7th Mediterranean Conference and Exhibition on Power Generation, Transmission, Distribution and Energy Conversion

7-10 November 2010, Agia Napa, Cyprus (Paper No. MED10/161)

Grid Voltage Harmonics Effect on Distribution

Transformer Operation
Themistoklis D. Kefalas, Member, IEEE and Antonios G. Kladas, Member, IEEE

Abstract-- Distribution transformer operation is sensitive to

the distortion of the supply voltage waveform. According to
Strategies for development and diffusion of Energy Efficient
Distribution Transformers (SEEDT), reactive power and
harmonic losses add a further 5 TWh/year to the losses of
European Union (EU-27) distribution transformers. In the
present paper a systematic experimental procedure is
developed in order to evaluate the effect of voltage harmonics
on distribution transformer operation. Also, a theoretical
analysis based on the hysteresis design tool of Matlab and a
finite element code integrating the hysteresis phenomena is
carried out.
Index Terms--Computer aided analysis, finite element
methods, industrial power system harmonics, magnetic cores,
nonlinear magnetics, numerical analysis, power transformers,
magnetic hysteresis, magnetic losses.

I. INTRODUCTION

ISTRIBUTION transformers losses are equal to almost

2% of the electricity generated worldwide and only in
the European Union are estimated at about 33 TWh/year
according to Strategies for development and diffusion of
Energy Efficient Distribution Transformers (SEEDT) [1].
This fact renders distribution transformers the second most
energy intensive component of the power grid.
Approximately 75% of total losses are due to core losses as
a result of the loading characteristics of distribution
transformers [1], [2]. Consequently, the reduction of
distribution transformer no-load losses has recently become
top priority of many countries [3].
The deterioration of the supply voltage waveform is
caused by industrial and domestic loads, e.g. switch mode
power supplies that inject current harmonics in the power
grid. Key operational characteristics of transformers such as
no load loss and magnetizing current are sensitive to the
distortion of the supply voltage waveform [2]. According to
SEEDT, reactive power and harmonic losses add a further 5
TWh/year to the losses of European Union (EU-27)
distribution transformers [1].The accurate evaluation of the
operating characteristics of transformers under distorted
supply voltage conditions is of great importance for
transformer manufacturers as it will aid in evaluating the
technically and economically optimum transformer [3].
In this paper a systematic experimental procedure is
developed in order to estimate the supply voltage harmonic
impact on transformer operation. Particular emphasis is
given to the distorted line-to-neutral and line-to-line voltage
T. D. Kefalas and A. G. Kladas are with the School of Electrical and
Computer Engineering, National Technical University of Athens, 15780
Athens, Greece (e-mail:thkefala@central.ntua.gr, kladasel@central.ntua.gr).

waveform of the power grid. Finally, a theoretical analysis

based on a Matlab model and a FE code integrating an
inverse hysteresis model is presented [2].
II. ANALYSIS BASED ON THE HYSTERESIS DESIGN TOOL OF
MATLAB
The Matlab model used for the no-load loss analysis of
one-phase wound core transformers is shown in Fig. 1. The
model uses the Saturable Transformer block from the
SimPowerSystems library of Simulink. The Simulate
hysteresis option of the Saturable Transformer block
must be selected. The parameters of the hysteresis model
were set using the graphical user interface of the hysteresis
design tool. The value of the parameters for the
conventional grain oriented electrical steel M4 (thickness
0.27 mm) used in the present paper are shown in Table I,
where Fr is the remanent flux, Fs is the saturation flux, Is is
the saturation current, Ic is the coercive current, and dF/dI is
the gradient of the flux at coercive current. The input
voltage can be modeled either directly by inserting
experimental data to a 1D lookup table, or by using a
function block. The output of the model is the no-load
current of the Saturable Transformer block.

Fig. 1. Matlab model based on the hysteresis design tool.

TABLE
PARAMETERS OF THE HYSTERESIS MODEL
Parameter
Value
Fr (Vs)
0.31500
Fs (Vs)
0.33579
Is (A)
8.01450
Ic (A)
0.78660
dF/dI
3.0

III. FE ANALYSIS INTEGRATING HYSTERESIS MODEL

The hysteresis model is similar to the Jiles-Atherton
model, but the independent variable is the flux density B
instead of the magnetic field H [4]. In this way the
hysteresis model can be integrated in a two dimensional
(2D) finite element (FE) analysis, where B in each element
can be directly evaluated by the curl of the magnetic vector
potential A . Simplicity, computational efficiency, and ease
of implementation are its main advantages in contrast with
other hysteresis models like the Preisach [5]. According to
[4], H is partitioned in two components the anhysteretic
H AN , and the hysteretic H H as follows
H = H AN + H H .

