Progress In Electromagnetics Research B, Vol.

31, 357–379, 2011

SMART ELECTROMAGNETIC SIMULATIONS: GUIDE-

LINES FOR DESIGN OF EXPERIMENTS TECHNIQUE

H. R. E. H. Bouchekara* , G. Dahman, and M. Nahas

Department of Electrical Engineering, Faculty of Engineering and
Islamic Architecture, Umm Al-Qura University, P. Box: 5555, Makkah
21955, Saudi Arabia

Abstract—Electromagnetic design problems usually involve a large

number of varying parameters. A designer can use different kinds
of models in order to achieve optimum design. Some models, e.g.,
finite-element model, can be very precise: however, it requires large
computational costs (i.e., CPU time). Therefore, the designer should
use a screening process to reduce the number of parameters in
order to reduce the required computational time. In this paper,
using the Design of Experiments (DOE) approach to reduce the
number of parameters is explored. The benefits of this technique are
tremendous. For example, once researchers realize how much insight
and information can be obtained in a relatively short amount of time
from a well-designed experiment, DOE would become a regular part of
the way they approach their simulation projects. The main objective
of this paper is to apply the DOE technique to electromagnetic
simulations of different systems and to explore its effectiveness on a new
field, namely the magnetic refrigeration systems. The methodology of
the DOE is presented to assess the effects of the different variables and
their interaction involved in electromagnetic simulations design and
optimization processes.

1. INTRODUCTION

Designing a system in engineering is a cooperative and iterative

process. It is usually a multidisciplinary task involving skilled engineers
specialized in different areas. The aim of the designing process is to
generate a system (a product) with a predefined set of requirements and
Received 21 May 2011, Accepted 27 June 2011, Scheduled 5 July 2011
* Corresponding author: Houssem Rafik El Hana Bouchekara
([email protected]).
358 Bouchekara, Dahman, and Nahas

constraints. Traditionally, this process has involved the development

of prototypes (physical models) to test. This is a time consuming and
expensive approach for developing new systems and can represent a
large percentage of the total system. Reducing time and costs has
always been, and still is, a key issue [1].
The first and most obvious place to reduce both costs and time
is by simulating the physical models rather than building actual
prototypes. Computer-aided analysis, also known as simulation, has
revolutionized engineering design. In the span of about 35 years,
simulation software has evolved into a variety of complex packages
equipped with sophisticated layout and visualization modules [2]. The
numerical engines employ computational methods whose efficiency and
versatility have improved greatly since their time of inception, mostly
in the 1970s. Such packages, which are now commercially available,
have allowed designers to leave behind the drawing board for the most
part and to shortcut through several stages of expensive prototyping
and manual cut-and-try tuning [2].
The process of building, verifying, and validating a simulation
model can be arduous, but once completed, it can be used to
explore different aspects of the modeled system. One extremely
effective way for accomplishing this is to use experimental designs, also
called Design of Experiments (DOE), to help explore the simulation
models [3]. According to [4], many simulation practitioners could
obtain more information from their analysis if they used statistical
theories, especially with the use of DOE developed specifically for
computer models. If the input variables to the process are varied,
the outputs will vary, even though the variation may only be due to
random effects or noise. The question is which input variables (factors)
are causing the majority of the variability in the output (responses)?
In other words, which factors are the significant “drivers”? It is
desirable to determine where the variability is coming from (also
known as “sensitivities”) with an optimum expenditure of resources [5].
Like most statistical methods, DOE has the primary objective of
obtaining maximum information at minimum cost: its goal is to
find the best ratio between benefits of information and information
costs [5]. More specifically, in the case of DOE, the desired information
is the quantification of the influence of several factors on a given
phenomenon. Due to this quantification, it is possible to predict the
behavior of the system studied in different possible configurations and,
consequently, optimize the operation of such systems. To achieve this,
the DOE methodology offers a testing strategy, where one of its main
characteristics leads to minimizing the number of tests to be performed.
Recently, the DOE technique has been adopted in the design
Progress In Electromagnetics Research B, Vol. 31, 2011 359

and testing of various applications including automotive assembly [6],

computational intelligence [7], bioassay robustness studies [8] and
many others.
The objective of this paper is to apply the DOE technique as
a screening tool for electromagnetic simulations. The effect of the
input variables on the output variables is intended to be evaluated.
In particular, the input variables — which are also called factors or
parameters — will include (for example) the shape parameters of an
electric machine, where the output variables — which are the objective
functions to be evaluated — will include (for example) the torque
produced by an electric machine.
The remaining sections of the paper are organized as follows.
Section 2 describes the main concepts and typical methods used in
the design process of any system comprising several factors. Section 3
discusses a practical implementation of the DOE technique using a tool
built on Matlab and Java for efficient use of the technique. Section 4
presents three different applications to which the DOE is applied in
order to facilitate their design and optimization processes. These
applications are magnetizer, Problem 25 of Team Workshop and a
magnetic refrigeration system. Results from such systems are also
presented in Section 4. Finally, the overall paper conclusions is drawn
in Section 5.

