IEEE TRANSACTIONS ON MAGNETICS, VOL. 48, NO.

2, FEBRUARY 2012 351

Comparison Study of Finite Element Methods to Deal With Floating

Conductors in Electric Field
W. N. Fu1 , S. L. Ho1 , Shuangxia Niu1 , and Jianguo Zhu2
Department of Electrical Engineering,Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong
School of Electrical, Mechanical and Mechatronic Systems, Faculty of Engineering and Information Technology,
University of Technology, Sydney (UTS), NSW 2007, Australia

In transient magnetic field computation, it is highly desirable if the stray capacitances among all conductors can be computed and their
effects addressed conveniently. Because of the existence of floating conductors, the general finite element (FE) solver of Poisson’s equa-
tion, which is a commonly used magnetic field solver, cannot be used to extract the capacitance matrix directly. In this paper, methods to
deal with floating conductors in the electrostatic field for field solution, and capacitance matrix extraction using finite element method
(FEM) are compared; their merits and shortcomings are discussed. A method to compute the electric field and extract the capacitance
matrix to include the effect of floating conductors inside the solution domain is put forward. The merit of the proposed algorithm is that
the general FE solver can be used without the need for any special program modification. A FE formulation to automatically include the
capacitances in transient magnetic field is presented. By using the proposed method, users only need to input the information as how the
conductors are connected. All the stray capacitances among conductors are automatically included in the solution of transient magnetic
field with the proposed algorithm.
Index Terms—Capacitance, circuit parameter, electric field, finite-element method, floating conductor, magnetic field.

I. INTRODUCTION is that for different charge excitations on the conductors, only the
right-hand side (RHS) of the system equation changes, whereas
T O SIMULATE power electronic electric drives, it is im-
portant to address high-frequency effects in the solutions
because of high speed switching and repetitive steep rising and
the coefficient matrix remains unchanged. Hence a multi-RHS
solver can be used conveniently to reduce the computing time.
The shortcoming of the charge-excitation method is that it
falling edges in voltage waveforms [1]–[3]. In addition, the
requires complicated internal modifications to existing FE pro-
solving of full-wave electromagnetic fields that include stray
grams. The proposed method is to compute the complete ca-
capacitance is very time consuming. A practical and promising
pacitance matrix first. For the computation of the capacitances
method is to couple the stray capacitances among conductors
among floating conductors and other conductors, each floating
with the transient magnetic field using field—circuit coupled
conductor will have its own voltage excitation or zero-value
method [4]. It is highly desirable if the general finite element
Dirichlet boundary condition. The complete capacitance matrix
(FE) solver of Poisson’s equation, which is commonly used by
can be obtained without dealing with the floating conductors.
the magnetic field solver, can also be used to extract the capac-
The entries of the floating conductors are then removed and a
itance matrix. However, because of the existence of floating
reduced-size capacitance matrix is then automatically obtained.
conductors which represent equipotential volumes but which
Lastly, from the charge-potential relationship, the potentials on
do not constitute the Dirichlet boundary conditions, a special
the floating conductors can be obtained and the field distribution
numerical constraint needs to be implemented. Reference
of the nominal problem can also be computed using the voltage
[5] presents the constraint equations for floating conductors
excitation method again. This method does not require any mod-
in finite-difference methods for determining the capacitance
ification of normal FE programs, and multi-RHS solver can also
relationships among conductors. The method used in [6]–[8] is
be used to reduce the computing time. For transient magnetic
to regard the floating conductors as dielectric materials having
field computation, the capacitances obtained are automatically
infinite permittivity. In program implementation, infinity is
included in the formulation. Users only need to input the data
usually realized using a very large number, and numerical
of the connection of the conductors into the software; the stray
errors from the aforementioned formulation become inevitable.
capacitance effect is automatically addressed.
In this paper a simple and direct method to deal with floating
conductors inside the solution domain for the electro-static field II. CHARGE-EXCITATION METHOD
finite element method (FEM) using electric charge excitations
is discussed. The equivalent capacitance coefficient matrix A. Computation of Field
including the effect of floating conductors is subsequently For a two-dimensional (2-D) problem on the - plane, dis-
extracted. The merits of such approach are that the floating cretizing the Poisson’s equation using Galerkin method gives
conductors are automatically taken as equipotential objects even
though they are in the solution domain; the effect of floating r 1 r V ) dx d y =
"( W W  dx dy + "W
@V
@n
dc (1)
conductors, which are not the circuit ports, can be taken into ac-
where is the electric potential; is the material permittivity;
count in the extracted lumped capacitances. Another advantage
is the volume charge density; is the weighting function;
Manuscript received July 04, 2011; revised October 12, 2011; accepted Oc- and edge is the boundary. For a conductor with a total electric
tober 21, 2011. Date of current version January 25, 2012. Corresponding author: charge as shown in Fig. 1, the conductor region is denoted as
W. N. Fu (e-mail: eewnfu@polyu.edu.hk).
Color versions of one or more of the figures in this paper are available online
and is bounded by edge . The last term in (1) is:
at http://ieeexplore.ieee.org. @V
Digital Object Identifier 10.1109/TMAG.2011.2173917 "W dc = W  dc (2)
@n

