IEEE TRANSACTIONS ON MAGNETICS, VOL. 26, NO.

2, MARCH 1990 925

2-D FINITE ELEMENT CALCULATION OF SHELL LOSSES IN POWER TRANS-
FORMERS, TAKING THE 3RD DIMENSION INTO ACCOUNT BY MEANS OF A
CORRECTIONFACTOR.
Triomphant NGNEGUEU, GBrard MEUNIER, Jean Claude SABONNADIERE
Laboratoired'Electrotechnique de Grenoble (U.A CNRS 355)
E.N.S.I.E.G, m a i m universitaireBP 6.38402 st Martin DHWS -~rance.
Mihel MAILHOT. Jean Pierre ARTHAUD.
Jeumont-Schneider, Division Transformateurs de Puissance
84. Avenue Paul Santy 68371 Lyon Cedex 08 -France.

Abstract
In a previous study, windings losses in shell As the steady state lines of flux plotting tends
form power transformers were investigated, using to show it (fig.2), the leakage flux not only induces
a code written for the purpose and based on some extra losses in the windings but would also induce
analytical methods [2]. In the present study, a losses in the tank's walls.
transfert of data is made to FLUXLAB-2D, a
software based on the finite element method. Then
shell losses are investigated and some alternative
solutions are compared. A correction factor is
introduced when the 2-D approximation may yield
poor results. For the purpose of this study a 13.333
MVA shell form power transformer, built for the
previous study is still being shielded. Once the test
results are available, they are to be compared with
calculation results.

Although the method of analysis proposed in this

paper is maint to be general, we focuse our
attention on the particular case of shell form
Fg.2 :p w
. -. 3
power transformers. Such transformers are those
in which the windings are surrounded be the
magnetic circuit; some diagrammatic sectional For fear of local heatings in the tank's wall, two
views of such a transformer are presented below. alternative means are currently used. The tank's
, walls are usualy covered with magnetic steel
strips. The purpose of these magnetic shunts is to
channel the leakage flux back to the meant
magnetic circuit. When precalculations give clues
that the leakage flux at the vinicity of a tank's
wall would not be important enough, this latter is
covered with nonmagnetic plates. Alumnium plates
are currently used. The purpose of these plates is
to expel1 the leakage flux by means of induced
currents.
In this paper, a 2-D finite element formulation
of induced losses in nonmagnetic plates and bare
magnetic walls is presented.
. *
onal fi-nts f o r m o n of induced
losses in-on
For this calculation, Maxwell's equations for the
quasi-stationnary effects are used.

CurlH =J (1)
Curl E -
= 6Bht (2)
DivB =O (3)
H = VB (4)
J = oE (5)
OO18-9464/90/03OO-0925$01 .OO 0 1990 IEEE

I
926
Where H is the magnetic field, E the electric
field, B the magnetic flux, J the current density, o
the electric conductivity and v the magnetic
reluctivity. Introducing a magnetic vector
potential A by : B o Curl A (6) P
Curl (v Curl A) = J (7)
From (2) and (6), Curl(E + 6A/6t) = 0. Hence, the
total current density at a point is obtained :
J = -o(6A/6t + Grad V) (8)
In the sinusoidal hypothesis of pulsation w this
becomes :
J = -iowA - oGrad V (9),
and by (7),the field equation to be solved is given Fig. 3:
by just to provide the supply's current densities. In
Curl (v Curl A) + iaoA + oGrad V = 0 (1O), this calculation, no eddy current is to be induced in
with i2 = -1. V is the electric potential due to the these subregions. Hence their electric conductivity
supply and the induced currents and it's important
is set at zero. Current densities are those used in
to notice that in 2-D the value of G r a d V is
the calculation of windings losses [2],[3]. The
constant on every subregion [I], [4].
magnetic permeability is that of the air.
After the finite element discretization
following the Galerkine procedure, equations (9)
and (10) are solved. For this solution the current
approach is the intregro-differential approach in These are the actual conducting subregions.
which the term oGrad V is substituted for in (9) They are made of the plates which are assigned a
and (lo), leading to a dense system's matrix and to constant magnetic permeability p which is that of
the coupling of all the nodes of every conducting the air and a constant electric conductivity o. The
subregion. Another approach is the simultaneous only currents are the eddy currents and the plates
approach in which both A and Grad V are are asumed to be independant, hence the total
considered as the unknowns. Though the number of current in each plate is zero. By (9), the current
lines of the system to be solved is higher than in density at a point is given by : J = -iowA - oGrad V
the previous apprach, the system's matrix is less Altough in the adopted simultaneous approach
dense and only neighbouring nodes are coupled. As of solution , the Grad V are part of the unkwons, it
will be the case in the examples we are to present, is more convenient to express Grad V in terms of
the simultaneous approach is particularly efficient the magnetic vector potential A. Expressing that
when the total currents are specified in the the total current in each of the subregions of the
subregions [SI. -
third type is zero, J3 (-iowA oGrad V) ds = 0 .
Both the integro-differential appraoch and the Bearing in mind that Grad V is constant on each
simultaneous appraoch are available in the finite of the subregions, Grad V is obtained on a given
element based software FLUXLAB-2D of the subregion of the grd type :
Laboratoire d'Electrotechnique de Grenoble. G r a d V = -io0 13 A ds / 13 ods . Hence the
FLUXLAB-2D .has been used for the solution of our
problems. current density.
J = -iowA + ioo J3 A ds / 13 ods (11).
ormet's window into subre- The induced losses are given on each of these
subregions by :
Three kinds of subregions are defined.
Po I 1/20. J3 I J I* ds (12)-
(11) and (12) have been implemented in the
FLUXLAB-2D software.
This is the air. It's characterized by a zero
conductivity and a current density equal to zero,
since there is no supply an no induced current.
The surrounding iron is assumed to be of infinite
s of the permeability, and not saturated. This leads to
These subregions are made of the coils. Neumann conditions on the outside rectangle.
Windings losses have already been investigated by
other means [2] and the task of the coils here is
 927

