Numerical techniques for design and

modeling of distribution transformers

DOI : 10.1016/j.jmatprotec.2004.07.044

Dear Reader,

You can cite this work as follows:

M. Tsili, A. Kladas, P. Georgilakis, A. Souflaris, D. Paparigas, “Numerical techniques for design

and modeling of distribution transformers,” Journal of Materials Processing Technology, Vol. 161,
No. 1−2, pp. 320−326, April 2005.

Thank You.
 Journal of Materials Processing Technology 161 (2005) 320–326

Numerical techniques for design and modeling

of distribution transformers
M. Tsili a , A. Kladas a,∗ , P. Georgilakis b , A. Souflaris b , D. Paparigas b
a Faculty of Electrical and Computer Engineering, Laboratory of Electrical Machines, Electric Power Division,
National Technical University of Athens, 9 Iroon Polytechneiou Street, 15780 Athens, Greece
b Schneider Electric A.E., Elvim Plant, 32011 Inofyta Viotia, Greece

Abstract

Power transformer analysis and design focusing on the equivalent circuit parameter evaluation by magnetic field numerical calculation
is presented. The proposed method adopts a particular reduced scalar potential formulation enabling 3D magnetostatic problem solution.
This method, necessitating no source field calculation, in conjunction with a mixed finite element – boundary element technique, results in
a very efficient 3D numerical model for power transformer design office use. Computed results are validated through measurements. Such a
methodology is very promising for investigation concerning losses and short circuit voltage variations with the main geometrical parameters.
© 2004 Elsevier B.V. All rights reserved.

Keywords: Distribution transformers; Methodology; Geometrical parameters

1. Introduction tion. The leakage inductance evaluation has been extensively

analyzed, as well as eddy current loss in transformer tank
Transformers are electric machines that enable the trans- walls, iron lamination characteristics and design considera-
mission and distribution of electric energy in a simple and tions. The systematic increase of computer efficiency along
cost-effective way, since their efficiency overcomes 95%. The with the evolution of numerical methods of magnetic field
modern industry requirements necessitate the construction of simulation enable the detailed transformer magnetic field
a great variety of transformers that do not fit into standardized analysis with the use of low cost and widely popular com-
large-scale constructions. In such cases, experiential ways of putational systems. The finite element method is one of the
electric characteristics calculation do not afford satisfying ac- numerical methods that have prevailed in the field analysis
curacy, as they concern particular geometries. Moreover, the of three-dimensional configurations that comprise materials
limited delivery time does not allow the experimental verifi- with non-linear characteristics, like transformers, and may
cation of the predicted transformer characteristics. be applied within reasonable time in an appropriate personal
Numerical modeling techniques are now-a-days well es- computer [9,10].
tablished for power transformer analysis and enable repre- On the other hand, the boundary element method is a nu-
sentation of all important features of these devices [1,2]. merical field analysis technique that uses the integral form of
More particularly, techniques based on finite elements present magnetic field equations and discretizes only the boundaries
interesting advantages for non-linear characteristics simula- of the considered areas (in comparison to the finite element
method which discretizes the whole field). Therefore, this
∗ Corresponding author. Fax: +30 210 772 3593. method is suitable for open-boundary problems as well as
E-mail addresses: [email protected] (M. Tsili), geometries with extensive parts of air. Moreover, the combi-
[email protected] (A. Kladas),
pavlos [email protected] (P. Georgilakis),
nation of boundary and finite elements is widely used for elec-
thanassis [email protected] (A. Souflaris), tromagnetic problems since the electromagnetic field is not
dimitris [email protected] (D. Paparigas). only confined to the conductors but it expands over extensive

0924-0136/$ – see front matter © 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.jmatprotec.2004.07.044
 M. Tsili et al. / Journal of Materials Processing Technology 161 (2005) 320–326 321