0018-9464/$31.00 © 2012 IEEE

352
Fig. 1. A conductor with electric charge Q in the domain.
Fig. 2. The capacitances among N conductors and ground.
where is the surface charge density. Line integration of (2)
will only be carried out on the edge of the conductor.
Conductors in static or quasistatic fields can be considered as
perfect electric conductors. The body of a perfect electric con-
ductor is an equipotential volume. If the equipotential value is
known, a Dirichlet boundary condition can be applied. If the
equipotential value is unknown, that is, the conductor is not con-
nected to a voltage source, the floating potential is a variable.
The nodal potential values associated with the conductor are all
the same but the actual value of the potential is part of the so-
lution. When it comes to program implementation, all potential
variables on the conductor are taken as one variable. The term in
the right-hand side of (2) contributes to one entry in an element
assembly, which is located at the same row and column in the co-
efficient matrix of the system equation which is associated with
the DoF (degree of freedom) of the potential on the conductor.
A constraint that enforces the total charge, which is equal to ,
is applied on the conductor. Because (2) is only contributing to
the same entry in the coefficient matrix, the constraint can be
implemented readily. The floating conductor is a special case of
charge excitations on conductors in which is equal to zero.
B. Computation of Capacitance Matrix
Here the charge-excitation method is reviewed to compute the
capacitance matrix. Because of the relationship of ,
for the system having conductors and one ground (all con-
ductors connected to a ground can be considered collectively
as one single ground conductor) as shown in Fig. 2, in which
represent the potentials of the conductors, the
net charge on each conductor is:
.
.
.
.
.
.
.
.
(3)
.
IEEE TRANSACTIONS ON MAGNETICS, VOL. 48, NO. 2, FEBRUARY 2012
By defining an elastance matrix, which is denoted as the
matrix:
(4)
one has:
. .
. .
. .
. (5)
.
.
If the charge excitations are set as:
(6)
then:
.
.
.
.
(7)
. .
After the field is solved, , and are known. One
column of the matrix is obtained:
(8)
By using similar methods, other columns of can be
obtained.
After the matrix is obtained, the matrix can be derived
according to (4). With this algorithm, the coefficient matrix of
the FE equations can be kept unchanged. Only a multi-RHS
problem needs to be solved. By using multi-RHS algebraic
equation solvers, the computing time required to extract the
capacitance matrix can be greatly reduced.
III. VOLTAGE-EXCITATION METHOD
In voltage-excitation method, the Poisson’s equation on all
conductors with only Dirichlet boundary conditions needs to be
solved. The interiors of conductors are excluded from the solu-
tion domain because the potentials are constant throughout the
volumes of the conductors. If there are conductor charge exci-
tations or floating conductors in the nominal problem, the com-
plete capacitance matrix, including the entries of floating con-
ductors, is computed first with the proposed algorithm. Each ca-
pacitance can be computed according to the energy of the elec-
tric field for each specified voltage excitations.
If there are floating conductors, the total charge on the con-
ductor is zero and a matrix reduction method can be used. The
row and column corresponding to that conductor in the capac-
itance matrix can be eliminated without solving the FE equa-
tions. Without loss of generality, a 3-conductor system is taken
as an example. The - relationship is
(9)
FU et al.: COMPARISON STUDY OF FINITE ELEMENT METHODS TO DEAL WITH FLOATING CONDUCTORS IN ELECTRIC FIELD 353

