You are on page 1of 12

JOURNAL OF APPLIED

COMPUTER SCIENCE
Vol. 16. No 1 (2008), pp. 89-100

3D Finite Element Estimation of Stray Losses


in Three-Phase Transformers

M.A. Venegas Vega1, R. Escarela Pérez2, T. Niewierowicz3


1
Arteche Transformadores y Tecnología
Ant. Carr. México-Querétaro km 73.5 Tepeji, Hidalgo, México
(e-mail: avenegas@arteche.com.mx)
2
Universidad Autonoma Metropolitana- Azcapotzalco,
Departamento de Energia
Av. San Pablo No. 180, Col. Reynosa, C.P. 02200, México, D.F.
(e-mail: r.escarela@ieee.org)
3
Instituto Politécnico Nacional, SEPI, ESIME-Zacatenco
C.P. 07738, México, D.F. (e-mail: tniewi@ipn.mx)

Abstract. Stray losses and flux leakages of a three-phase three-limb


transformer are determined and analyzed using 3D finite element analysis.
The problem is not current driven but voltage fed. Thus, a circuit-field
problem is established and solved. The operating condition analyzed is the
load-loss test. Since low excitation voltages are excited during this test,
the transformer works under unsaturated conditions. As a result, the
transformer is modelled using a time-harmonic approach. The transformer
geometry is truly three dimensional, forbidding the use of conventional 2D
models. The clamp plates and transformer tank are conveniently modelled
with surface impedance boundary conditions.

1. Introduction
Prediction of electromagnetic phenomena in structural metallic parts of
power transformers is an important step for their correct design, where optimal
reduction of local overheating due to leakage magnetic fluxes is sought.
Numerical determination of losses in massive conductors has been treated by
several authors [1]-[4], where the losses have been determined through the use
of 2D and 3D finite element approaches. For instance, references [1] and [2]
tackled the problem of losses in the regions near the high-current bushings of
transformers. However, the problem has been solved by withdrawing specific
portions from the whole geometry.
90 M.A. Venegas Vega, R. Escarela Pérez, T. Niewierowicz

The purpose of this work is to determine and analyse the losses produced by
leakage fluxes in a three-phase three-limb transformer, using a 3D time-
harmonic finite-element formulation of the full geometry. The operating
condition considered here is the three-phase load-loss test which injects
impedance voltage in one side of the transformer windings while keeping short-
circuited the other side. Since the skin depth of the transformer massive
conductors is much smaller than any of their defining dimensions, surface
impedance boundary conditions can be used to avoid explicit meshing of these
volume regions. This not only results in requiring smaller computational
resources but also allows the use of magnetic scalar potentials at all volume
regions, further reducing computational costs.

2. Model of the transformer

Fig. 1 shows the geometry of a three-phase three-limb distribution


transformer. The tank is not shown but can be visualized as a box enclosing
the transformer geometry. The nominal and geometry data of the transformer
are presented in Table I and in Fig. 2. The following constraints must be
incorporated to properly model the load-loss test. The high voltage (HV)
windings are connected to a three-phase external source whereas the low voltage
(LV) windings are short circuited. The voltage source is feeding impedance
voltage, that is, the nominal voltage times the nominal short circuit impedance
(14.3% of the nominal voltage for the transformer considered in this work, see
Table 1). The conductivity of the tank and clamps is numerically given by 5x106
S/m. The core relative permeability was chosen as 20000 since it is a typical
value for non-saturated steel.

Fig. 1. Complete transformer geometry: windings, core and clamps of transformer.


The tank is not shown
3D Finite Element Estimation of Stray Losses in Three-Phase Transformers 91

Table 1. Data of the transformer

Quantity Value
Rated power 31.5 MVA
Frequency 50Hz
Rated Voltages HV/LV 132/33kV
Rated Currents HV/LV 138/318 A
Number of turns HV/LV 1000/433

1090 1090

560

130

H L L H H L L H H L L H 1520
V V V V V V V V V V V V

560 130

560

Core Tank

150 150

LV LV LV

HV HV HV

150 Windings

All dimension in mm.

Fig. 2. Geometry dimensions of the transformer

Due to the geometry and excitation symmetry that exists in the transformer,
a quarter of the geometry is only modeled to reduce the numerical size of the
finite element model (Fig. 3). External boundary conditions are imposed in
the tank as impedance surfaces while parallel and perpendicular flux conditions
are considered in the two remaining faces as shown in Fig. 4.
92 M.A. Venegas Vega, R. Escarela Pérez, T. Niewierowicz

Fig. 3. Three-dimensional finite element mesh

Fig. 4. Surface impedances and boundary conditions in the transformer

The finite element model consists of three-dimensional elements that include


tetrahedrons, prisms and cubes. the surfaces of impedance are also employed in
the clamps (Fig. 4) the windings of the transformers are connected to external
sources of voltage by means of external resistances as depicted in Fig. 5.