parts of air, where the use of a boundary element representa-

tion can significantly decrease the computational effort [5,6].
The application of these methods to the transformer mag-
netic field simulation can afford accurate equivalent circuit
parameters evaluation as well as contribute to the prediction
of the transformer operating characteristics.
The upgrade of the distribution networks voltage (from
15 to 20 kV) has created the need for construction of trans-
formers suitable for both voltage levels by multiple high
voltage windings connection. In these cases, a difficulty
of the transformer parameters evaluation through the ex-
isting design methodology arises, which can be overcome
by incorporating the magnetic field analysis techniques to
the accustomed approximating methods used by manufac- Fig. 1. Active part configuration of the three-phase shell type distribution
turers. In addition, the accurate prediction of the trans- transformer considered.
former characteristics can result to its cost reduction, since
the short-circuit voltage value is critical for the choice
of dimensions that ensure its durability under short-circuit the conductor, that satisfies Ampere’s law and is perpendic-
conditions. ular on the subdomain boundary.
Fig. 1 shows active part the of the shell type distribution
transformer considered. Fig. 2 illustrates the perspective view
of the one-phase transformer part modeled, comprising the
2. Modeling techniques
iron core, low and high voltage windings.
The use of this one-phase model instead of the whole three-
2.1. Finite element method
phase transformer model was conducted for the following
reasons:
In the present paper a particular scalar potential formu-
lation has been developed, enabling the 3D magnetostatic
(i) The smaller model size enables the construction of more
field analysis. According to our method the magnetic field
dense tetrahedral finite element mesh without great com-
strength H is conveniently partitioned to a rotational and an
putational cost (given that the exact representation of the
irrotational part as follows [3]:
transformer magnetic field requires great accuracy which
H = K − ∇Φ (1) is dependent on the mesh density and the total execution
time of the finite element calculations).
where Φ is a scalar potential extended all over the solution (ii) The representation of one-phase of the active part does
domain while K is a vector quantity (fictitious field distribu- not affect the accuracy of the equivalent circuit parame-
tion), defined in a simply connected subdomain comprising ters calculation.

Fig. 2. Perspective view of the one-phase transformer part modeled.

322 M. Tsili et al. / Journal of Materials Processing Technology 161 (2005) 320–326

required, in the BEM formulation a mesh of the boundary

only is required, resulting to the significant reduction of the
problem size.
As the solution domain, in case of transformer simulation,
comprises extensive parts of air, the adoption of a boundary
element technique to represent the respective subdomains,
improves the model performance.

2.3. Mixed FEM–BEM method

Let us consider a coupled finite element/boundary element

Fig. 3. Fictitious field distribution corresponding to low voltage winding.
solution domain, comprising m FE nodes, n BE nodes and r
common nodes in the interface boundary. The matrices for
Due to the symmetries of the problem, the solution do- the BE region can be written as follows:
main was reduced to one-fourth of the device (although there
is a slight disymmetry due to the terminal connections in HU = GQ (3)
one side). These symmetries were taken into account by the
where U is a vector of the BE nodal potential values, Q a
imposition of Dirichlet boundary condition (Φ = 0) along xy-
vector of the BE nodal ∂Φ(s′ )/∂n values, while Hij and Gij
plane and Neumann boundary condition (∂Φ/∂n = 0) along
correspond to the Eq. (2) integrals Γj (∂Gi (s′ , s)/∂n)ds′ and
the other three faces of the air box that surrounds the trans-  ′ ′
former active part. Γj Gi (s , s)ds , respectively [6]. The matrices of the FE re-
In such a case the distribution of K is straightforward. gion can be written as SU = F, where S is the stiffness matrix
As an example such a distribution corresponding to the low and F the source vector. Therefore, the global system matrix
voltage winding is shown in Fig. 3. Such a formulation is has the form
well suited for finite element discretization [1,4].
As shown in Fig. 2, the high voltage winding area is di-
vided into four subcoils. This division was made in order to
take into consideration the winding connection which gives
the second high voltage level (i.e. 15 kV). In particular, in the
case of the first connection (20 kV) all the subcoils are con-
sidered to undergo the nominal current, while in the second
one, two of them are connected in parallel, therefore, half of
the nominal current flows through them. Each subcoil con-
sists of the respective number of turns which are given by the
manufacturer.

where Tij is the term used to link the finite element region
2.2. Boundary element method
to the boundary element region (involving the potential and
normal derivative values of the FE–BE interface boundary
The boundary element method is derived through dis-
nodes), [7,8].
cretization of an integral equation, that is, mathematically
equivalent to the original partial differential equation. The
boundary integral equation corresponding to Laplace equa-
tion is of the form: 3. Results and discussion

∂G(s′ , s) ∂Φ(s′ )

 
′ The proposed reduced scalar potential formulation has
c(s)Φ(s) + Φ(s) − G(s , s) ds′ = 0
∂n ∂n′ been applied in the 3D numerical analysis of a transformer
Γ under short circuit for its leakage reactance calculation. Two
(2)
study cases were considered, and the respective results of the
where s is the observation point, s′ the boundary Γ coordinate, finite element analysis were compared to the experimental
n′ the unit normal and G the fundamental solution of Laplace results (local field values and short-circuit voltage value).
equation in free space.
The re-formulation of the PDE that underlies the BEM 3.1. Study Case 1
consists of an integral equation that is defined on the bound-
ary of the domain and an integral that relates the boundary The case of the one-phase part of a 1000 kVA, dual voltage
solution to the solution at the points in the domain. Therefore, 20–15 kV/400 V three-phase shell type power transformer,
while in the finite element method an entire domain mesh is has been considered.
 M. Tsili et al. / Journal of Materials Processing Technology 161 (2005) 320–326 323

Fig. 6. Comparison of measured and computed field values along the line
AB.