2. METHODOLOGY

Designing an electromagnetic device or system, is usually a study of a

phenomenon depending on different factors. The intuitive (classical)
method is not always the best choice. The classical method consists
generally of fixing the level of all variables except one and then
measuring the response for several values of the variable factor.
Let’s assume for example that an electric machine is aimed to
be optimized. To simplify the problem we assume that only the global
dimensions (the length and the external radius) are the varying factors.
Thus the length is the first factor and is denoted by x1 and the external
radius is the second factor and is denoted by x2 . Each factor can take
several values between {x1min , x1max } and {x2min , x2max }. We desire to
study the influence of each of these factors on the system response or
output (torque) called Y . The classical or traditional approach is to
study the two factors x1 and x2 , separately.
First we fix x2 at the average level x2average and study the response
of the system when x1 varies from x1min to x1max following, for example,
4 steps (experiments or simulations) as shown in Figure 1. We then
repeat the same experience to study the influence of x2 . The total
360 Bouchekara, Dahman, and Nahas

Figure 1. Traditional method of experiments.

Figure 2. One experience at each node of the mesh.

number of tests is 8. Nevertheless, one can ask if we have a good

knowledge about the system with these 8 experiences. It is obvious
that the answer of this question is no.
To get a better knowledge about the system, we have to mesh the
validity domain of the two factors and test each node of this mesh as
shown in Figure 2. In this case, we have to achieve 4 × 4 = 42 = 16
experiences.
However, in this example only two factors are taken into account.
If the number of factors increases to 7 for example, the number of
tests to be performed will rise to 47 = 16384 experiences, which is a
time- and cost-consuming process. Knowing that it is impossible to
reduce the number of values that one factor can take less than 2, the
designer often reduces the number of factors, which leads to incertitude
of results. To reduce both cost and time, the DOE is used where it
aims to establish a design experiment with less number of tests. The
DOE, for example, allows identifying the influence of 7 factors with 2
Progress In Electromagnetics Research B, Vol. 31, 2011 361

Permeability Speed of rotation Teeth

Length External radius Temperature
Number of poles

Length Temperature

Figure 3. Schematic of a screening experimental design, where many

factors are reduced to a significant few.

points per variable with only 8 or 12 tests rather than 128 tests with
the traditional method [5]. When the number of factors increases a
screening process using the DOE technique should first be adopted to
reduce this number. This screening process is schematically shown in
Figure 3. It shows an example of parameters screening where the initial
factors are: permeability, length, speed of rotation, external radius,
teeth, temperature and number of poles. Then, after the screening
process, only two parameters are influent on the output (temperature
and length).

2.1. Mathematical Concept

Assume that y is the response (or output) of an experiment (or
a simulation) and {x1 , x2 , x3 , . . . , xk } are k factors acting on this
experiment where each factor has two levels of variation xi− and xi+ .
To predict the value of y, it is approximated by an algebraic model
given by the following equation:
y = a0 + a1 x1 +a2 x2 + . . . + ak xk + . . . + a1 x1 x2 + . . . + a1 x1 xk
+a1...k x1...k (1)
where aj are coefficients which represent the effect of factors and their
interactions on the response of the experiment.

2.2. Full Factorial Design

The study of full factorial design consists of exploring all possible
combinations of the factors considered in the experiment [9]. Note that
the design X k means that this experiment concerns a system with k
factors with X levels.
362 Bouchekara, Dahman, and Nahas

Usually, two values of the X’s (called levels) are used. The use of
only two levels implies that the effects are monotonic on the response
variable, but not necessarily linear [5]. For each factor, the two levels
are denoted using the “rating Yates” notation (named after its author)
by: −1 the low level of each factor, +1 the high level of each factor
(Figure 4). Thus, the number of experiments carried out by a full
factorial design with 2 levels is given by:

n = 2k (2)
where k is the number of factors to be considered.
Figure 4 shows the design matrix of a full factorial design for 2
factors and the mesh of the experimental field where points correspond
to nodes.