The subregion of the third type defined in the

last paragraph is replaced by the a magnetic wall
of a given thickness. The finite elements
formulation of the problem is the same as for the
calculation of the losses in the plates. Except for
boundary conditionq and the wall permeability
which do change.

Fig 5.2: p W 3 of a

The induced losses are also calculated and

compared for the two types of problems.
For the bare magnetic wall, the following B(H)
curve has been used, corresponding to a soft steel.
9 .
T' r
:5 ~ . S1.1

%
dGldH

Fig. 4: p

On the one hand, the same window of the

commercial shell form transformer mentioned in
figure (1) is considered and the two types of
problems are solved. The lines of flux are plotted
at 50 Hertz (fig 5.1 and fig 5.2). As expected, the
plate expells the leakage flux whereas in the case
of the naked wall, the leakage flux reaches the
wall in a 90" incidence since the wall permeability
remains far higher than that of the air. It's to be
noticed that figure (5.2) remains quite unchanged
compared to figure (2).

Baremaoneticwall A l W m plate
Cdculatedinducedl o s s s
perunitlength (Watts) 6266 1736.

I I
928
bear some errors,
L
< >

J I
Currents' li ne3 of flow pattern accordi ng
to e 2-D calculation
L
>

1-

In this paper, shell losses in shell-form power

transformers has been investigated, using the 2 - 0
finite element method. Lines of flux have been
plotted and alternative calculation results have
been compared. To take the 3 rd dimension into
account, a correction factor has been introduced.
For the purpose of this study, a 13.333 MVA shell
form power transformer is still being built and
once the test results are available, they are to be
compared with our calculations.

[l] GQardMEUNIER,Dazhong SHEN,Jean Louis COUIDMB.

"Modelisationof Zland axisymetric magnetodynamk domain by the finite
elements method." COMPLJMAG88.
[2] TriomphantNGNEGUEU, Grard MEUNIER, Jean-cbde SABONNADJERE
Michel MAILHOT, J ~ - P k i ARTHAUD.
t
"Calculationof extra losses in shell form power transformers windings"
INTERMAG 89.
[3] TriomphantNGNEGUEU,G6rardMEUNIER, Jean-ClaudeSABONNADJERE
Michel UAIIlIoT,J m - b ARTHAUD.
"Quelquesm&hc&s de calcul des pertes supplheneairesdans lesbobinaga
desaansformatwrdep~."
Under submission to LA REWE DE PHYSIQUE APPLIQUEE 89.
[4] [Link]
"The numerical solution of steady state skin effect problems. An -
integmdifferencial approach." IEEETrans Mag Vol Mag-17, 1 9 d .
[5] R. [Link], K. € NORSWORTHY.
I.
"Eddy currents and wall losses in screened-rotor induction motors"
IEE Paper No 2525 U, Apr. 1958.
Posibontothctmk'l ZDinducedlossa [watts) CotrededloJsesIwatts: [SJ Dazhongs".
bottom [mm) Po KP(580mtt1)~po "ContributionB la mod6lisation num6rique des phenomhx 6lectroma-
0. D4 .067 gnktiques" T h h de doctorat de l'[Link].G., ch 1.55, Page 25, Sept 1987.
[7] [Link].
490. 41. 79. Electrotechnique l'usage des inghieurs
I 650. I 1002. I 1674. I Tome 1.6lktriCit6 fondamentale,Page 319, Dunod 1%9.

T&2:& ..