Supposing conductor 3 is a floating conductor, = 0. can

be expressed by and , and (9) is then simplified as:
Q = C 0C 0C 0 C V : (10)
Q 0C 0 C C 0C V
For the computation of nominal field, if the original problem
has conductor charge excitations or has floating conductors, ac- Fig. 3. The plot of equipotential lines when V = 1 V, V = 01 V and the
cording to the - relationship, the potentials on the conduc- conductors 2 and 3 are floated.
tors can be obtained. The voltage excitations are then applied to
all conductors and the field solution of the nominal problem can Adding the capacitance circuit equation (16) into (14), one
be computed. obtains:
IV. INCLUSION OF CAPACITANCES IN TRANSIENT
MAGNETIC FIELD
For transient magnetic field computation, all capacitances
among the conductors can be included in the system equations
by virtue of internal circuit coupling. When the capacitances (17)
among the conductors have to be considered, it is obvious that
the frequency of the excitations is high. Therefore, all wires are The branch voltage column matrix and the branch current
considered as solid conductors to include eddy-current effect. column matrix are:
The basic 2-D magnetic field equation in the - regions of the (18)
solid conductors is [4]: (19)

r 1 (rA) 0  @A d 
@t + l V = 0 (11) Using the nodal method, the relationship between the branch
voltage and the nodal voltage is:
where is the axial component of magnetic vector potential;
is the reluctivity of material; is the conductivity; is the fV g = A fV g (20)
polarity ( or ) to represent the forward path or return path,
where is the node-to-branch incidence matrix, which can be
respectively; is the model depth in the -axis; is the voltage
determined easily according to the connection of the conductors.
difference between the two terminals of the conductor. The cir-
The Kirchhoff’s current law can be expressed as
cuit branch equation of the conductor is
d  @A d
0 1 V = 0 1 I (12)
(21)
l @t lR l Substituting these relationships into the system equations, the
where is the total d.c. resistance of the conductor; is final global equations are obtained as follows:
the current in the solid conductor. After applying a standard S + S A
A S0 00 1tS 0
Galerkin procedure, the coupled field and circuit equations
in the magnetic field region can be written in a block matrix A 0 0 C A
P + A
2 VA = A
format:
S S A D 0 S A
= 0 0I + P0
(22)
0 S u + S 0 (13) 0 CV
where the coefficient matrix is symmetrical. Here users only
where is the column matrix associated with the excitations. need to input the incidence matrix to the program, as all
Using the backward Euler’s method to discretize the time vari- other procedures are realized by the software automatically.
able and multiplying the time step size to the circuit branch
equation, one obtains the recurrence matrix formula at the th V. TEST CASES
step for the time stepping process as follows: A. Capacitance Extraction of 2-D Field
S + S A = 0 0I + P S+ A A
A 4-conductor system above a ground plane as shown in
S 1tS V Fig. 3, where an analytical solution is available, is used as a
test case [9]. The diameter of each conductor is 10 mm. The
(14)
where the superscript means the values at the th step of the distance between two adjacent conductors is 100 mm. The
time stepping computation. distance between the center of the conductors and the ground is
According to (3), the branch matrix equation of the capaci- 150 mm. If V and V, and conductors 2 and 3
tance circuits can be discretized as: are floated, a plot of the computed equipotential lines is shown
in Fig. 3.
(15) The capacitance matrix computed by using analytical method
is:
14:840556 03:8479870 00:93025803 00:44076465
Multiplying to the two sides of (15), one has the 03:8479870
C = 00:93025803 15:825204 03:6344104 00:93025803 (pF):
branch equation of the capacitance circuits as follows: 03:6344104 15:825204 03:8479870
0 1tl C V = 0 1tl I + 0 1tl C V : (16)
00:44076465 00:93025803 03:8479870 14:840556
(23)
354 IEEE TRANSACTIONS ON MAGNETICS, VOL. 48, NO. 2, FEBRUARY 2012

