See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/229005714

Modeling of nonlinear and hysteretic iron-core

inductors in ATP

Article

CITATIONS READS

3 462

2 authors:

Nicola Chiesa Hans Kristian Høidalen

Statoil ASA Norwegian University of Science and Techno…
33 PUBLICATIONS 180 CITATIONS 98 PUBLICATIONS 597 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

ProSmart - Power system protection in a smartgrid perspective View project

All content following this page was uploaded by Hans Kristian Høidalen on 25 March 2014.

The user has requested enhancement of the downloaded file. All in-text references underlined in blue are added to the original document
and are linked to publications on ResearchGate, letting you access and read them immediately.
Modeling of nonlinear and hysteretic iron-core
inductors in ATP

Nicola Chiesa Department of Electric Power Engineering

NTNU, Norway O.S. Bragstadspl. 2E
N-7491 Trondheim, NORWAY
Voice tel. +47 735 94232
[email protected]

Hans Kristian Høidalen Department of Electric Power Engineering

NTNU, Norway O.S. Bragstadspl. 2E
N-7491 Trondheim, NORWAY
[email protected]

Abstract - A crucial element of transformer models for transient simulation is the

representation of the core. The modeling of non-linear hysteretic inductor required to
properly represent a transformer core is a challenge in ATP. The simulation of transient such
as inrush and ferroresonance requires a correct handling of nonlinear and frequency
dependent losses, accurate hysteresis loop representation, possibility of flux initialization,
and a proper automatic initialization by disconnection transients. In addition to the
investigation of standard non-linear inductor models of ATP, an advanced hysteretic model
based on the Jiles-Atherton theory is implemented in MODELS and tested. The comparison of
the models shows several weaknesses and the need of further investigations. A practical table
summarizes the main quality and weaknesses of each model together with recommendations
useful for the choice of the most appropriate model.

Keywords: test report, nonlinear inductor, hysteresis, nonlinear losses, frequency dependent
losses, residual flux, Jiles-Atherton.

1 Introduction

A transformer iron-core and any other ferromagnetic nonlinear inductance modeled in ATP
suffer of low accuracy. The purpose of this paper is to investigate the different possibility that
ATP offers for modeling nonlinear inductances. The main limitation is the lack of input data;
advanced models require detailed measurements for the estimation of parameters, while
standard test report is usually the only source of data. An accurate representation of the losses
(nonlinear and frequency dependent) is also a required feature of an accurate model.

The first part of this paper addresses a method for dealing with the lack of data by curve
fitting. The second part of the paper compares different ATP nonlinear inductor model with
focus on losses, shape of the hysteresis loop, and residual flux initialization. An advanced
nonlinear inductor model based on Jiles-Atherton theory is implemented in MODELS and
tested. The response of the different models to a deenergizarion-reenergization operation is
compared.
2 Extension of the test report data

A core model is usually based on the transformer open-circuit test measurement.

Conventionally data at rated excitation is reported. Additional points are required in order to
characterize the nonlinearity of the core and built a piecewise linear characteristic.
Transformer manufacturers most commonly provide additional data for 90% and 110%
excitation levels. Table II in Appendix shows the test report data for a 290 MVA step-up
transformer. In this case five results of the open-circuit test are provided, spanning from 75%
to 106.25% of the rated voltage. The data provided by the open-circuit test report is sufficient
when the purpose of the transient simulation does not involve heavy saturation of the
transformer core. Note that the rms V-I characteristic specified in a test report has to be
converted to a peak Ȝ-i characteristic by use of the ATP CONVERT/SATURA routine [1,2]
or a more advanced method that that can take into account the three-phase coupling of
transformers [3,4].

Open-circuit test report seldom reports data above the 110% of excitation level; testing a
power transformer in heavy saturation requires a large and stiff source, with rated power
comparable to the power of the transformer. Indeed, no testing facility has the capability of
performing such test on large units. Thus, the extension process is not straight forward.

