1432

IEEE Transactions on Power Delivery. Vol. 10. No. 3, July 1995

Saturation Curves of Delta-Connected Transformersfrom Measurements

Washington L. A. Neve;
Student Member, IEEE

Hermann W. Dommel
Fellow, IEEE

Department of Electrical Engineering

University of British Columbia
Vancouver, B.C.,
Canada V6T 124

Abstract
An algorithm is described for the computation of

the saturation
characteristics of three-~hase deltacoMected transformers from
measured V- Z- curves and no load losses at rated frequency.
For these transfor",
positive sequence excitation tests are usually
carried out with the delta connection closed. While triplen harmonic
currents are present in the deltaamected windmgs, they do not
appear in the line connections to the three-phase source. This paper
presents an algorithm which accounts for the missing triplen harmonics on the line side in the derivation of the saturation chancteristics.

For three-phase transformers, the standard excitation test data

- Z- curves and no-load
available are the positive sequence Vlosses. Figure 1 shows a symmetrical three-phase voltage source
supplying a no-load delta-connected transformer. The delta branches
consist of nonlinear elements. In general, excitation tests are carried
out with a closed delta. In that case, ammeters, placed in series with
the line, will not contain the triplen harmonic currents because these
circulate in the delta connection. This paper presents a method for
generathe piecewise linear saturation curves (nonlinear resistance
and nonlinear inductance), which accounts for the fact that triplen
harmonics circulate in the closed delta, but do not appear in the
measured line currents.

Key words: Transformer, delta connection, saturation, harmonics.

1 Introduction
insulation coordination studies are needed to specify the insulation
levels of power apparatus and systems. These studies are usually done
with digital computer programs, such as the EMTP (Electromagnetic
Transients Program). They require mathematical models for the
various power system components.
The type of model depends on the type of study. For inrush current
and ferroresonance studies, models for transformers and reactors must
represent saturation enects reasonably well. For this, the instantaneous saturation characteristic, which gives flux llnkage R as a
function of current i , is neded. Numerical methods have been used
for some time to produce these peak fluxcurrent curves from
curves[lJ].
measured Vms- -I
A simple circuit consisting of a nonlinear resistance in parallel with a
nonlinear inductance represents the transformer core reasonably
we11[3,4]. An algorithm to obtain the nonlinear resistance (piecewise
linear v - i, curve) and the nonlinear inductance (piecewise linear L il curve) from the measured V- Z- curve and no-load losses was
presented in [SI. This algorithm assumed all cdd harmonic current
components to be present in the measured values.

Figure 1: D ella-con n e c l e d transform erPosit i v e seq u en c e e: c i t at i on test.

2 Basic Considerations
In the circuit of Figure 1, the three branch elements of the delta
connection are assumed to be nonlinear and identical. The branch
currents can be Written as a Fourier time series containing odd
harmonic components only. Then:

- -

On leave from Universidade Federal da Paraiba, Campina Grande PB

Brazil.

The triplen harmonic currents (Z3. Zg...) are in phase (zero sequence
harmonics).
The nns current in each branch is
(4)

94 SM 459-8 PWRD

A paper recommended and approved

by the IEEE Transmission and Distribution Committee
of the IEEE Power Engineering Society f o r presentation at the IF,EE/PES 1994 Summer Meeting, San
Francisco,
CA, July 24 - 28, 1994.
Manuscript
submitted December 30, 1992; made available for
printing April 20, 1994.

Thelinecurrentsare:

0885-8977/95/$04.00 0 1994 IEEE

ia(r)= id(r)-ic,,(r)
ib(r) = ik(r)-id(t)

i c ( r ) = ic,,(r)-t&)

1433

From equations (1) to (8) one can make the following observations:
0 triplen harmonic currents, although present in each branch, are
not present in the line currents.
if triplen harmonic currents in each delta branch are removed
from their nns values (equation (4)) and scaled by 4 , the nns
line currents (equation (8)) ax obtained. This information is the
basis of the algorithm developed next.

