Nested Multigrid Finite Element Analyses

of Eddy Current Losses in Power Transformers

Erich Schmidt1, Joachim Schöberl2 , Peter Hamberger3

1
Institute of Electrical Drives and Machines, Vienna University of Technology
A–1040 Vienna, Austria, Email: [email protected]
2 Institute of Computational Mathematics, Johannes Kepler University Linz
A-4040 Linz, Austria, Email: [email protected]
3
VA TECH EBG Transformatoren GmbH & Co
A-4020 Linz, Austria, Email: [email protected]

Abstract: The distribution of eddy currents in the core clamping parts and the tank wall of
power transformers is obviously different between measurements of load losses and normal op-
erational conditions. Thus, the calculation of these losses shows the regions of possible hot-spot
temperatures with normal operational conditions. The presented analysis method uses 3D non-
linear time-harmonic finite element analyses with a nested multigrid solver. Consequently, the
occurring short calculation times allow to take into account for ambitious technical specifications
and individual customer demands.
Keywords: Eddy currents, Power transformer losses, Multigrid solvers, Finite element methods

1. Introduction

Apart from core and winding losses, the eddy current losses in the steel tank wall and in
the core clamping parts represent the largest components of stray load losses [1], [2]. The eddy
currents in these regions are obviously different between measurements of load losses and normal
operational conditions. Due to this fact, the calculation of these eddy current losses is a matter
of interest with the initial design and design optimization of large power transformers.
With the aim of an improvement of the design calculation methods, the performed analyses
should have capabilities of assessing or revalidating the design accurately under all anticipated
operating conditions. Thus, fully automated geometry descriptions as well as short calculation
times are necessary to take into account for ambitious technical specifications according to
individual customer demands.

2. Transformer Modeling

As described in [3], the finite element discretization can be generated automatically from the
detailed geometry data in particular of windings, laminated iron core, clamping parts, tank wall.
Thereby, second and, within the core regions, forth order tetrahedral edge elements are used
as introduced by [4], [5]. The inner regions of the laminated core are modeled with anisotropic
permeabilities with regard to different magnetization directions of core and yoke [6]. Appropriate
constraints inside the core regions take into account for the lamination. To encounter for eddy
currents in the outer sheets of the laminated iron core, two dimensional current carrying sheets
with a lateral conductivity are modelled [2], [7]. The excitation is carried out with uniform
current densities within the winding regions in accordance to the specified operating conditions.

©2005 ACES
 3. Finite Element Formulation

The Maxwell equations with neglected displacement currents read as

~
~ =σE
curl H ~ + J~0 , ~ = − ∂B ,
curl E (1)
∼
∂t

where σ∼
denotes an anisotropic conductivity tensor, J~0 an applied source current density dis-
tribution [8], [9]. Additionally for completion, the nonlinear anisotropic constitutive relation is
given by H~ = ν · B,~ where ν denotes an anisotropic reluctivity tensor. Introducing the magnetic
∼ ∼
vector potential as B ~ = curl A
~ leads to the differential equation

~
~ + σ ∂ A = J~0 in Ω
 
curl ν∼ · curl A (2)
∼
∂t
with appropriate Neumann and Dirichlet boundary conditions
 
ν · curl ~ × ~n = K
A ~ on ΓH , A
~ × ~n = ~0 on ΓB , (3)
∼

where ΓH denotes the boundary where ~n × H ~ is specified and ΓB the boundary where ~n · B
~
is specified. In non-conducting regions, the vector potential is unique only up to gradient
fields [8], [9]. Thus, we apply gauging by introducing a very small conductivity everywhere in
non-conducting regions.
To avoid the nonlinear time-stepping solution process, an iterative time-harmonic solution is
introduced. Thereby, the unknown degrees of freedom are replaced by their complex magni-
tudes according to Uk (t) = Re Uk e jωt . The nonlinear magnetic characteristics of the utilized

materials are included by evaluating the local time-dependent components of the magnetic flux
density and the corresponding

magnetic field strength according to the local magnetization di-
~ ~ ~
rections as H(t) = ν∼ |B(t)| · B(t). The fundamental harmonics obtained from Fourier analyses
are used to evaluate equivalent anisotropic permeabilities for the governing equations.

4. Multigrid Solver

Multigrid methods have been established as among the most efficient solvers for discretized
electromagnetic elliptic problems [10], [11]. A considerably faster solution strategy compared
to conventional ones is established by using a nested multigrid solver. Thereby, a hierarchy of
discretizations represented by


jωCq + Kq Uq Uq = Pq , q = 1, 2, . . ., l , (4)

from the very coarse level q = 1 up to the finest level q = l is constructed. Special attention
has to be given on a suitable prolongation of the degrees of freedom with the refinement to
guarantee the conservation of the magnetic flux as shown with Fig. 1 and Fig. 2.
The nonlinear matrix equation (4) of the eddy current problem is solved by an approximate
Newton-Raphson method. In each iteration step, the preconditioner is constructed from a
hierarchy of edge finite element spaces and choosing appropriate smoothing procedures for each
of these spaces [10].
 a q,2 a q,3
Prolongation a q,8
a q-1,1 a q-1,2
aq,1 a q,4
a q,7 a q,9

a q-1,3 a q,6 a q,5

Fig. 1: Multigrid solution with recursively Fig. 2: Prolongation of the edge degrees of freedom
restarted two-grid algorithms with the refinement

5. Analysis Results

Fig. 3 shows the distribution of eddy current losses in the core clamping parts with two
different core properties for rated current excitation of LV and HV windings of a 160 MVA
transformer. With regard to the measurement tests, there is no significant saturation of the
core. Thus, a high effective core permeability is established and the stray flux portions between
the legs are negligible. On the other hand with normal operational conditions, the saturated core
yields to a lower effective core permeability. Therefore, significant stray flux portions between
two legs result in high local eddy current losses in these regions of the core clamping parts.