(1)

Furthermore, H H is divided in two parts, one associated

with the irreversible part of wall domains motion H HW , and
one associated with the reversible motion of wall domains
H RET . Thus, it follows that the magnetic energy of the
hysteretic component H H , is given by

H dB = H
B

HW

dB H RET dB.

(2)

The anhysteretic component H AN is obtained by solving

(3), where M S is the saturation magnetization, L( AN ) is

the Langevin function of AN given by (4) and (5), and ,
a are parameters of the Langevin function.
H AN = B / 0 M S L( AN )

(3)

L( AN ) = coth ( AN ) 1 / AN

(4)

AN =

H AN (1 ) + B( / 0 )
a

(5)

Application of the Euler scheme to (8) yields the

following nonlinear equation from the solution of which the
hysteretic component H H is obtained.
I D H H H = (H HS L(H ) H H )B

In 2D FE analysis the Poissons equation is solved which

is a function of the magnetic vector potential and the
reluctivity. When hysteresis is taken into account, numerical
difficulties arise due to the discontinuity of the reluctivity
when B is equal to zero. The aforesaid problem is tackled
by considering the constitutive equation for ferromagnetic
materials [4], [5]
B = 0 (H + M ).

M y M x
v0 Az +

y
x

M x
dM x Bx
dM = M

y
y

B
x

H RET = I D H

dH H
dB

(7)

M x
B y dBx

.
M y dBy

By

(12)

By using (12) and the 2D vector potential, the second left

hand term of (11) is given by
M y

M x M y Az M y Az

y
Bx x y By x x

M x Az M x Az

.
Bx y y By y x

respectively, where I D is a variable that takes the value +1

(6)

(11)

M y = f (Bx , B y ) , the differential of M is given by

of the hysteretic component H H , are given by (6) and (7)

H H + I D H HS
aH

+ J z = 0.

Considering that the two components of the

magnetization are a function of B [6], M x = f (Bx , B y ) ,

The irreversible part H HW , and the reversible part H RET

H HW = H HS L(H ), H =

(10)

Combining the above equation and Amperes law yields

the following equation for the 2D problem

when B > 0 and 1 when B < 0 , and H HS , aH , H are

parameters of the hysteresis model.

(9)

(13)

Also the relationship between the differentials of the

magnetization and the flux density can be expressed by
means of a tensor [6], [7]. By using the Euler scheme to
represent the differentials yields
M x M cos
M =

B sin
y

sin Bx

cos By

(14)

The differential equation of hysteresis is obtained by

substituting (6), (7) into (2) and isolating dH H / dB

where is the rotation angle between M and B . By

multiplying and dividing the right hand side of (14) with
B it follows that

dH H H HS L(H ) H H
.
=
dB
I D H

M x
M =
y

(8)

Bx

By

(15)

where and are given by (16).

M B
B

, =

M B
B

(16)

From (12) and (15) it follows that

M x
B
x
M y
B
x

M x
By
=
M y
By

(17)

The output of the probes was connected to a noise

rejecting, shielded BNC I/O connector block (BNC-2110)
via passive probes (TP6060). A noise rejecting, shielded
cable (SHC68-68-EP) connects the data acquisition (DAQ)
device directly to the BNC-2110 connector block.
The DAQ device used was National Instruments (NI)
6143. NI 6143 has 8 differential channels, an ADC
resolution of 16-bit, a maximum sampling rate of 250
kSamples / s per channel analog input, and a 5 V analog
input signal range. Finally, the DAQ device was placed into
a PCI slot of a desktop PC (processor: 1.8 GHz Intel Core 2
Duo, memory: 1 GB DDR2).

The problems solution is obtained by discretizing (18),

that is derived by substituting (13) and (17) into (11).
v
0

Az + J z = 0

v0

(18)

The coupled field-circuit global system of equations is

given by (19) where D is the potential-current coupling
stiffness matrix, G is the inductive damping matrix, and
, I are the time derivatives of the magnetic vector
A
potential and electric current vectors
the time-dependent system (19),
numerical integration technique
nonlinearities are taken into account
Raphson iterative scheme.

respectively. To solve
the Euler Backward
is used, and the
by using the Newton-

S D A 0
0 0 A
G L + 0 R I = V

Fig. 2. Experimental setup.