3 Saiuration Curves
Each delta branch in Figure 1 is represented by a nonlinear inductance
in parallel with a nonlinear resistance (Figure 2). Their nonlinear
characteristicsare computed with the following assumptions' :

the v irA and A i[A c w e s (Figures 2b and 2c) are symmetric

with respect to the ongin (Rk and & are the slopes of segment k
of the v - ij-A and 2 i / A curves, respectively);
the transformer winding resistances and leakage inductances are
ignored.
The algorithm works then as follows:
1. For the constructionof the v - irA curve (Section 3.1):
compute the peak values of the branch current irAl, i r A p . .
point by point from the n*load losses.
2. For the " c t i o n

of the A - i/A curve (Section 3.2):

0 from the v - irA c w e , compute the nns values ZrA-m, remove

the triplen harmonic currents and obtain the resistive h e current
Ir-nns;

obtain the nns values I/-nodinear inductance fkom I,---,

the applied voltage V ;

Figure 2: ( a ) N on lin ear elem en Is;

(b) y - i,, curve;
(c) 1- i,,curve.

of the line current due to the

the total line current It--,
end

compute the peak values of the inductive current i/Al, i/&, ...
point by point iteratively.

3.1 Computali'onof the v - irA curve

Let us assume that the three-phase ndoad losses Pi, P2, ..., P m are
available as a function of the branch voltages VmSl, Vm2, ...,V-,,
(Figure 3).
If we assume that the applied voltage is sinusoidal, the conversion of
rms voltages to peak values (vertical axis of Figure 2b) is simply
vk =

for k = L Z . * * , m .

fi

(9)

Figure 3: V r m r - p o w e r loss curve.

1434

Due to symmetry reasons, voltage and current waveforms need only be

evaluated over 114 of a cycle. For a sinusoidal voltage v(6) = vk sin e,
pk can be written in the form2:

fl

6. if the absolute value of the difference id = I 1 - w

is less
than the specified tolerance, convergence is achieved. Otherwise,
the residue id is added to ig and the iterative process is repeated
fitnu step 2 onwards.

In general, convergence is achieved in less than 20 iteration steps with

ilA1 and initial guess
a tolerance of
For the f i linear segment in the v-rd curve, the current is
sinusoidal. The computation of the frrst peak current frAl is therefore
straightforward. Since P 1 = 3 v m ~ , I m ,in the linear case,

for every k 2 2. In some cases as many as 40 iteration steps may be

n-==Y.

From the second segment onwards (t2 2), equation (10) is evaluated
at each segment A , with only i r k being unknown, as explained in
more detail in Section 2.1 of (51. The computation of the peak current
i r b is done segment by segment, starting with i.4 and endug with
the last point i r b . Whenever a point i r 4 is found, its mu value is
calculated as well. A Fourier program (Appendix A) is used to campute the triplen harmonics (13. 19,...). They are removed from I r A m s
to obtain Ir-ms, which is needed later for the construction of the peak
flux peak current II ilA curve.

.c o r r e c t c u r v e

3.2 Computation of the

curve

The conversion of the peak branch voltages vk to flux 4 is again a rescaling procedure. Hence, for each linear segment in the II - i/A
me,
A&+

81

ZT
2

(12)

Let us now compute the peak values of the currents il4 through the
nonlinear inductance. First, the mu values of the line currents II+,,,~
are evaluated with

8
F i g u r e 4 : G e n e r a t i n g c u r r e n t waveform
from sinusoidal f l u x .

where the line current It+,,,, is available fhm the measurements, and
has already been computed h m the previous section.
where
For the first linear segment, the computation of iiA, is straightforward
since there are no harmonics yet. Therefore,

From the second segment onwards (k 2 2) the algorithm works

iteratively as follows (see R ilA curye in Figure 4):

1. guess ig;

2. with A((B)=Aksine, find 1/4 of a cycle of the distorted current

d*cally,

4 Casestudy
Consider a 50 Hz three-phase five-legged core type transformer. The
following infomation is known[6]:
1. rated power - 750 MVA (three-phase);
2. rated voltages - 420 kV/27 kV (line to line values);
3. wye connection on 420 kV side, delta connection on 27 kV side.

Table I shows the data from the positive sequence excitation tests
measured at the closed delta 27 kV side.