Test Factor Factor Response Factor

number x1 x2 y
x2 Y Y
1 -1 -1 Y1

2 -1 +1 Y2 Mes

3 +1 -1 Y3 Yh Y

4 +1 +1 Y4 1 3
Factor
(a) (b)
Figure 4. Full factorial design for 2 factors and 2 levels. (a) Design
matrix. (b) Strategy of experimentation; points corresponding to
nodes in the mesh of the experimental field.

The advantage of full factorial designs, is the ability to estimate

not only the main effects of factors, but also all their interactions in
two by two, three by three, etc., up to the interaction involving the k
factors. However, when the number of factors increases, the use of such
a design leads to a prohibitive number of experiments or simulations
to perform.
The question to be asked is then: is it necessary to perform all
experiments of the full factorial design to estimate the system’s model?
In other words, is it necessary to conduct a test at each node of the
mesh?

2.3. Fractional Factorial Design

It is not necessary to identify the effect of all interactions of the
analytical model given by Equation (2), because the interactions
Progress In Electromagnetics Research B, Vol. 31, 2011 363

of order ≥ 2 (like x1 x2 x3 ) are usually negligible. To illustrate

this phenomenon, an analogy can be made with a Taylor series
approximation where the information given by each term decreases
if the order of this one increases. So, fractional factorial designs can
be used to estimate factors effect and interactions that act more on
the experiments with a reduced number of experiments [10]. Taguchi
tables [9], or G. Box generators [11], give the fractional factorial design
matrix of experiments [12].
To illustrate fractional factorial designs let’s take an example,
if the number of factors is k = 3, the design matrix of these three
factors is given by G. Box generators in a way that the third factor
is the product of the two other factors. It is said that factor x3
and interaction x1 x2 are confused, or x3 and x1 x2 are aliased and
there is a confusion of these aliases because only their sums are
reachable [9, 10, 12].

2.4. Estimation of Model Coefficients

The value of the coefficient a0 is estimated from the arithmetic average
of all observed responses and it is given by:
n
1X
a0 =ȳ= yi (3)
n
i=1
where yi is the response observed for the experiment i and n is the
number of experiments.
The effect of a factor xj at the level xj+ can be calculated thus,
the coefficient associated with this effect can be identified by using the
following equations:

aj = ej = yx+j − a0 (4)
and
n
1 X +
yx+j = yi (5)
n+
i=1

where yx+j is the response observed for experiment i when xj is at level

xj+ , n+ is the number of experiments where xj is at the level xj+ and
eaj is the effect of coefficient aj .
Once the method of how to calculate the coefficients of the model
and how to identify the existing confusion between these factors has
been presented, we can evaluate the contributions of contrasts (the
sum of confusions) and therefore the most significant factors (affecting
364 Bouchekara, Dahman, and Nahas

the response). In [11] the identification of the significant factors has

been proposed by evaluating the coefficients contribution (or contrasts,
for fractional designs) on the model response from the normalization of
their values compared to the sum of squared responses, such as given
in the following equations:
SCE (aj )
Caj = [%] (6)
SCE (y)
with
n
X
SCE (y) = (yi − y)2 (7)
i=1
s
n X ¡ ¢2
SCE (aj ) = eaj (8)
s
j=1

where S is the number of levels (equals to 2 in this case), eaj is the effect
of coefficient aj , and Caj is the contribution of the contrast associated
with the coefficient aj .
According to [13]:
• The contribution given by (6) is deemed significant if Caj ≤ 5%.
• The interactions of order higher than two are negligible.
• If a contrast is negligible, all effects composing this contrast are
negligible also.
• Two significant factors can generate a significant interaction.
On the other side, two insignificant factors do not generate a
significant interaction.

3. IMPLEMENTATION

For an efficient use of the DOE methodology, it has been implemented

in the form of interactive tool called Design of experiments Tool
(DOET) using a combination of Matlab and Java. Matlab is
an efficient software which puts the powerful calculation function,
visual and program designing together in an easily used development
environment. Java is a cross-platform program development language
which is created by Sun. It is the most advanced program language
which also has the richest characteristic and has the most powerful
function. In DOET, Matlab is used for all calculation functions, and
java is used to generate interactive User Interfaces.
Progress In Electromagnetics Research B, Vol. 31, 2011 365

(a) (b)
Figure 5. Magnetizer to be studied; (a) geometry, (b) contour of the
solution.

4. APPLICATIONS

4.1. Application to Magnetizer

4.1.1. Description
The application of the DOE technique is first demonstrated on a
magnetizer where the geometry is shown in Figure 5. The study
and optimization of this magnetizer is given in [9, 10]. The shape
of the pole face is to be optimized. In the finite element model the
object to be magnetized is treated as if it were made of nonmagnetic
material. A permeability value very close to that of air is assigned to
that region (µr = 10). A high current is applied to the coil. The linear
magnetostatic field analysis is carried out in 2-D using “finite element
method” (FEM).

