Beruflich Dokumente
Kultur Dokumente
Abstract
We propose a method to decide the appropriate arrangement of both fictitious charges and
contour points in the charge simulation method for the case that the shape of electrodes can be
expressed in the spherical coordinates. In this paper, we use genetic algorithm (GA), which is
inspired by the mechanism of natural selection where stronger individuals are likely the
winners in a competing environment, as a search method. We calculated the potential
distribution around two spherical electrodes with different potentials above a grounded
electrode plate. The surface potentials of the electrodes were assumed to be 5000 and 30 V;
respectively, for example. The total number of the charges was 22 for the electrode system.
First, we placed fictitious point-charges and contour points at random inside the electrodes
and on the surfaces of electrodes, respectively. The arrangement was automatically adjusted by
using GA until the potential reached the desired values at all the test points placed on the
electrode surfaces. We show that the GA can be useful to determine the appropriate
arrangement of both fictitious charges and contour points simultaneously.
r 2002 Elsevier Science B.V. All rights reserved.
Keywords: Electric field analysis; Charge simulation method; Fictitious charges; Contour points; Genetic
algorithm; Arrangement problem; Spherical electrode; Numerical field calculation
$
Original version presented at ICAES’2001, Fourth International Conference on Applied Electro-
statics, 8–12 October 2001, Dalian, China.
*Corresponding author. Fax: +81-857-31-0880.
E-mail address: ryo@ele.tottori-u.ac.jp (R. Nishimura).
0304-3886/03/$ - see front matter r 2002 Elsevier Science B.V. All rights reserved.
PII: S 0 3 0 4 - 3 8 8 6 ( 0 2 ) 0 0 1 7 0 - 5
338 R. Nishimura et al. / Journal of Electrostatics 57 (2003) 337–346
1. Introduction
The principle of the charge simulation method (CSM) is easy to understand and to
implement computationally. Because of these advantages, CSM is often used for
electric field analyses, such as calculation of potential distribution around electrodes.
However, appropriate arrangements of fictitious charges and contour points are
indispensable to obtain accurate solutions. To date, these arrangements have been
carried out empirically.
2. Purpose of study
In CSM calculation, point charges must be used when an electrode system with
3-D asymmetrical structures are assumed. The more complex the electrode shape is,
the larger the number of point charges are required. For such cases, it takes times
and efforts to decide the appropriate arrangement of the fictitious charges. In order
to remove the inconvenience, it is desired that the appropriate arrangement should
be carried out automatically. We have been shown that an appropriate arrangement
of both fictitious charges and contour points can be obtained by using genetic
algorithm (GA) for the electrode system that consists of cylindrical electrodes [1,2].
In this paper, we propose a method to determine an appropriate arrangement of
both fictitious charges and contour points by using GA for spherical coordinate
system. We assume an electrode system that consists of two spherical electrodes
above a grounded conductive electrode plane. In order to enable asymmetrical
calculations, we consider the angles y and j used in polar-coordinate expressions for
yA½0; p and jA½0; 2pÞ:
4. Numerical model
Fig. 1 shows the two-electrode system that consists of two spherical electrodes.
The centers of both electrodes are on z-axis. We assume that the electrodes
are placed above a grounded plate electrode. So we can use the mirror images
of the fictitious charges inside the electrodes. As shown in this figure, a point
P inside an each electrode can be expressed by using spherical coordinate
system whose origin o0 is set at the center of each electrode as Pðr; y; jÞ; where
r is the distance from the center of the sphere to which the point P belongs, yA½0; p is
an angle between the z0 axis and the line drawn from the origin to P; and jA½0; 2pÞ is
an angle between x0 axis and the line vertically projected on x0 y0 -plane from the point
P: In this figure, the axes x0 ; y0 and z0 are parallel to the global axes x; y and z;
respectively.
5. Calculations
5.1. Procedure for decision of arrangements of fictitious charges and contour points
using GA
In this study, ten individuals (parents) make breeding pairs in round-robin system.
Two children are produced from a single pair by using the crossover and mutation
procedures. The chromosomes of the 100 individuals, sum of the numbers of the
parents and the children, are decoded into the positions of the charges and contour
points. The set of simultaneous equations expressing the relations among the
potentials at each contour point on the electrode surface and the arrangement of
both the contour points and fictitious charges is constructed for each individual. If
the coefficient matrix of an individual is singular, the individual is regarded as a
‘‘dead individual’’ and deleted because the simultaneous equation cannot be solved.
The evaluation parameter s; which is described later, is calculated for the rest of NL
individuals ðNL p100Þ: The parameter s should be defined by a concept that as s gets
closer to zero, the surface potentials approach the given values all over the surfaces
of the electrodes. Ten elite individuals are selected by using the smallest values of s in
order and become the parents of the ‘‘next generation’’.
The bit sequences of ten individuals of the ‘‘0th generation’’, the initial condition,
are created by using uniformly random numbers. In this paper, the calculation
terminates when the iteration number of generation reaches 5000.
In this study, we assume an electrode system that consists of two electrodes. The
surface potentials of both the electrodes have to approach the given values all over
the surfaces of the electrodes. We introduce the evaluation parameter s defined by
the following procedure:
342 R. Nishimura et al. / Journal of Electrostatics 57 (2003) 337–346
First, the maximum values of the relative errors of the electrode surface potentials
are calculated for each electrode ðs1 ; s2 Þ:
f1 ðyi ; jj Þ f10
s1i;j ¼ 100;
ð7Þ
f10
f2 ðyi ; jj Þ f20
s2i;j
¼ 100; ð8Þ
f
20
s1 ¼ max½s1i;j ; ð9Þ
s2 ¼ max½s2i;j ; ð10Þ
where f1 ðyi ; jj Þ and f2 ðyi ; jj Þ are the potentials at test points on the surfaces of the
Electrodes 1 and 2, respectively. In this calculation, yi and jj are calculated by the
following formulae:
i
yi ¼ p ði ¼ 0; 1; y; 20Þ; ð11Þ
20
j
jj ¼ p ðj ¼ 0; 1; y; 39Þ: ð12Þ
20
Potentials f10 and f20 are given surface potentials of the Electrodes 1 and 2,
respectively, and s1i and s2j are relative errors at each test point.