3. compute the nnr inductive branch current whose peak is ig ;

4. use a Fourier program to find the triplen harmonic inductive
currents in the delta branch (Appendix A);

5. remove the triplen harmonics from the estimated nns branch

current. Scale the estimated result Ilest by Js and compare it to
Il-in equation (13);

Table I: Three-phase transformer test data

I
It-m
I
P

vm

orv)
22.76
24.29
25.64
27.00
27.50
28.47
29.10
32.50

(A)

8.20
11.35
15.50
21.16
24.68
31.63
38.30
80.97

orw

206.21
240.26
270.13
311.00
323.03
355.48
385.41
560.00

1435

Vms is the m line to line excitation voltage, Ir-ms is the nns

excitation current (three-phase average) and P are the ndoad losses
(three-phasevalues).
Figure 5 shows two computed 2 - jib, c w e s (points connected by
straight line segments). One of them assumes that all odd harmonic
components of the current are present in the measured values and it is
therefore incorrect. The other curve is the correct one; it has been produced with the algorithm of Section 3.2. Figure 6 shows the piecewise
linear v - irA curve.
1.2

1.0 -

0.8 -

o'2

1
0.002

0.004

0.006

0.008 0.010

0.012

C u r r e n t ( p. U .)
Figure 5:

A-

Figure 5 shows that the correct curve goes deeper into saturation. For
the flu value of 1.2 P.u., there is a difference of approximately 14%
between the peak currents of the incorrect curve and the wrmt one.
In facf for transient studies, it is often neceSSafy to know peak fluxpeak current curves beyond that point. The usual way is to extend the
curve up to a value necessary for the study (this extension is
sometimes done with a straight line passing through the previous to
the last and the last point in the peak flux-peak current curve). This
may lead to larger errors for the peak values of the cun-ent since the
curves diverge as the flux goes up towards deep saturation. The
transformer magnetizing current would always be underedmated if
triplen harmonics inside the delta windings were not taken into
accoullt Errors can also affect the air core inductance value. A
parametric study was done considering typical values for the air core
reactances (0.2 p.u to 0.5 P.u.). For the study, a straight line segment
was connected to the last point of the correct curve of Figure 5 . Errors
on the slope of the saturation characteristics,for this case, are between
18% and 25% when the magnetizing current reaches the transformer
rated current. These differences may be important in ferroresonance
or inrush current studies.
The proper way to represent saturation effects is to look at the
transformer magnetic circuit. Saturation is related to fluxes in the wre
and tank. There tire several types of core construction. To model the
core reasonably well, one should know what kind of geometry the
transformer has. One should h o w if the core is three-legged, five
legged or shell-type. The knowledge of the zero sequence magnetizing
impedance is also required. It can be estimated for each type of core
coIlstruction [A. The saturation curve must be obtained for each leg
of the transformer and placed across the lowest voltage terminal.
In the iteration scheme of Section 3.2, harmonics up to the 99* order
were included. An average of 23.86 iteration steps was necessary (the
"um number of iterations was 34). In order to check the
numerical accuracy of the method, the m line currents were
errom
recomputed back from v irb, and 1 - jib, Curves.
were found to be very small (less than 0.001%).

iiacurvc.

5 Uncertainties

0.0005

Current (P.u.)
Figure 6 : V - i f A c u r v e .

0.0010

To model transformers exactly is very complicated, if not impossible.

The ferromagnetic properties of steel laminations are not well
understood yet. There is no theory available today that could predict
losses accurately as a function of frequency, even for low kequencies
[8,9,10]. There is a big discrepancy between calculated and measured
losses due to the very complicated domain structure of these materials,
There is also the further discrepancy #at transformer core loss pa
kilogram is always greater than the nominal loss of the steel as
measured in standard testers. The ratio between the transformer per
unit loss and the nominal or standard unit loss is the so-called
building factor of the core. Building factors usually range from 1.1 to
2.0. The extra loss is due to factors such as [l 11:
0 Non-uniform flux distribution due to differencein path lengths
among magnetic circuits;
0 Distortion of fluxwaveform due to magnetic saturation;
0 Circulating flux due to magnetic anisotropy of m a w ,
0 Flux directed out of the rolling direction;
0 Transverse flux between layers due to joints.

ne flux distribution in transformalaminationsisn0tl"eVtn

at low

fkqueames. For a Sinusoidal applied flux,the flux m each lamhation is, in

genaal,notsinwidal although the fluxcomponents addup to produce the

sinusoidal total flux [12. 131. These at-Mcertaintiesmironm

modelling. Amther major problem is the availabilityof data

1436

The core nonlinear elements, in this paper, are obtained from one
single frequency. According to this model. transformer no-load losses
would then be underestimated for frequencies below the rated
frequency and overestimated for frequencies above the rated
frequency. However, ferroresonance and inrush current studies show
that the power system is much more sensitive to variations of the
saturation curves than to variations of core losses.
Figure 7 is the measured positive sequence excitation characteristic of
a three-phase five-legged 75 kVA distribution transformer (secondary
side 120 V). The x-axis is the ratio of the no-load exciting current
I , (measured in each phase) to the no-load current at rated voltage
I-,a#d
(average value of rated current in all phases) for the same
transformer. The solid line represents the average curve.