Fig. 5. Connections of windings, voltage sources and external resistances during


the load-loss test
3D Finite Element Estimation of Stray Losses in Three-Phase Transformers 93

3. Electromagnetic formulation
The electromagnetic field is formulated using a time-harmonic model. This
representation means that the ferromagnetic materials show linear behavior,
avoiding, this way, costly time-stepping simulations. The assumption of linearity
is admissible since the load-loss test is performed with low-excitation voltages.
Thus, the transformer is operating in the linear regions of the BH curves of
the transformer ferromagnetic materials. The low-frequency time-harmonic
representation of electromagnetic fields can naturally incorporate impedance
surfaces as boundary conditions.
The model consists of three different regions: coils, transformer core and air.
Tank and frames are not explicitly included as they are represented with
impedance surfaces. This substitution is allowed because the skin depth of the
ferromagnetic materials considered is smaller than the dimensions of the tank
and frame. Explicit consideration of the tank and frames is possible but they
would have to be meshed, leading to a huge mesh that in turn implies high
computational costs. If the reduced magnetic scalar potential is employed,
transformer windings do not need to be meshed. It is also assumed at this stage
that transformer windings are filamentary, that is, the current density is uniform
over the cross section of the winding conductors. The transformer core is
modeled using the total magnetic scalar potential. The regions of eddy currents
use a semi-analytic formulation that uses the total magnetic scalar potential.
A brief description of the electromagnetic formulation using magnetic scalar
potentials is given in the next paragraphs.
The intensity of magnetic field in eddy current free regions can be obtained
as the contribution of two components as follows:
H = H s − ∇φ (1)
where Hs is the magnetic field produced by the density currents JS when the
existence of material media is disregarded and the problem region is considered
unbounded. It is calculated from Biot-Savart's law
1 1
Hs = Js ×∇ dΩ (2)
4π Ω R

The magnetic potential φ in (1) is called the reduced magnetic scalar


potential. It accounts for the presence of magnetic materials. The divergenceless
property of the magnetic field density (∇ ⋅ B = 0 ) and the constitutive relationship
between the magnetic field intensity and density (B = µH ) can be combined
with (1), resulting in the following Poisson type equation:
∇ ⋅ µ∇φ = ∇ ⋅ µH s (3)

Equation (3) also fulfills Ampere’s law since ∇ × H s = J s and ∇ × ( H − H s ) = 0 .


94 M.A. Venegas Vega, R. Escarela Pérez, T. Niewierowicz

Interface boundary conditions between magnetic media 1 and 2 can be


written as:
n ⋅ H s (µ2 − µ1 ) + n ⋅ (µ1∇φ1 − µ 2∇φ2 ) = 0 (4)

for the continuity condition of the normal component of B . The continuity of


the tangential component of H is automatically enforced with any continuous
φ [7].
The main problem of the φ formulation is its poor numerical accuracy in
highly permeable materials, where H s and ∇φ have similar magnitudes since
B is small there.
Cancellation errors in regions with high permeabilities (free of currents of
sources) can be avoided with the use of the total magnetic scalar potential ψ.
The magnetic field intensity in these regions is then given by
H = −∇ψ (5)

Combining the divergenceless property of B and the constitutive


relationship between H and B , the following Laplace type equation is obtained:

∇ ⋅ µ∇ψ = 0 (6)

Interface boundary conditions between magnetic media 1 and 2 can be


written as:
n ⋅ (µ1∇ψ 1 ) + n ⋅ (µ 2 ∇ψ 2 ) (7)

Similarly, the continuity condition of the tangential component of H is


fulfilled with a continuous ψ. Scalar potentials are very attractive when
calculating 3D magnetic fields since the computational times are largely
reduced. The core of the transformer is modeled using this formulation [5].
The whole transformer is then modelled by coupling (3) and (6) at
boundaries where interface boundary conditions are enforced:
∂φ ∂ψ
- µ1 + µ1 H sn ⋅ = - µ 2 (8)
∂n ∂n

− n × ∇φ + n × H s = − n × ∇ ψ (9)
Where n is a normal vector to the surface. The subscript n in the magnetic
field intensity indicates normal component. Thus, the magnetic field Hs is only
calculated in the interface of the regions 1 and 2 [5] for solving (7).
The theory shown in this section has been coded in MEGA [8] a general
purpose electromagnetic software which has 3D graphics capabilities for
displaying of results. This software has also been tested through benchmark
problems that have been accepted by the finite-element community.
3D Finite Element Estimation of Stray Losses in Three-Phase Transformers 95