(19)

IV. EXPERIMENTAL SETUP

A 16 turns excitation coil was supplied from a
programmable ac power supply (California Instruments
MX30) in order to magnetize the wound core with a number
of voltage waveforms, Fig. 2.
The voltage across the excitation coil terminals was
captured using an active differential voltage probe. The
specific voltage probe provides safe means of measuring
floating potentials by converting the high input differential
voltage ( 1400 V peak) into a low voltage ( 5 V). The 3
dB frequency of the voltage probe is 18 MHz, the rejection
rate on common mode at 50 Hz is 90 dB, the dc output
offset voltage is 0.15 mV and the attenuation factor is equal
to 205.
A current probe based on the Hall Effect is used for
capturing the no-load current. It provides a galvanic
isolation between the primary circuit (high power) and the
secondary circuit (electronic circuit), and very good
linearity (< 0.2 %). The current measurement range is 36 A
peak and the 1 dB frequency of the probe is 150 kHz.
The voltage and current probe were designed and
manufactured by the authors at the Laboratory of Electrical
Machines at National Technical University of Athens
(NTUA). The circuit analysis of the probes was carried out
by using OrCAD-PSpice. Both probes are depicted in Fig. 3.

Fig. 3. DAQ system, wound core, current and voltage probes.

Three virtual instruments (VI) were created with the use

of LabVIEW software [2]. The purpose of the first VI was
the real-time measurement, of the excitation coil voltage and
no-load current waveforms, fast Fourier transform (FFT),
peak, rms, and THD values. The second VI was used for
capturing the voltage and no-load current waveforms and
FFT, for two periods of the fundamental frequency and for
the maximum sampling rate of the DAQ device. The third
VI was used for manipulating the acquired voltage and
current data in order to compute the no-load loss.
The repeatability error of the experimental setup is 0.3
%, whereas the absolute error is within 0.5 %. Thus, the
specific experimental setup permits the evaluation of the
harmonic impact on no-load loss even in cases of distorted
supply voltage waveforms of low THD.
The experimental setup was also used for local flux
density measurements in order to evaluate the FE computed
flux density distribution of wound cores [8]. Two turns

search coils wound around the total width of the steel sheet,
were inserted in order to determine the peak flux density
distribution along the limb, yoke and corner of the wound
cores. Solderable enameled copper wire 0.1 mm in diameter
was used for the search coils. The overall loss of the core
did not change noticeably after the coils were wound in
position, showing that the flux distribution did not change
much due to the search coils being introduced. The voltages
induced in the search coils were captured by directly
connecting the search coils into the NI6143 DAQ card
inputs. Necessary calculations and analysis of the captured
data were carried out by using LabVIEW software [8].
V. RESULTS AND DISCUSSION
A total of seven different voltage waveforms, for five
rms values (15 V, 20 V, 23 V, 24 V, and 25 V)
corresponding to magnetization levels from 1.0 T to 1.6 T
have been studied. The results presented hereafter are for an
rms input voltage value of 20 V and the simulated no-load
current resulted by using the Matlab model.

Fig. 7. Experimental and simulated no-load current for square voltage.

A comparison of Figs. 5, 7 show that the impact of the

square voltage waveform on no-load current and
consequently on no-load loss is more significant than that of
the triangular voltage waveform.

Fig. 4. Triangular voltage waveform.

B. Sinusoidal and distorted sinusoidal voltage waveforms

Figs. 8, 9 show the sinusoidal voltage waveform and the
resulting experimental and simulated no-load current
respectively. Figs. 10, 11 show a sinusoidal waveform with
a superimposed 3rd harmonic and the resulting no-load
current. The percentage of the amplitude and the phase
angle relative to the fundamental, of the 3rd harmonic are
10% and 0o respectively. The specific distorted sinusoidal
waveform resembles a line-to-neutral voltage waveform.
Figs. 12, 13 show a distorted sinusoidal waveform with a
superimposed 3rd harmonic of the same percentage
amplitude and a phase angle relative to the fundamental of
90o, as well as the resulting no-load current. The specific
distorted sinusoidal waveform resembles a line-to-line
voltage waveform. Figs. 9, 11, 13 show that the distorted
sinusoidal waveform has a more adverse effect on no-load
operation than the sinusoidal waveform. Furthermore, the
distorted sinusoidal having a phase angle relative to the
fundamental of 90o, affects no-load current and
consequently no-load loss more than the distorted sinusoidal
resembling line-to-neutral waveform.