4.1.2. Objective Function and Constraints

The goal of the optimization is to achieve a constant magnetic flux
density distribution along chord A-B positioned halfway through the
width of the magnetized piece and subtending an angle of π/3.
The constraints of x1 , x2 , x3 , x4 and x5 are represented in Table 1.
To avoid intersection of the pole with the material to magnetize
and to ensure an air-gap at least equal to 2 mm, the parameters are
366 Bouchekara, Dahman, and Nahas

Table 1. Parameters and their constraints.

Parameter Minimum Value [mm] Maximum Value [mm]

x1 0 15
x2 35 45.9
x3 25 26.5
x4 15 20
x5 70 80

subject to an additional geometric constraint. This constraint is given

by q
x1 + x22 + (x3 − x1 )2 ≤ 53 (9)
The optimization problem can be formulated as follows:
1
Fobj = (10)
1 + fobj
s µn ¶2
Pn
2
P
n Bi − Bi
i=1 i=1
fobj = (11)
P
n
Bi
i=1
where Bi is the value of magnetic flux density at a point i of the arc
A-B, and n = 50 is the total number of points on the arc.
When the geometric constraint is violated, fobj is expressed by a
penalty term given by:
µ q ¶
2 2
fobj =λ1 +λ2 x1 + x2 +(x3 −x1 ) −53 (λ1 =λ2 = 10) (12)

4.1.3. Identification of Significant Parameters Using the DOE

The aim of this section is to identify the most influential factors on the
objective function using the DOE. As explained earlier the use of two-
level full factorial design needs 25 = 32 runs (simulation) to evaluate
the objective function. Using Generators of G.Box, the possible designs
are 25−2 = 8 runs and 25−1 = 16 runs.
Design 1: the choice of a 25−2 design means that we have a 2 levels
design with 5 factors where 2 of these factors are generated using the
other 3 factors. In other words; the factor (4) will be generated using
the product of factors (1) and (2). The factor (5) will be generated
using the product of factors (1) and (3) (see Table 2 and Figure 6).
Progress In Electromagnetics Research B, Vol. 31, 2011 367

Table 2. Generators of G. Box for 5 factors.

Resolution Design name Number of tests Generators
3 25−2 8 4 = ±1, 2 5 = ±1, 3
5 25−1 16 5 = ±1, 2, 3, 4

Table 3. Design matrix of generated by the 25−2 fractional factorial

design.
x1 [mm] x2 [mm] x3 [mm] x4 [mm] x5 [mm] F
0 35 25 2 8 0.6102
0 35 26 2 7 0.4973
0 46 25 15 8 0.6085
0 46 26 15 7 0.4969
15 35 25 15 7 0.4903
15 35 26 15 8 0.5884
15 46 25 2 7 −0.0019
15 46 26 2 8 −0.0019

Figure 6. Contributions obtained with a 25−2 fractional factorial

design: The contrasts (left side), and the influence on the objective
function (right side)

Design 2: the choice of a 25−1 design means that we have a 2 levels

design with 5 factors where one of these factors is generated using the
other 4 factors. This means that the factor (5) will be generated using
the product of factors (1), (2), (3) and (4) (Table 2 and Figure 7).
So, using a fractional design will effectively reduce the number of
runs from 32 to 16 or 8 depending on the chosen design. Thus the time
of simulation using Design 1, and Design 2 is reduced to 50% and 75%
of the initial time, respectively.
368 Bouchekara, Dahman, and Nahas

Figure 7. Contributions obtained with a 25−1 fractional factorial

design: The contrasts (left side), and the influence on the objective
function (right side).

Figure 8. Graph of the main effects.

Progress In Electromagnetics Research B, Vol. 31, 2011 369

Table 4. Design matrix of generated by the 25−1 fractional factorial

design.

x1 [mm] x2 [mm] x3 [mm] x4 [mm] x5 [mm] F

0 35 25 15 8 0.6102
0 35 25 2 7 0.4973
0 35 26 15 7 0.4973
0 35 26 2 8 0.6085
0 46 25 15 7 0.4972
0 46 25 2 8 0.6085
0 46 26 15 8 0.6048
0 46 26 2 7 0.4969
15 35 25 15 7 0.4903
15 35 25 2 8 0.6065
15 35 26 15 8 0.5884
15 35 26 2 7 0.5
15 46 25 15 8 −0.0019
15 46 25 2 7 0.0019
15 46 26 15 7 −0.0019
15 46 26 2 8 −0.0019