Then, the evaluation parameter s is defined by the following formula:
s ¼ max½s1 ; s2 : ð13Þ
Table 1
Parameters used for GA calculation
Table 1 shows the numerical conditions used for the calculation. The numerical
conditions in this table are given to the computing program as an input data file.
Though this electrode system has an axisymmetric structure, the calculation is
carried out for j that ranges from 0 to 2p:
Figs. 3 and 4 show the convergences of surface potentials of the electrodes at
y ¼ 0 cm plane. We can see that the surface potential of Electrode 1 is imitated well.
However, pronounced deviations from the specified potential are seen for Electrode
2 at a region near Electrode 1. This problem will be solved by increasing the numbers
of fictitious charges and contour points belong to Electrode 2. The evolution was
completed and convergence achieved by about 1000th generation.
6. Discussions
A small displacement of the charges or the contour points from their appropriate
positions gives a small value of the electrode potential error. As shown in Figs. 3(b)
and 4(b), the absolute potential error of Electrodes 1 and 2 at 5000th generation is
almost the same value. However, for a low-voltage electrode (Electrode 2), the small
value of the absolute error becomes large value of relative error. Because of this, the
deviations occurred only for Electrode 2.
In this research, we tentatively assign 10 and 12 fictitious charges to Electrodes 1
and 2, respectively, as shown in Table 1. In order to improve the accuracy of the
solution, appropriate numbers of the charges and the contour points should be
assigned to each electrode.
Fig. 3. Surface potential of Electrode 1 at y ¼ 0 cm: (a) 0th generation and 5000th generation. (b) 5000th
generation (enlargement of (a)).
344 R. Nishimura et al. / Journal of Electrostatics 57 (2003) 337–346
Fig. 4. Surface potential of Electrode 2 at y ¼ 0 cm: (a) 0th generation and 5000th generation. (b) 5000th
generation (enlargement of (a)).
Table 2
Computer resources used for the calculations
The computer resources used for the calculation are shown in Table 2. The
computing time for a single run (5000 generations) is about 14 h and the convergence
time (1000 generation) is about 3 h: When good individuals are being selected during
the calculation, a descending sort of the evaluation parameters is carried out by using
bubble-sorting method at each generation of the evolution. Also the computer
program writes some data on the hard disc during the calculation. Because of these,
the calculation time can be shortened by improving the program. Also, it seems that
the values of crossover rate Cv and mutation rate Mt affect the convergence of the
solution. So, the convergence time can probably be shortened by adopting appropriate
values of Cv and Mt : Usually, Cv is set less than 10% and Mt is set smaller than Cv
though there is no standard of these values because a clear information-theoretical
explanation has not yet been given about GA. The values of Cv and Mt used in this
research are relatively small. Because of this, the evolution goes on slowly.
The calculation time for a single individual is roughly proportional to the
following parameters:
(ii) NðN þ 1Þ; where N is the total number of the fictitious charges, because
Gauss–Jordan elimination is used when the simultaneous equations are
solved.
7. Conclusions
In this paper, we show the possibility that GA can be used as a way to arrange
both the fictitious charges and the contour points in CSM to calculate the potential
distribution around the electrode system with 3-D structure. The accurate solution
will be obtained by the following improvements:
(1) Increasing the number of the charges. This is the easiest method to improve the
accuracy of the solutions.
(2) Assigning the appropriate numbers of the charges and the contour points to
each electrode that constructs the electrode system [2]. This method is effective
when the total number of the charges is limited by some reason such as the limit
of computer resources.
It is also effective to increase the number of the test points, which are used to
calculate the evaluation parameter s; to improve the accuracy of the solution.
However, the calculation time for a single generation will be roughly proportional to
the number of the test point when the number becomes larger. Because of this, the
method to calculate the parameter s by using a limited number of the test point
should also be investigated.
Because this study is an investigation whether GA can be applied to the
problem of the arrangements of fictitious charges and contour points, we do not
consider the terminating condition from the magnitude of the error. For real
problems, the terminating condition will be changed considering the magnitude of
the error, i.e., the calculation is repeated until the magnitude of error reaches
tolerance.
If iteration (generation) number is large, it may take more calculation time to
obtain accurate (practical) solutions by using GA than by using a try-and-error
method by a researcher who acquires experience in CSM calculations.
It is effective to increase the number of the fictitious charges to improve the
accuracy of the solution. However, as the charge number becomes larger in 3-D
calculations, it is more difficult to get the picture of the whole charge arrangement
correctly when a try-and-error method is used, especially for beginners of CSM.
Because of this, it is desirable to investigate automatic arrangement methods of the
fictitious charges and contour points.
If a CSM program already exists, this automatic arrangement method can be used
by appending a decision routine using GA, in which case, it is not necessary to
rewrite the CSM program. The computer resources required for this automatic
arrangement are the same for conventional CSM calculations, however, it is
suggested to use a higher-speed computer.
346 R. Nishimura et al. / Journal of Electrostatics 57 (2003) 337–346
References