1.4,-

---,- -

the current at any point in between A and B. For instance at point B,

the open-circuited capacitancecan be obtained by:
Copon
=

+.
m AB

Table II shows *corrupted*fluxcurrent points obtained using Figure

Sa as input data.
The following information is needed:
peak current at turning point A = 1.240726 A;
peak current at tUming point B = 0.123851 A;
peak flux at turningpoint B = 0.189066 Vs.
Then, the open circuit capacitancereferred to the 120 V side is
1.240726- 0.123851 ~ ~ 4 1 . 5 6 ~ .
( 2 z x 60)2x0.189066
is referred to the primary side (14400 V),
Copen
=41.56~(120/14400~~2.89~F.
Copen =

7 - 7

If.,C

Table II: Corrupted fluxcurrent curve.

Flux Linkage(V.s)
0.000000

Cment(A)
0.0000000

0.040327

0.426281

lexcdexc-rated

Figure 7: V,,

- 1,characteristicfor three-phase transformers

There are discrepancies below the rated voltage, but the curves are
very close as the transformer goes into deep satmition.
I, curve shown in Figure 8 was obtained from a recently
The V ,
manufactured 10 kVA single-phase distribution transformer . The nns
current at 90% of the rated voltage is smaller than the m s current at
50% of the applied voltage. Here, the exciting current in the
unsaturated region is very much affected by stray capacitances. As the
transformer goes into deep saturation, stray capacitances aEkct very
little the computation of saturation curves.

0.059046

0.747657

0.075777

0.905318

0.095659

1.1933 16

0.114790

1.240726

0.151553

0.815392

0.169560

0.396883

0.189066

0.123851

0.207823

0.162665

To represent the transformer exactly, one should represent saturation,

hysteresis and eddy currents in the core as well as eddy currents and
stray capacitances effects in the windings. Modern power transformer
cores have very low losses and saturation is the predominant effect. In
this paper the main focus is on estimation of the saturation curves of
transformers.

4 Conclusions
An approach for the computation of instantaneous saturation curves of
rms C u n m t (A)

(a)
(b)
Figure 8 a)V, - L curve for a recently manufactured transformer.
b) Corrupted fluxcurrent curve (not to scale).
A crude estimation of the saturation curves and the open-circuit
capacitance can be found iteratively. First, the algorithm generates the
peak fluxcurrent characteristic "corrupted" by stray capacitance
effects as shown in Figure 8b. The peak current is decreasing in the
region between points A and B. The inductance is very high in this
region. It can be assumed that it is infinite between turning points A
and B. The peak cucrent 1. through the open circuit capacitance C,
is given by the difference between the current at turning point A and

deltaannected transformers has been presented. It uses positive

sequence excitation test data as input, and is suitable for situations in
which the tests are performed with a closed delta.

Once the v irA and R - ilA curves have been obtained, they can then
be used to model the excitation branch of transformers in transient and
harmonic studies.

7 Acknowledgments
The authors would like to thank the r e v i m for their valuable
suggestions.Also. the authors would like to thank Powatech Labs Inc. for
providmg the t"mtest data of Section 5.
The financial support of Mr.Washington Nevw from Univmidade
Federal da Paraiba, Campina Grande PB Brazil, and from The
University of British Columbia is gratefully acknowledged.

1437

For segment R = 2 ,

8 References
1. S.N. Talukdar, J. K. Dickson. R. C. Dugan, M. J. SprinZen, C. J.
Lenda, On Modeling T m f o n n e r and Reactor Satumtion
Chamcteristicsfor Digital and Analog Studies, IEEE Trans. on
PAS, vol. PAS-94.1975, pp. 612621.
2. S. Prusty d M.V. S.Rm,A Dimxtp'ecewise Linean.zedAppd
toGmveli musalumtionchomcrenstr
. 'cto Instantaneous Wmtim
CUM,IEEE Trans. Mag., vol. Mag-16, NO. 1, J ~ ~ M1980,
I Y p ~ 1.5 6
160.
3. L. 0. Chua and K. A. StrO"oe, Lumped Circuit Models for
Nonlinear Inductors Exhibiting Hysteresis Loops, IEEETrans. on
Circuit Theory, vol. CT-17, NO.4, NOV.1970, pp. 564-574.
4. J. G. Santesnases,J. Ayah, S. H Cachero, Analytical Approximation
of Dynamic Hysteresis Loops and its Application to a Series
FerroresonanceCircuit, Proc. IEE 117,No. 1, Jan 1970, pp. 234-240.
5. W. L A N e v e H. W. Donrmel, On Modelling Imn Core
Nmlinamities. IEEE T d o n s on Power Systems, vol. PWRS-8.
h4ay 1993,pp. 417425.