The DOET allows running this problem. The design matrix

(values of the parameters used in each experience, as well as the values
of the objective function for each configuration), is shown for Design 1
in Table 3 and for design 2 in Table 4.
The contributions of contrasts on the objective function of this
case study are given for designs 1 and 2 in Figure 6 and Figure 7,
respectively. Each figure illustrates the contrasts and their influence
on the objective function. The results of Figures 9 and 10 are similar;
this shows the accuracy of the DOE technique.
Figure 11 depicts the main effects of each factor. Based on the
results shown in Figure 6, Figure 7 and Figure 8 it can be concluded
that among the factors analyzed in this study, factor x1 and x2 are the
ones that influence most significantly the response variable. Moreover,
the interaction x1 x2 also influences significantly the response variable.
It is important to highlight that the number of factors to be considered
in this optimization process is significantly reduced from 5 to 2 factors
only.
370 Bouchekara, Dahman, and Nahas

Table 5. Parameters and their constraints.

Parameter Minimum Value [mm] Maximum Value [mm]

R1 5.0 9.4
L2 12.6 18
L3 14.0 45.0
L4 4.0 19,0
A1 170.0 190.0
A2 70.0 90.0
A3 86.0 88.0
A4 9.5 11.0

(a) Whole view (b) Enlarged view

Figure 9. Model of die press with electromagnet.

4.2. Application to the Problem 25 of Team Workshop

4.2.1. Description
The aim of this problem is to obtain the shape of a die molds used for
producing anisotropic permanent magnet by using the optimization
method. The model can be assumed as two-dimensional. The die
mold is described by an internal circle of radius R1 and by an external
ellipse represented by L2 , L3 and L4 . It is required to find the values
of R1 , L2 , L3 and L4 such that a constant radial magnetic induction
on ten different points defined on the arc ef , as shown in Figure 9, can
be obtained.
Progress In Electromagnetics Research B, Vol. 31, 2011 371

To show the efficiency of the DOE Method in the identification

of significant parameters, 4 new parameters (A1 , A2 , A3 and A4 )
have been added to the original problem [5]. We can identify these
parameters in Figure 9, and their respective constraint in Table 5.

4.2.2. Objective Function and Constraints

The objective function describing this problem W is given by
Xn n o
W = (Bxip − Bxio )2 + (Byip − Byio )2 (13)
i−1
where n = 10 is the number of specified points. The subscripts p and
o refer to the calculated and specified values respectively. Bxo and Byo
along the line e-f are specified as follows:
(
Bxo = 0.35cos θ (T)
(14)
Byo = 0.35sin θ (T)
The constraints on R1 , L2 , L3 , L4 , A1 , A2 , A3 and A4 are given
in Table 5.

4.2.3. Identification of Significant Parameters Using the DOE

The aim of this section is to identify the most influential factors on
the objective function using the DOE. As explained earlier, the use
of two-level full factorial design needs 28 = 256 runs (simulations) to
evaluate the objective function. Using a 28−4 fractional factorial design
will reduce effectively the number of runs from 256 to 16. Thus the
time is reduced to around 93% of the initial time. The 28−4 design is
given using Generators of G. Box as shown in Table 6 The choice of
a 28−4 means that we have a 2 levels design with 8 factors where 4 of
these factors are generated using the other 4 factors. So we can write
that:

Table 6. Generators of G. Box for 8 factors.

Resolution Design name Number of tests Generators

5 = ±2, 3, 4 6 = ±1, 3, 4
4 28−4 16
7 = ±1, 2, 3 8 = ±1, 2, 4
6 = ±1, 2, 3 7 = ±1, 2, 4
4 28−3 32
8 = ±2, 3, 4, 5
5 28−2 64 7 = ±1, 2, 3, 4 8 = ±1, 2, 5, 6
372 Bouchekara, Dahman, and Nahas

Figure 10. Design matrix generated by the 28−4 fractional factorial

design.