Fig. 5. Experimental and simulated no-load current for triangular voltage.

Fig. 8. Sinusoidal voltage waveform.

A. Triangular and square voltage waveforms

Triangular and square voltage waveforms arise in cases
where transformers connect PV and wind power plants to
the medium voltage grid. Figs. 4-7 show the triangular and
square voltage waveform and the resulting experimental and
simulated no-load current.

Fig. 6. Square voltage waveform.

Fig. 9. Experimental and simulated no-load current for sinusoidal voltage.

and the resulting experimental and simulated current.

Analogous procedure was followed for the deformed
line-to-line voltage waveform. The corresponding harmonic
content is shown in Table III. The line-to-line waveform and
the resulting experimental and simulated currents are shown
in Figs. 16, 17.

Fig. 10. Deformed sinusoidal voltage waveform (0o phase).

TABLE I
DEFORMED LINE-TO-NEUTRAL VOLTAGE HARMONIC CONTENT
Harmonic
Percentage (%)
Phase (Degrees)
3
6.9667
162.80
5
7.3618
-177.19
7
2.9613
-169.40
9
1.8317
-174.60
11
0.4802
-39.044
13
1.0927
-1.60
15
0.5094
21.89
17
0.0800
67.37
19
0.2693
-148.93
21
0.2610
-165.49
23
0.1610
133.49
25
0.0145
-15.27
27
0.0775
-32.10
29
0.0939
26.64

Fig. 11. Experimental and simulated no-load current for deformed

sinusoidal voltage (0o phase).

Fig. 14. Deformed line-to-neutral voltage waveform.

Fig. 12. Deformed sinusoidal voltage waveform (90o phase).

Fig. 15. Experimental and simulated no-load current for deformed line-toneutral voltage.

Fig. 13. Experimental and simulated no-load current for deformed

sinusoidal voltage (90o phase).

C. Deformed line-to-neutral and line-to-line voltage

waveforms
The programmable ac power supply was used to simulate
the deformed line-to-neutral waveform. Table II shows the
amplitude percentage and the phase angle relative to the
fundamental of the odd higher harmonics up to the 29th
harmonic. The data shown in Table II were obtained
experimentally from the deformed line-to-neutral voltage
waveform of the Greek grid. Figs. 14, 15 show the
corresponding deformed line-to-neutral voltage waveform

TABLE II
DEFORMED LINE-TO-LINE VOLTAGE HARMONIC CONTENT
Harmonic
Percentage (%)
Phase (Degrees)
3
9.0438
165.31
5
5.8257
176.23
7
4.9526
-175.03
9
1.1832
-147.55
11
0.7256
-31.62
13
1.3199
3.33
15
0.5981
23.35
17
0.0813
-86.25
19
0.3052
171.63
21
0.3616
-162.15
23
0.0748
138.15
25
0.0822
-177.50
27
0.1891
-175.85
29
0.0764
-32.15

In both cases the impact on no-load current is significant

in comparison with the case of the sinusoidal supply voltage
waveform.

VI. CONCLUSION
The phase of the voltage harmonics relative to the
fundamental, as well as their amplitude influences
transformer operation. The paper clearly demonstrates that
the higher harmonic content of deformed line-to-neutral and
especially line-to-line voltage waveform has an adverse
effect on the no-load operation of distribution transformers
and this fact must be taken in consideration by transformer
manufacturers and utilities.
VII. REFERENCES
[1]

Fig. 16. Deformed line-to-line voltage waveform.

[2]
[3]
[4]

[5]
[6]
Fig. 17. Experimental and simulated no-load current for deformed line-toline voltage.