conductors, special constraints are necessary. The drawback is

that special formulation is needed in the FE package to deal with
the floating conductors.
2) The voltage-excitation method just needs to solve stan-
dard Poisson’s equation with only Dirichlet boundaries on all
Fig. 4. (a) Two cylinder conductors represent the two feet of an electronic com- the conductors. Therefore general FE solver, which is originally
ponent. (b) Its FEM mesh. used by magnetic field solver, can also be used to extract the
capacitance matrix. It also needs to simply solve the set of al-
gebraic matrix equations with the same coefficient matrix. Re-
duction matrix method can be used to avoid complications in
dealing with floating conductors. After the potentials on all con-
ductors are available from the potential—charge relationship,
the field distribution of the original nominal problem can then
Fig. 5. (a) Two cylinder conductors with a shielding conductor behind them. be computed. The merit of this method is that it requires no mod-
(b) Its FEM mesh. ifications to existing FE programs.
The capacitance matrix computed by using the voltage-exci- 3) The voltage excitation method and charge excitation
tation FEM is: method are interchangeable to each other. Each one can solve
14:851 03:8486 00:93902 00:44385 its field distribution and extract its capacitance matrix, no
03 8486
C = 00::93902 15:842 03:6324 00:93902 (pF): (24)
matter whether there are charge excitations on the conductors
03:6324 15:842 03:8487 or there are floating conductors. The charge-excitation method
00:44385 00:93902 03:8487 14:851 is a direct method whereas voltage-excitation method is an
indirect method.
If conductors 2 and 3 are floated, the capacitance matrix com- 4) After the capacitance matrix is extracted, it can be automat-
puted by applying the matrix reduction method to (24) is:
ically included into the transient magnetic field by an internal
13:69499 01:16517 (pF):
C = 01 (25) circuit coupling and its effect can then be correctly included in
:16517 13:69493 the solutions.
ACKNOWLEDGMENT
The capacitance matrix computed by using the charge-exci-
tation method of the floating conductors is: This work was supported by the Research Grant Council of
13:695 01:1652 (pF):
C = 01 (26)
the Hong Kong SAR Government under projects PolyU 5176/
:1652 13:695 09E, 5184/09E and 87RX.
REFERENCES
Comparing the results in (23) and (24), also in (25) and (26), [1] L. Gubbala, A. von Jouanne, P. N. Enjeti, C. Singh, and H. A. Toliyat,
it can be seen that both proposed methods are giving precise “Voltage distribution in the windings of an ac motor subjected to high
solutions. dv/dt PWM voltages,” in IEEE PESC Conf. Proc., 1995, pp. 579–585.
[2] D. B. Hyypio, “Simulation of cable and winding response to steep-
B. Capacitance Extraction of 3-D Field fronted voltage waves,” in IEEE IAS Conf. Rec., 1995, pp. 800–806.
[3] Y. Tang, “Analysis of steep-fronted voltage distribution and turn insu-
Two identical cylinders with radius 1 mm and limited height lation failure in inverter fed ac motor,” in IEEE IAS Conf. Rec., 1997,
5 mm are separated by 2 mm as shown in Fig. 4, which may rep- pp. 509–516.
resent two feet of an electronic component. The computed ca- [4] W. N. Fu, P. Zhou, D. Lin, S. Stanton, and Z. J. Cendes, “Modeling of
pacitance between the two cylinders is 0.16108 pF by using the solid conductors in two-dimensional transient finite-element analysis
and its application to electric machines,” IEEE Trans. Magn., vol. 40,
voltage excitation method and is 0.16136 pF by using the charge no. 2, pp. 426–434, Mar. 2004.
excitation method. If a shielding conductor is placed at the dis- [5] C. N. Dorny and R. A. Mathias, “Finite-difference methods for de-
tance of 1 mm from the two cylinders as shown in Fig. 5, the termining capacitance relationships among arbitrarily shaped conduc-
tors,” IEEE Trans. Power App. Syst., vol. PAS-90, no. 2, pp. 876–881,
computed capacitance between the two cylinders is 0.187140 Mar. 1971.
pF by using the voltage excitation method; it is 0.187625 pF by [6] A. Konrad and M. Graovac, “The floating potential approach to the
using the charge excitation method. The shielding conductor is characterization of capacitive effects in high-speed interconnects,”
dealt as a floating object in the 3-D FEM. It is reasonable for IEEE Trans. Magn., vol. 33, no. 2, pp. 1185–1188, Mar. 1997.
[7] A. Konrad and M. Graovac, “The finite element modeling of conduc-
the capacitance to increase if the shielding conductor is added tors and floating potentials,” IEEE Trans. Magn., vol. 32, no. 5, pp.
nearby the two cylinders. 4329–4331, Sep. 1996.
[8] G. J. Cokkinides, B. Beker, and A. Templeton, “Direct computation of
capacitance in integrated passive components containing floating con-
VI. CONCLUSION ductors,” IEEE Trans. Compon., Packag., Manuf. Technol., vol. 20, no.
2, pt. part B, pp. 123–128, May 1997.
1) For capacitance matrix extraction, the charge-excitation [9] P. Silvester, Modern Electromagnetzc Fzelds. Englewood Cliffs,
method only needs to solve a multi-RHS problem. For floating N.J.: Prentice-Hall, 1968, ch. 2.