Figure 11. Contributions obtained: the contrasts (left side), and the
influence on the objective function (right side).
Progress In Electromagnetics Research B, Vol. 31, 2011 373

• The factor (5) will be generated using the product of

factors (2), (3), and (4).
• The factor (6) will be generated using the product of
factors (1), (3), and (4).
• The factor (7) will be generated using the product of
factors (1), (2), and (3).
• The factor (8) will be generated using the product of
factors (1), (2), and (4).
The design matrix (values of the parameters used in each
simulation, as well as the values of the objective function for each
configuration), is shown in Figure 10.
The contributions of contrasts obtained are given in Figure 11.
As set earlier a contribution is significant if it is higher than 5%.
High order interactions (higher than 2) are considered negligible and
only interactions of significant parameters are also significant. Thus,
factors R1 (14, 9%), L2 (25%), L4 (6.2%) and interactions R1 L2 (33%),
R1 L4 (8.3%) and L2 L4 (6.1%) are the only significant factors for the
objective function value. So, for an optimization process only R1 , L2
and L4 are considered. This represents a reduction of 93% of the
number of factors and consequently causes a significant time saving.

4.3. Application to a Magnetic Refrigeration Prototype

Magnetic refrigeration (MR) becomes a promising competitive
technology to the conventional gas-compression/expansion [14]. MR
is a compact, reliable and efficient technology since it does not
require compressor (which is the most inefficient part in conventional
refrigeration). The MR is based on the magneto-caloric effect
(MCE) [14–16]. The MCE was first discovered in iron compound
by Warburg 1881 [14]. The MCE is defined as the response of a
solid material to an applied magnetic field (generated with permanent
magnet for instance), which appears as a change in its temperature.
When such solid materials are placed in a magnetic field, their
temperature increases and when the materials are removed from the
magnetic field, they cool down.

4.3.1. Description
The MR structure studied here is quite similar to a rotating machine as
shown in Figure 12. The stator of this machine consists of a cylindrical
yoke and two pairs of containers called beds (b1 , b2 , b3 and b4 ) filled
with MCE materials (for instance Gadolinium). The beds are placed
around a rotating magnet (rotor). The yoke has two major roles: the
374 Bouchekara, Dahman, and Nahas

Figure 12. Configuration and operating principle of the PM cooling

machine.

first one is to canalize the magnetic flux and the second one is to
support the containers. The rotation of the permanent magnet (PM)
will generate cycles of varying magnetic field between Bhigh and Blow .
An electromagnetic study is undertaken to compute the profiles of
the variation of magnetic field, torque and forces. The electromagnetic
study is achieved by simulating the system using FEM. Due to
symmetrical geometry of the chosen configuration, the investigations
were performed in two dimensions. The problem has been solved
in magnetostatic formulation. In these studies, the remanent
magnetization and the relative permeability of the magnet are fixed
respectively to 1.46 T and 1.046 (NdFeB magnets). The stator is
described by a constant permeability equals to 1000 and the MCE
material by an isotropic bulk gadolinium material with a relative
permeability of 5.

4.3.2. Objective Function and Constraints

Since the magneto-caloric effect depends on the magnitude of the
magnetic field, the performance of this machine are directly related
to the magnetic field range ∆B. Thus, the ∆B should be the first
criterion to be taken into account in the design and the optimization
process. The objective for this criterion is to maximize the ∆B. The
second criterion is the minimization of the magnetic efforts (Forces and
Torque). The objective function for the optimization process is given
by:
F objective = {f1 , f2 , f3 } (15)
where
f1 = max (∆B) (16)
f2 = min (Forces) (17)
f3 = min (Torque) (18)
Progress In Electromagnetics Research B, Vol. 31, 2011 375

4.3.3. Identification of Significant Parameters Using the DOE

Since eight parameters define the shape of the machine, it is advisable
to determine the effect of each parameter on the objective function.
Thus, it is very important to provide proper parameter ranges. The
considered parameters (names, definitions, ranges and types) are listed
in Table 7. There are two types of parameters: continuous parameters
(which can take any value inside the defined range like the length of
the machine, the radius of the rotor, etc.) and discrete parameters
(which can take only the limits of the defined range like the number
of blocks; since a 2 level design is used). Using two-level full factorial
design needs 28 = 256 runs (simulations) to evaluate the objective
function. Using a 28−4 fractional factorial design will significantly
reduce the number of runs from 256 to 16. The 28−4 design is given
using Generators of G. Box as shown in Table 8. In this application,
the output function is a multi-objective function rather than a single-
objective function as in application 1 and 2. The results obtained
when the procedure described above is run are given in Table 8 and

Table 7. Considered parameters.

Initial Minimum Maximum
Name Description Type
Value Value Value
Length of
L 100 mm 90 mm 110 mm Continuous
the machine
External radius
Re 88.5 mm 80.5 mm 99.5 mm Continuous
of the yoke
Internal radius
Ri 73.5 mm 73.5 mm 77.5 mm Continuous
of the yoke
Rr Radius of the rotor 53.5 mm 47.5 mm 53.5 mm Continuous
Angular size of
ar 60◦ 45◦ 60◦ Continuous
the rotor
Angular size of the
ab magnetocaloric 45◦ 30◦ 45◦ Continuous
material Blocks
Width of the
wb magnetocaloric 17 mm 10 mm 17 mm Continuous
material Blocks
Number of the
Nb magnetocaloric 4 2 6 Discreet
material Blocks
376 Bouchekara, Dahman, and Nahas

Table 8. Design matrix generated by the 28−4 Box-Wilson fractional

factorial design and the simulation results.