the variation of the perpendicular flux density component Bn

along the line AB, positioned as shown in Fig. 5, in case
Fig. 4. Three-dimensional finite element mesh of the transformer active part of short-circuit with the high voltage winding connections
(89,603 nodes). corresponding to 20 kV voltage supply. This figure illustrates
the good correlation of the simulated results with the local
Fig. 4 illustrates the finite element tetrahedral mesh leakage field measurements.
adopted for the calculation of the transformer magnetic field Table 1 compares the measured short-circuit voltage value
under short-circuit conditions. The mesh consists of 89,603 deduced by the short-circuit test and the calculated one with
nodes, a density which is considered satisfactory (although the use of different mesh densities, in order to evaluate the
not optimum) for a three-dimensional problem. As shown in respective variation of the error. The error of the short-circuit
this figure, it is significantly more dense in the windings area, voltage calculation reduces significantly as the mesh density
while it is sparser in the iron cores. This configuration was increases, while it appears to be similar for the two high
chosen in order to represent in great detail the magnetic field voltage levels.
sources (i.e. the windings) without extreme increase of the
total number of nodes (by reducing the respective number of 3.2. Study Case 2
nodes in the cores area).
Fig. 5 shows the magnetic flux density magnitude distri- The case of the one-phase part of a 630 kVA, dual voltage
bution during short-circuit test, as it was calculated by the 20–15 kV/400 V three-phase shell type power transformer,
proposed 3D finite element method with the use of the above was also considered. The three-dimensional finite element
mesh. model that was used is similar to the one used for the first
These field values have been compared to those measured study case, consisting of a 100,999 nodes tetrahedral mesh.
by a Hall effect probe during short-circuit test. Fig. 6 gives The computed values of the perpendicular flux density
component Bn have been compared to those measured by a
Hall effect probe during short-circuit test along the lines AB
and CD of Fig. 7. Figs. 8 and 9 show the variation of Bn along
the line AB with the high voltage winding connections corre-
sponding to 20 and 15 kV voltage supply, respectively, while
Figs. 10 and 11 give the variation of Bn along the line CD. In

Table 1
Comparison of calculated and measured short-circuit voltage values for the
1000 kVA transformer modeled
Number of Calculated Measured Error High voltage
nodes short-circuit short-circuit (%) level (kV)
voltage voltage
14491 6.38 6.13 4.00 20
49047 6.30 2.84
86903 6.19 0.98
14491 6.13 5.95 3.01 15
49047 6.09 2.30
Fig. 5. Magnetic flux density magnitude distribution during short-circuit
86903 5.97 0.34
test.
324 M. Tsili et al. / Journal of Materials Processing Technology 161 (2005) 320–326

Fig. 7. One-phase 3D model of the 630 kVA distribution transformer.

Fig. 8. Comparison of measured and computed field values along the line AB (short-circuit at 20 kV).

Fig. 9. Comparison of measured and computed field values along the line AB (short-circuit at 15 kV).
 M. Tsili et al. / Journal of Materials Processing Technology 161 (2005) 320–326 325

Fig. 10. Comparison of measured and computed field values along the line CD (short-circuit at 20 kV).

Fig. 11. Comparison of measured and computed field values along the line CD (short-circuit at 15 kV).

the case of AB line, the computed and measured field values the use of different mesh densities, in order to evaluate the
at 20 kV are very close, with the exception of two points in respective variation of the error. The variation of the er-
the “phase c” curve (due to errors during the measuring pro- ror is similar to that of Table 1, although in the case of
cess). The proximity is also good for the case of 15 kV, where the 630 kVA transformer the error is relatively greater for
we can observe the field variation due to the different current medium mesh densities (>4% for a 23,696 nodes mesh)
level going through the two high voltage subcoils connected and reaches a satisfactory low value for a mesh of 100,000
in parallel. The same conclusions are drawn from the obser- nodes.
vation of the curves corresponding to the line CD for the case
of short-circuit at 20 and 15 kV.
Table 2 compares the measured short-circuit voltage value Acknowledgement
deduced by the short-circuit test and the calculated one with
The authors express their gratitude to the General Secre-
tariat for Research and Technology of Greece for supporting
Table 2
this work under Grant PAVET 00BE457.
Comparison of calculated and measured short-circuit voltage values for the
630 kVA transformer modeled
Number of Calculated Measured Error High voltage
nodes short-circuit short-circuit (%) level (kV) References
voltage voltage
[1] P. Georgilakis, N. Hatziargyriou, D. Paparigas, AI helps reduce trans-
23696 5.95 5.61 5.97 20
former iron losses, IEEE Comput. Appl. Power 12 (4) (1999) 41–46.
47044 5.73 2.07
[2] P.S. Georgilakis, N.D. Doulamis, A.D. Doulamis, N.D. Hatziar-
100999 5.67 1.14
gyriou, S.D. Kollias, A novel iron loss reduction technique for dis-
23696 5.75 5.50 4.55 15
tribution transformers based on a Combined Genetic Algorithm –
47044 5.62 2.13
Neural Network Approach, IEEE Trans. Systems Man Cybernetics
100999 5.57 1.18
C: Appl. Rev. 31 (1) (2001) 16–34.
326 M. Tsili et al. / Journal of Materials Processing Technology 161 (2005) 320–326