D. Comparison of computed and experimental no-load loss

Table IV shows that the computed no-load loss is in
relatively good agreement with the experimental results in
the case where the Matlab hysteresis model is used.
Computed results are improved when the FE hysteresis
model is used. Nevertheless, the computational cost of the
latter is significantly larger than that of the Matlab model.
Also, Table IV shows that the deterioration of no-load loss
is more significant in the cases of distorted sinusoidal, lineto-neutral, and especially line-to-line voltage waveforms,
than in the case of the sinusoidal voltage waveform.
TABLE V
COMPUTED AND EXPERIMENTAL NO-LOAD LOSS
Waveform
No-load loss (W) No-load loss (W) No-load loss (W)
Experiment
Matlab model
FE hysteresis
Triangular

9.626

8.856

9.237

Square

11.510

10.709

10.975

Sinusoidal

9.629

8.767

9.292

Sinusoidal,
3rd harmonic:
10%, 0o

9.654

8.891

9.219

Sinusoidal,
3rd harmonic:
10%, 90o

9.930

9.255

9.543

Line-to-neutral

10.10

9.282

9.615

Line-to-line

10.38

9.345

9.717

[7]

[8]

F. V. Topalis, R. Targosz, W. Irrek, A. Rialhe, A. Baginski,

Strategies for development and diffusion of energy efficient
distribution transformers in Europe, in Proc. IEEE CEFC, 2008, pp.
582.
T. D. Kefalas and A. G. Kladas, Harmonic impact on distribution
transformer no-load loss, IEEE Transactions on Industrial
Electronics, vol. 57, no. 1, pp. 193-200, Jan. 2010.
P. Georgilakis, N. Hatziargyriou, and D. Paparigas, AI helps reduce
transformer iron losses, IEEE Computer Applications in Power, vol.
12, no 4, pp. 41-46, Oct. 1999.
P. I. Koltermann, L. A. Righi, J. P. A. Bastos, R. Carlson, N.
Sadowski, N. J. Batistela, A modified Jiles method for hysteresis
computation including minor loops, Physica B, vol. 275, pp. 233237, 2000.
H. L. Toms, R. G. Colclaser, M. P. Krefta, Two-dimensional finite
element magnetic modeling for scalar hysteresis effects, IEEE Trans.
Magnetics, vol. 37, no 2, pp. 982-988, Mar. 2001.
P. I. Koltermann, J. P. A. Bastos, N. Sadowski, N. J. Batistela, A.
Kost, L. Jnicke, K. Miethner, Nonlinear magnetic field model by
FEM taking into account hysteresis characteristics with M-B
variables, IEEE Trans. Magnetics, vol. 38, no 2, pp. 897-900, Mar.
2002.
L. A. Righi, P. I. Koltermann, N. Sadowski, J. P. A. Bastos, R.
Carlson, A. Kost, L. Jnicke, D. Lederer, Non-linear magnetic field
analysis by FEM using Langevin function, IEEE Trans. Magnetics,
vol. 36, no 4, pp. 1263-1266, Jul. 2000.
G. Loizos, T. D. Kefalas, A. G. Kladas, A. T. Souflaris, Flux
distribution analysis in threephase SiFe wound transformer cores,
IEEE Trans. Magn., vol. 46, no. 2, pp. 594597, Feb. 2010.

VIII. BIOGRAPHIES
Themistoklis D. Kefalas (M09) was born in Greece in 1977. He received
the Electrical Engineering Educator degree from the School of Pedagogical
and Technological Education, Athens, Greece, in 1999 and the Diploma
and the Ph.D. degree in electrical engineering from the National Technical
University of Athens, Athens, Greece, in 2005 and 2008, respectively.
He is currently with the School of Electrical and Computer Engineering,
National Technical University of Athens. His research interests include
transformer and electric machine modeling and optimization.
Dr. Kefalas is a member of the Technical Chamber of Greece.
Antonios G. Kladas (S80A99M02) was born in Greece in 1959. He
received the Diploma in electrical engineering from the Aristotle University
of Thessaloniki, Thessaloniki, Greece, in 1982 and the D.E.A. and Ph.D.
degrees from Pierre and Marie Curie University (Paris 6), Paris, France, in
1983 and 1987, respectively.
He was an Associate Assistant with Pierre and Marie Curie University
from 1984 to 1989. From 1991 to 1996, he was with the Public Power
Corporation of Greece, where he was engaged in the System Studies
Department. Since 1996, he has been with the School of Electrical and
Computer Engineering, National Technical University of Athens, Athens,
Greece, where he is currently a Professor. His research interests include
transformer and electric machine modeling and design, as well as the
analysis of generating units by renewable energy sources and industrial
drives.
Dr. Kladas is a member of the Technical Chamber of Greece