L Re Ri Rr ar ab w b ∆B Torque Forces
No Nb
[mm] [mm] [mm] [mm] [ o] [o ] [mm] [T] [N.m] [N]
1 90 85.5 73.5 47.5 45.3 0 10 2 0.56579 8.3397 -124.95
2 90 85.5 73.5 53.5 60.4 5 10 6 0.81441 4.0532 -532.66
3 90 85.5 77.5 47.5 60.4 5 17 2 0.61438 19.4440 -354.97
4 90 85.5 77.5 53.5 45.3 0 17 6 0.85772 16.5120 -290.60
5 90 99.5 73.5 47.5 60.3 0 17 6 0.85831 6.1528 -266.31
6 90 99.5 73.5 53.5 45.4 5 17 2 0.91797 40.4441 -645.45
7 90 99.5 77.5 47.5 45.4 5 10 6 0.44317 2.0692 -154.36
8 90 99.5 77.5 53.5 60.3 0 10 2 0.77045 16.5523 -253.22
9 110 85.5 73.5 47.5 45.4 5 17 6 0.63619 6.1388 -375.38
10 110 85.5 73.5 53.5 60.3 0 17 2 1.17272 55.1549 -617.04
11 110 85.5 77.5 47.5 60.3 0 10 6 0.55053 3.1761 -163.76
12 110 85.5 77.5 53.5 45.4 5 10 2 0.61404 16.9514 -363.43
13 110 99.5 73.5 47.5 60.4 5 10 2 0.60799 14.7476 -372.99
14 110 99.5 73.5 53.5 45.3 0 10 6 0.80641 14.1823 -293.53
15 110 99.5 77.5 47.5 45.3 0 17 2 0.61751 16.7636 -181.47
16 110 99.5 77.5 53.5 60.4 5 17 6 0.85549 5.8050 -749.81

Table 9. Contrasts and contribution obtained.

Contrasts ∆B Significant? Torque Significant? Forces Significant?

L 0 No 1 No 3 No
Re 0 No 0 No 0 No
Ri 14 Yes 6 Yes 6 Yes
Rr 46 Yes 18 Yes 36 Yes
ar 8 Yes 0 No 9 Yes
ab 6 Yes 2 No 22 Yes
wb 23 Yes 16 Yes 18 Yes
Nb 0 No 35 Yes 0 No
L*Re+Ri*wb+Rr*Nb+ar*ab 1 No 5 Yes 0 No
L*Ri+Rr*ab+Re*wb+ar*Nb 0 No 4 No 1 No
L*Rr+Ri*ab+Re*Nb+ar*wb 0 No 0 No 0 No
L*ar+Re*ab+Rr*wb+Ri*Nb 1 No 4 No 3 No
L*ab+Ri*Rr+Re*ar+wb*Nb 0 No 9 Yes 0 No
L*wb+Re*Ri+Rr*ar+ab*Nb 0 No 1 No 1 No
L*Nb+Re*Rr+Ri*ar+ab*wb 1 No 1 No 0 No
Progress In Electromagnetics Research B, Vol. 31, 2011 377

Table 9. The significance of each parameter on the output functions

are summarized in Table 9.

5. CONCLUSION

Traditional design methodologies — based on physical prototyping of

a system — are often seen expensive and time-consuming processes.
In order to optimize a design process while reducing design costs, the
Design of Experiments (DOE) approach is a widely advisable solution.
The use of DOE technique, in conjunction with computer simulation,
allows for more efficient analysis of the simulated models. In this paper,
the DOE technique was introduced and evaluated. In particular, the
effect of significant factors and their interaction with the objective
functions of electromagnetic simulations were investigated. The use
of this technique was illustrated using three case studies. The results
presented in the paper demonstrated that the DOE approach has a
great potential to reduce the required simulation time in all cases. In
more detail, the number of factors was reduced from 5 to 2 and from
8 to 3 in the first and second cases, respectively. In the third case,
the DOE was applied to a multi-objective design and optimization
process related to a magnetic refrigeration system. The study showed
that the application of DOE can reduce the number of simulations
from 256 to 16, thereby reducing the computational cost significantly
(by approximately 94%). From the results of this case study, one can
analyze the influence of each parameter on each output of the multi-
objective function. Since magnetic refrigeration systems involve many
design parameters, it was demonstrated that the DOE technique would
be very helpful in the design and optimization of such systems.
The approach suggested in this paper is aimed at trying to
quantify the influence of parameters on an objective function. This
approach is very interesting and it could be a first step into a
long design optimization process. Moreover, the effectiveness of the
presented approach can be further explored through applying it to a
broad range of design applications where the design problem depends
on a large set of varying parameters. Areas of interest may include
antenna design, electronic circuit design and automatic control design.