[3] A.G. Kladas, M.P. Papadopoulos, J.A. Tegopoulos, Leakage Flux A. Kladas was born in Greece, in 1959. He received the Diploma in Elec-
and Force Calculation on Power Transformer Windings under short- trical Engineering from the Aristotle University of Thessaloniki, Greece
circuit: 2D and 3D models based on the Theory of Images and in 1982 and the DEA and PhD degrees in 1983 and 1987, respectively,
the Finite Element Method compared to measurements, IEEE Trans. from the University of Pierre and Marie Curie (Paris 6), France. He
Magn. 30 (5/2) (1994) 3487–3490. served as Associate Assistant in the University of Pierre and Marie Curie
[4] Z.X. Feng, The treatment of singularities in calculation of magnetic from 1984 to 1989. During the period 1991 to 1996 he joined the Public
field by using integral method, IEEE Trans. Magn. 21 (6) (1985) Power Corporation of Greece, where he was engaged in the System Stud-
2207–2210. ies Department. Since 1996 he joined the Department of Electrical and
[5] G. Meunier, J.L. Coulomb, S.J. Salon, L. Krahenbul, Hybrid Fi- Computer Engineering of the National Technical University of Athens
nite Element Boundary Element Solutions for three-dimensional (NTUA), where he is now Associate Professor. His research interests in-
scalar potential problems, IEEE Trans. Magn. 22 (5) (1986) 1040– clude transformer and electric machine modeling and design as well as
1042. analysis of generating units by renewable energy sources and industrial
[6] S.J. Salon, J. D’Angelo, Applications of the hybrid finite element – drives.
boundary element method in electromagnetics, IEEE Trans. Magn.
P. Georgilakis (S’1998, M’2001) was born in Chania, Greece in 1967.
24 (1) (1988) 80–85.
He received the Diploma in Electrical and Computer Engineering and the
[7] A.J. Moses, Comparison of transformer loss prediction from com-
PhD degree from the National Technical University of Athens, Greece in
puted and measured flux density distribution, IEEE Trans. Magn. 34
1990 and 2000, respectively. In 1994 he joined Schneider Electric AE,
(4) (1998) 1186–1188.
Greece. He has worked in the Development and also the Quality Control
[8] C. Lin, C. Xiang, Z. Yanlu, C. Zhingwang, Z. Guoqiang, Z.
Departments of the Industrial Division of the company. From 1999 to
Yinhan, Losses calculation in transformer tie plate using the fi-
2001 he was the R&D Manager of Schneider Electric AE. At present
nite element method, IEEE Trans. Magn. 34 (5) (1998) 3644–
he is Business and Activity Manager in the Marketing Division. He is
3647.
member of IEEE, CIGRE, and the Technical Chamber of Greece.
[9] I.L. Nahas, B. Szabados, R.D. Findlay, M. Poloujadoff, S. Lee, P.
Burke, D. Perco, Three dimensional flux calculation on a three- A. Souflaris was born in Athens, Greece in 1956. He received the
phase transformer, IEEE Trans. Power Deliver. 1 (3) (1986) 156– Diploma in Electrical Engineering from the Technical University of
160. Pireaus, Greece in 1981. He joined Schneider Electric AE in 1985 as
[10] K. Zakrzewski, B. Tomczuk, Magnetic field analysis and leak- Transformer Design Engineer and from 1988 he is the Transformer De-
age inductance calculation in current transformer by means of sign Manager of Schneider Electric AE.
3D integral method, IEEE Trans. Magn. 32 (3) (1996) 1637–
1640. D. Paparigas was born in Komotini, Greece in 1945. He received the
Diploma in Electric Machines and Instruments from the Moscow Energy
Institute, USSR in 1972. From 1974 to 1976 he worked in the Engineering
M. Tsili was born in Greece, in 1976. She received the Diploma in Elec- Department of Masina Xrisolouris, Greece. Since 1976, he has been with
trical and Computer Engineering from the National Technical University Schneider Electric AE (former Elvim), Greece. He has been engaged in
of Athens in 2001 where she follows post-graduate studies. Her research several positions, such as Quality Control Manager, Production Manager,
interests include transformer and electric machine modeling as well as Transformer Design Manager and Quality Assurance Director. At present,
analysis of generating units by renewable energy sources. he is Technical Division Director and Industrial Division Director.