4. Surface impedances
The transformer has structural metallic parts such as the tank and frames that
are exposed to leakage fields. As a result, losses are generated in these massive
conductors. Direct modeling of these regions is possible with the use of vector
potentials. However a great deal of elements is required to mesh these zones
since the skin depth of the metallic parts is small when compared with their
geometry dimensions. This means that the magnetic field does not penetrate
deep into the conductor. So, direct modeling of eddy current regions not only
requires lots of elements but more degrees of freedom are needed for the vector
potential representation. The analytical solution of flux penetration and currents
into plane-faced boxes is well known [9] and can be used to avoid explicit
meshing of massive conductors. A finite element implementation of the
approach has been proposed in [6] and [4]. It basically uses the analytic solution
of magnetic field as surface impedances in the faces of the elements that
coincide with interfaces of conducting and non conducting regions.
The decay of magnetic field inside a massive conductor is therefore modeled
using the distribution of field penetrating a conductive semi-infinite slab. The
current per unit width [6], it is related with the current density in the surface by:

I = J S δ (1 − j ) / 2 (10)

where δ is the skin depth and given by:

2
δ = (11)
ωσµ

If the conducting region is facing a total magnetic scalar potential region, the
governing equation of the non-conducting region is (5). Finite element
discretization of (5) gives surface integral terms related of H×n. The integrals on
interfaces of eddy current region can be put in terms of the tangential component
of Ψ with the aid of the analytic solution, leading to the following boundary
condition
∂ψ 1− j µi 2
= ∇ ψ (12)
∂n 2 µo
This condition is only applied at interfaces of regions of thin eddy currents.
This way, explicit meshing of eddy current regions and use of vector potentials
has been avoided.

5. External circuits
The electromagnetic field of low-frequency transformers is produced by
currents circulating through its windings. However, these currents are not known
96 M.A. Venegas Vega, R. Escarela Pérez, T. Niewierowicz

a priori because the transformer is fed from voltage sources during the load-loss
test. Hence, the electromagnetic equations are not current driven but voltage
excited. It was already mentioned that transformer windings are considered
filamentary. Hence, uniform current densities circulate through their traverse
sections, neglecting skin and proximity effects.
For a non-current driven problem, it is necessary solve a circuit-field
coupled problem. The key to solve this sort of problems lies in the voltage
equations at the terminals of windings:
∂λ
V = − IR + (13)
∂t
where R is the total resistance of the coil, λ represents the magnetic flux
linkages of a winding and I is the winding current. Flux linkages are calculated
from the magnetic potentials yielding a linking route between field and circuit
equations. This way, additional equations are generated for each winding
connected to external circuits, which are added and solved simultaneously with
the finite element system.

6. Load-Loss test

After establishing boundary conditions and connecting external circuits to


the transformer, it can be proceeded to obtain a finite-element solution for a
specified frequency. The system of simultaneous equations has complex form,
that is, the finite-element global matrix, as well as the forcing vector, has
complex numbers as entries. This in turn means that the solution of potentials is
also complex.
The power absorbed during the load-loss test consists basically of Joule-type
losses (I2R) in winding conductors, losses produced by eddy currents in winding
conductors (neglected in this work) and stray losses in structural metallic parts
(main objective of this work). Losses in the transformer core are disregarded as
well, because the impedance voltage is small (compared with rated voltage), and
the magnetic flux is being forced into air leakage paths.
Transformer impedance voltage [5] is the necessary value to make nominal
current circulate through the transformer windings, when one of the transformer
sides is short-circuited and the other one is fed with impedance voltage. It is
usually expressed as percentage of the transformer rated voltage and equal to the
per-unit leakage impedance value times 100. This result is easily obtained from
the classical T equivalent circuit when the secondary side is short-circuited and
the excitation branch is neglected. Here, the short-circuit impedance nearly
equals the total leakage impedance.
Thus, during the load-loss test of a three phase transformer, the losses are
given by:

Pload = 3PI 2 R + Pstray (14)


3D Finite Element Estimation of Stray Losses in Three-Phase Transformers 97

where:

Pload: Load losses (W)


PI2R: Joule Losses (W)
Pstray: Stray losses (W)

Notice that losses, produced by eddy currents and proximity effects, have
been neglected inside the transformer windings. The stray losses are considered
here as those taking place in metallic structural parts of the transformer. They are
produced by the time-harmonic leakage fluxes reaching these massive conductors.
Fig. 6 shows the distribution of magnetic field in the transformer core and
windings at ωt = 0o, ωt = 120o and ωt = 240o. It can be immediately seen that the
magnetic flux is being expelled from the core legs and therefore forced into air
leakage paths. It is interesting to note that the magnetic fluxes are approximately
90 degrees behind the voltage sources according to Faraday’s law ( E = jωλ in
the frequency domain). This means that voltage is reaching its peak value when
the current is approaching or leaving a zero value. Flux linkages are nearly in
phase with currents. For instance, Fig. 6 shows that the currents on the left core
leg windings are peaking while voltage is zero. All these statements are
approximate since the presence of tank, core and winding resistances introduce
small phase shifts. The numerical results of the short circuit currents and
voltages are presented in Table 2.