REFERENCES

1. Mohammed, O. A., D. A. Lowther, M. H. Lean, and B. Alhalabi,

“On the creation of a generalized design optimization environment
for electromagnetic devices,” IEEE Transactions on Magnetics,
Vol. 37, No. 5, 3562–3565, 2001.
378 Bouchekara, Dahman, and Nahas

2. Gang, L., G. Y. Yang, K. R. Shao, Y. Guo, J. Zhu, and

J. D. Lavers, “Electromagnetic device design based on RBF
models and two new sequential optimization strategies,” IEEE
Transactions on Magnetics, Vol. 46, No. 8, 3181–3184, Aug. 2010.
3. Carcangiu, S., A. Fanni, A. Mereu, and A. Montisci, “Grid-
enabled tabu search for electromagnetic optimization problems,”
IEEE Transactions on Magnetics, Vol. 46, No. 8, 3265–3268, 2010.
4. Nikolova, N. K., “Solving design and inverse-imaging problems
through electromagnetic simulation,” 17th International Con-
ference on Microwaves, Radar and Wireless Communications,
MIKON 2008, 1–8, 19–21, 2008.
5. Sanchez, S. M., “Work smarter, not harder: Guidelines for
designing simulation experiments,” Proceedings of the Winter
Simulation Conference, 4–7, 2005.
6. Altayib, K. and A. Ali, “Improvement for alignment process
of automotive assembly plant using simulation and design of
experiments,” International Journal of Experimental Design and
Process Optimisation, Vol. 2, No. 2, 145–160, 2011.
7. Garcı́a, S., A. Fernández, J. Luengo, and F. Herrera, “Advanced
nonparametric tests for multiple comparisons in the design
of experiments in computational intelligence and data mining:
Experimental analysis of power,” Information Sciences, Vol. 180,
No. 10, 2044–2064, 2010.
8. Kutlea, L., N. Pavlovića, M. Dorotića, I. Zadroa, M. Kapustića,
and B. Halassy, “Robustness testing of live attenuated rubella
vaccine potency assay using fractional factorial design of
experiments,” Vaccine, Vol. 28, No. 33, 5497–5502, 2010.
9. Kleijnen, J. P. C., S. M. Sanchez, T. W. Lucas, and T. M. Cioppa,
“State-of-the-art review: A user’s guide to the brave new world
of designing simulation experiments,” Journal on Computing,
Vol. 17, No. 3, 263–289, 2005.
10. Uy, M. and J. K. Telford, “Optimization by design of experiment
techniques,” IEEE Aerospace Conference, 1–10, 7–14, 2009.
11. Pillet, M., “Les plans d’expériences par la méthode taguchi,” Les
Editions d’Organisation, ISBN 2-70-812031-X, 1997.
12. Costa, M. C., “Optimisation de dispositifs electromagnétiques
dans un contexte d’analyse par la méthode des eléments finis,”
Ph.D. Thesis National Polytechnic Institute of Grenoble, 2001.
13. Demonsant, J., “Comprendre et mener des plans d’expériences,”
Afnor, ISBN 2-124-75032-1, 1996.
14. Hong, S. C., S. J. Kim, E. J. Hahn, S. Park, and C. S. Kim,
Progress In Electromagnetics Research B, Vol. 31, 2011 379

“Magnetic refrigeration properties of La0.8 Ca0.2 Mn0.9957

Fe57
0.01 O3 ,” IEEE Transactions on Magnetics, Vol. 45, No. 6, 2565–
2568, Jun. 2009.
15. Bruck, E., O. Tegus, X. W. Li, et al., “Magnetic refrigeration-
towards room-temperature applications,” Physica B: Condensed
Matter, Vol. 327, Nos. 2–4, 431–437, Apr. 2003.
16. Dong, A. and X. Lu, “A new permanent magnet system for
rotating magnetic refrigerator,” IEEE Transactions on Applied
Superconductivity, Vol. 20, No. 3, 834–837, Jun. 2010.