper
Lower
(a)
(b)
56 N. León-Rovira, E. Uresti and W. Arcos
Figure 10(b) shows some individuals of the GA population GA shows that introducing changes in the profile
in the web page. The best fan profile design generated it is possible to get a better aerodynamic performance than
by the GA is shown into the red circle. It will be discussed the results obtained by former profile curves. The results are
in the next section. The web page can be accessed in shown in Figure 11.
the following link http://lizt.mty.itesm.mx/~euresti/genetico/ In Figure 12, the changes in the new fan profile
home.htm. generated by GA are shown. This kind of variation obtained
by GA optimisation procedures is finer than the changes
obtained by trial and error variations of experienced
10 Results designers, who commonly do not have the patience and
The best fan profile generated by the GA presents a time for such minimal changes. The use of splines for
significant increase in the velocity vector’s magnitude of changing profiles that are initially modelled with only
9.4%, with respect to the current design, while the profile circular arcs and lines seems to be a better solution for
curve presented in Section 7.1 had an increment of 4.7%. shape optimisation by using genetics algorithms
Due to the time limits, the GA was not run until its final for controlling the geometry through control points
convergence; however, the best fan profile obtained from (Lamping, 2003).
Figure 11 (a) Velocity vectors (profile generated by GA) and (b) pressure distribution on the blade (profile generated by GA)
(a)
(b)
Fan shape optimisation using CFD and genetic algorithms 57
Olhofer, M., Jin, Y. and Sendhoff, B. (2001) Adaptive Encoding Wright, A.H. (1991) ‘Genetic algorithms for real parameter
for Aerodynamic Shape Optimization Using Evolution optimization’, in Rawlins, G.J.E. (Ed.): Foundations of
Strategies, Congress on Evolutionary Computation (CEC). Genetic Algorithms, Morgan Kaufmann, San Mateo CA,
IEEE Press, Seoul, Korea, pp.576–583. pp.205–218.
Whitley, D. (1989) ‘The genitor algorithm and selection pressure:
why ranked-based allocation of reproductive trail is best’, in
Schaffer, J.D. (Ed.): Proceedings of the Third International Websites
Conference of Genetic Algorithms, Morgan Kaufmann,
http://www.usmotors.com/.
San Mateo CA, pp.116–121.
http://www.reliance.com/mtr/b7087_5/b7087_5_2.htm.