Fig. 6. Distribution of the magnetic field during the test of short circuit
98 M.A. Venegas Vega, R. Escarela Pérez, T. Niewierowicz

Table 2. Voltages and currents during the short circuit test

Voltage Current Current


PHASE (HV,RMS) [V] (LV,RMS) [A] (HV,RMS) [A]
A 10910.9 318.5943973 137.804226
∠0º ∠91.27109483º ∠-88.8610295º
B 10910.9 317.6049701 137.8780422
∠ 120º ∠-28.72440114º ∠151.2842585º
C 10910.9 318.5518172 137.7490197
∠120º ∠-148.6828969º ∠31.44052071º

The total power supplied by the voltage sources can now be determined
from the calculated voltages and currents of Table 2. Similarly, the I2R losses
of the transformer windings can be obtained. Losses in massive conductors are
semi-analytically calculated using the surface impedance approach of Section 4.
Input powers, winding losses and stray losses are reported in Tables 3 and 4.
Fig. 7 shows the distribution of losses in the frames and tank of the transformer.

Table 3. Input power supplied by the sources and losses in the primary
and secondary windings of the transformer
Real power
PHASE supplied by Losses, LV [W] Losses, HV [W]
voltage source [W]
A 29914.52928 13794.16702 19407.7721
B 33747.70946 13708.632 19392.29524
C 37818.18498 13790.51752 19428.67148
Total 101480.4237 41293.31655 58228.73882

Table 4. Losses in structural parts


Structural Parts Losses [W]
Tank 1767.49744
Plates Clamps 192.4532
Total 1959.95064

The total three-phase power supplied by the source is 101480.4237 W,


which equals the sum of I2R and stray losses: 101482.006 W. This is a simple
check that verifies the consistency of the finite-element code.
3D Finite Element Estimation of Stray Losses in Three-Phase Transformers 99

Fig. 7. Distribution of surface losses in the tank and frame during


the test of short circuit

7. Conclusions
A 3D finite-element model was developed to determine the magnetic field
and losses of a three-phase three-limb transformer. The load-loss test was
simulated to this end, making possible the use of a time-harmonic linear model.
The determination of stray losses in the tank and frame of the transformer was
achieved using scalar potentials and surface impedance boundary conditions.
This resulted in a relatively small 3D finite-element mesh with less degrees of
freedom than the required by vector potentials. The numeric model was able to
incorporate external circuits through a circuit-field coupling approach. The
numerical results show that the balance of powers during the test is completely
fulfilled, verifying the consistence of the computations. Currents, voltages and
losses also concur with expected experimental results.

References
[1] Olivares J.C., Escarela-Perez R., Kulkarni S.V., de León F., Venegas-Vega M.A.:
2D finite-element determination of tank wall losses in pad-mounted transformers,
Journal of Electric Power Systems Research (Elsevier), 71 (2004) 179-185.
100 M.A. Venegas Vega, R. Escarela Pérez, T. Niewierowicz

[2] Olivares J.C., Cañedo J., Moreno P., Driesen J., Escarela-Perez R.: Experimental
Study to Reduce the Distribution-Transformers Stray Losses Using Electromagnetic
Shields, Journal of Electric Power Systems Research (Elsevier), 63 (2002) 1-7.
[3] Guérin C., Tanneau G., Meunier G.: 3D Eddy Current Losses Calculation in
Transformer Tanks Using the Finite Element Method, IEEE Trans. Magn., Vol. 29,
(1993) 1419-1422.
[4] Holland S., O’Connell G., Haydock L.: Surface Impedance for 3D Non-linear
Eddy Current Problems Application to Loss in Transformer, IEEE Trans. Magn.,
Vol. 32, (1996) 808-811.
[5] Rodger D., Eastham J.F.: A formulation for low frequency eddy current solutions,
IEEE Trans. Magn., Vol. MAG 19, (1983) 2443-2446.
[6] Rodger D., Leonard P.J., Lai H.C., Coles P.C.: Finite element modeling of thin
skin depth problems using magnetic vector potential, IEEE Transactions on
Magnetic, Vol. 33, (1997) 1299-1301.
[7] Binns K.J., Lawrenson P.J., Trowbridge C.W.: The Analitycal and Numerical
Solution of Electric and Magnetic Fields, John Wiley and Sons, 1992.
[8] MEGA V6.27: USER’S MANUAL, Applied Electromagnetic Research Centre, Bath
University, UK, 2000.
[9] Carter G.W.: Electromagnetic Field and its Engineering Aspects, Second Edition.
Longman, London, 1967.

You might also like