Fig. 3: Eddy current losses in the core clamping parts for rated current excitation of the LV and HV
windings, high effective core permeability with measurement of load losses (left), low effective
core permeability with normal operational conditions (right)

TABLE I
Eddy Current Losses [W] in Various Tank Wall Regions

High Effective Core Permeability Low Effective Core Permeability

with Measurement of Load Losses with Normal Operational Conditions
Phase 1 Phase 2 Phase 3 Phase 1 Phase 2 Phase 3
left right left right left right left right left right left right
Region 1 3398 1089 1100 1097 1077 3402 3070 878 868 870 873 3072
Region 2 254 107 135 124 128 294 245 92 107 99 109 282
Region 3 102 125 121 118 127 95 67 53 49 50 59 62
Region 4 423 238 234 205 235 397 349 96 85 77 96 326
Region 5 205 142 166 158 142 193 172 63 55 57 62 161
Region 6 95 575 621 614 571 94 72 92 91 87 90 69
Region 7 219 287 297 296 263 203 179 165 130 139 150 165
Summary 4696 2563 2674 2612 2543 4678 4154 1439 1385 1379 1439 4137
 Table I lists the eddy current losses in the tank wall according to the above load condition
and the two core properties. The behaviour of the eddy currents in the tank wall is contrary
to those in the core clamping parts. A lower effective core permeability arising with normal
operational conditions yields to significantly reduced eddy current losses in the regions between
two phases. A comparison of the values listed in Table I points out the reduction of about 70%
in the region along the height of the yoke corresponds to the height of the eddy current carrying
core clamping parts.
Consequently, the completely different distribution of the stray flux between the legs arising
with core properties according to measurement tests and normal operational conditions will
significantly influence the investigated eddy current losses. In particular, there are lower losses
in the core clamping parts and higher losses in the tank wall with core properties according
to measurement tests. On the other hand, core properties arising with normal operational
conditions yield to higher losses in the core clamping parts and lower losses in the tank wall.

6. Conclusion

With the aim of an improvement of design calculation methods, the eddy current losses in the
core clamping parts and in the steel tank wall of power transformers in the range of 120 MVA to
200 MVA are investigated. The increasing requirements and challenges with the design of power
transformers ask for automated initial mesh generation based on detailed geometry data as well
as short calculation times. Therefore, a nested multigrid finite element calculation scheme for
the 3D nonlinear eddy current analyses has been presented. The obviously different distributions
of eddy current losses in the core clamping parts and the tank wall of power transformers caused
by different stray flux distributions between measurements of load losses and normal operational
conditions are described in detail.

References
[1] Guérin C., Tanneau G., Meunier G.: ”3D Eddy Current Losses Calculation in Transformer Tanks Using
the Finite Element Method”. IEEE Transactions on Magnetics, Vol. 29, No. 2, March 1993.
[2] Guérin C., Tanneau G., Meunier G., Labie P., Ngnegueu T., Sacotte M.: ”A Shell Element for Computing
3D Eddy Currents – Application to Transformers”. IEEE Transactions on Magnetics, Vol. 31, No. 3, May
1995.
[3] Schöberl J.: ”NETGEN - An advancing front 2D/3D mesh generator based on abstract rules”. Journal
Computing and Visualization in Science, Vol. 1, No. 1, 1997.
[4] Nédélec J.C.: ”Mixed Finite Elements in R3 ”. Numerische Mathematik, Vol. 35, No. 3, 1980.
[5] Nédélec J.C.: ”A New Family of Mixed Finite Elements in R3 ”. Numerische Mathematik, Vol. 50, No. 1,
1986.
[6] Schmidt E.: ”Representation of Laminated and Slotted Configurations in the Finite Element Analysis of
Electrical Machines and Transformers”. Digests of the 14th Conference on the Computation of Electromag-
netic Fields, COMPUMAG, Saratoga Springs (NY, USA), 2003.
[7] Biro O., Bardi I., Preis K., Renhart W., Richter K.R.: ”A Finite Element Formulation for Eddy Current
Carrying Ferromagnetic Thin Sheets”. IEEE Transactions on Magnetics, Vol. 33, No. 2, March 1997.
[8] Biro O., Preis K., Richter K.R.: ”Various FEM Formulations for the Calculation of Transient 3D Eddy
Currents in Nonlinear Media”. IEEE Transactions on Magnetics, Vol. 31, No. 3, May 1995.
[9] Biro O., Preis K., Richter K.R.: ”On the Use of the Magnetic Vector Potential in the Nodal and Edge
Finite Element Analysis of 3D Magnetostatic Problems”. IEEE Transactions on Magnetics, Vol. 32, No. 3,
May 1996.
[10] Arnold D.N., Falk R.S., Winther R.: ”Multigrid in H(div) and H(curl)”. Numerische Mathematik, Vol. 85,
No. 2, April 2000.
[11] Schinnerl M., Schöberl J., Kaltenbacher M.: ”Nested Multigrid Methods for the Fast Numerical Computa-
tion of 3D Magnetic Fields”. IEEE Transactions on Magnetics, Vol. 36, No. 4, July 2000.