Substituting

r - -!-=

'-4

i,A (@=

b,,sinne

R I

for n=l,3,--.,

with

(A. 1)

above equation, the

+4CO4)1

4-1
-rl~~~n,ef-l~~k+-C~n~j-l

r, = i
Lk '

'm,= $of

sin[(n-lP,I

13. A. Basak and AAA Qader, FtI"enta1 and Hanncnic Flux

Behaviour In a IOOkVA Disbibutim Tmnr/onner Core, IEEE Trans.
onh4ag.,Vol. Mag-l9,NoS,Septanber 1983,pp. 2100-2102.

Although only the computation of triplen harmonics of the current is

needed, it is appropriate to show the derivation of all odd harmonic
components that may be produced by saturation curves during
excitation tests. The equations below are developed for nonlinear
inductances. For nonlinear resistances, one needs to replace 1 by V
and L by R accordingly.
Consider the piecewise nonlinear inductance of Figure 2c. For a
sinusoidal flux A(@) = sin(@, the current can be written in a Fourier
series form containing odd harmonics only. So,

1 .
r, = mto the

4 = r24 + $rI - r2 x+z

fundamentalcurrent(n=l) anditsharmonics(n23)are:

6. ATP Rule Book, Leuven EMTP Center (LEC), Section XIXG,

Revision July, 1987.
7. H. W. Dommel, Electromagnetic Tmnsients Progmm Reference
Manual, Section 6, Department of Electrical Engineering-The
University of British Columbia, Vancouver, 1986.
8. G. E. Fish, So) Magnetic Materials, F'mxdqs of the IEEE,
Vol. 78, No. 6, June 1990, pp. 947-972.
9. T. H. O'Dell, Fe"amtoC+"ics, Chapters 1 and 5, Mcmillan
PressLtd,LonQns 198L
and
10. G. Hener and H. Hilzinger, Recent Developnents in Sop A h p d c
b,,=L?{-,
Matetials, physlat Scripta,Vol. T24,1988, pp. 22-28.
11. T. Nakata, Numerical Analysis of Flux and Loss Distribution in
1-2
Electrical Machinety, (Invited Paper), IEEE Trans. on Magnetics,
Vol. Mag 20, NO. 5 , September, 1984.
where
12. F. J. Willcins and A E. Drake, Meann"t and Intmpretation of
Ponw Larw in Elecbiml Sheet Steel. Proc. IEE Vol. 112, NO.4,
April, 1%5, pp. 71-785.
and

Appendix A Computation of HmmOniC Componenis

!!h and

gm(n.w =

qn-1)

-i s i n 20,)
- sin[(n+l)811
2(n+l)

- Washington L. A. Neves was bom in Brazil on March I, 1957. He received the B. Sc. and M. Sc. degrees in Electrical Engineaing h m
Universidade Federal da Paraiba in 1979 and 1982 respectively. From
1982 to 1985 he was with the Departmat of Electrical Engineerkg of
Faculdade de Engenharia de Joinville, Santa Catarina, Brazil. Since
November 1985, he is with the Department of Electrical hguleering
of Universidade Federal da Paraiba, Campina Grande-PB Brazil. He
is currently a Ph. D candidate at the University of British Columbia,
Vancouver, Canada.

Hennunn K Dommel was born in Germany in 1933. He received the

Dip1.-Ing. and Dr.-Ing. degrees in Electrical hgu~eerhg from
Technical University, Munich, Germany in 1959 and 1962
respectively. From 1959 to 1966 he was with the Technical University
Munich, and from 1966 to 1973 with Bonneville Power
Administration, Portland, Oregon. Since July 1973 he has been with
the University of British Columbia in Vancouver, Canada. Dr.
Dommel is a Fellow of IEEE and a registered professional engineer in
British Columbia, Canada.