Fig. 1 shows how a piecewise nonlinear characteristic can be extended by linear extrapolation
and curve fitting (saturation curve points are reported in Table III in Appendix). The linear
extrapolation method assumes a constant slope of the saturation curve after the last specified
segment of the piecewise nonlinear curve. This approach became doubtful when the last
points of the piecewise nonlinear curve lie in the 100% to 110% excitation level range: the
transformer has not reach the complete saturation during the open-circuit test and a linear
extension of the curve will result in a severe underestimation of current for any excitation
level above the last specified point. Curve fitting allows the definition of additional artificial
points of the saturation characteristic, such that new segments can be added to the piecewise
nonlinear curve. While linear extrapolation is the most commonly used, the curve fitting
approach should be preferred.

90

80
Flux-linkage [Wb-t]

70

60
Test Report
Extended Points
50
Frolich Equation
Linear Extrapolation

40
0 200 400 600 800 1000

Peak Current [A]

Fig. 1. Extension of the saturation curve from test report data.
Without specific measurements it is difficult to verify how precisely the curve fitting method
can represent the saturated region and mainly relay on the fitting function used. Five different
fitting functions are proposed:
λ=
i
a +b⋅ i + c⋅ i
(1)

λ = a ⋅ tanh ( b ⋅ i ) + c ⋅ i (2)

i = a ⋅ sinh ( b ⋅ λ ) + c ⋅ λ (3)

λ = a ⋅ ib + c ⋅ i
⎧ λ = L∞ ⋅ ( iE + iM )
(4)

⎪
⎨ λSAT a ⋅ iE + iE 2
⎪ = ⋅
(5)
⎩ L∞ c + b ⋅ iE + iE 2
iM

where Ȝ represent the flux-linkage.

• (1) is a modified version of the Frolich equation, [5].
• (2-4) are equation hard-coded in ATP as user-supplied FORTRAN, [2].
• (5) is a modified version of the equaton proposed by Annakkage, [7].
Fig. 2 shows the curves obtained from the fitting of the open-circuit test report data, while the
corresponding parameters are reported in Table IV in Appendix. Beside (2), the rest of the
curves give similar expansion of the saturated region. While (2) and (3) give poor agreement in
the knee area, (5) shows the best fitting thanks to the higher number of parameters. (1) and (5)
should be preferred since they allow a better controll of both the shape of curve in the knee
area and the saturation level.

Despite the fact that more advanced techniques exist [8], curve fitting is a simple method that
require no more than the standard test report data and allow the extension of the saturation
characteristic. The authors believe that curve fitting can increase the accuracy of transient
simulations like inrush and ferroresonance, and any other study that demand an accurate
model of the heavy saturated area of transformers and iron-core devices (the flux in the core
can be as high as three time the rated flux during an inrush transient).

120

100
Flux-linkage [Wb-t]

80
80
60
60 (1)
(2)
40 40 (3)
(4)
20 (5)
20
Test Report
0
0 10 20 30 40 50 60
0
0 200 400 600 800

Peak Current [A]

Fig. 2. Fitting of the saturation curve

3 Nonlinear and hysteretic inductor models in ATP

• type-98, pseudo nonlinear reactor;

The default nonlinear inductor models available in ATP are:

• type-93, true nonlinear inductance;

• type-96, pseudo nonlinear hysteretic reactor;
• User-supplied FORTRAN.
While the first three mentioned types are fairly common among users, the user supplied
FORTRAN type is not widely known. The same saturation curve can be used for both type-98
and type-93. However, if the saturation curve is intended for a type-98, a maximum of five
breakpoints is recommended to avoid numerical tracking problems, [2]. The solution method
used for type-93 ensures the operation on the correct slope of the characteristic so that many
breakpoints can be specified. Both these nonlinear inductor model are single valued and the
core losses have to be taken into account with a parallel resistor (R-L representation). Type-
96 is a hysteretic reactor, thus with direct handling of the core losses. The hysteresis loop data
required by type-96 can be generated in ATP with two routines: HYSDAT and HEVIA, [2,9].
The first routine rescale a hard-coded ARMCO Mh oriented silicon steel characteristic based
on saturation current and saturation flux values specified by the user. The second routine uses
a piecewise nonlinear single-valued characteristic and active core losses to create a hysteretic
characteristic suitable for the type-96. The input data for the HEVIA routine match the
information available in a test report and has been preferred in this paper. User-supplied
FORTRAN can be easily used in ATPDraw with user supplied library as shown in Fig. 3.
They are accessible with special call to type-93 card and special flags -333777, -444777 and
-555777 respectivelly for (2-4). Example of their use and how to define the equation parameter
can be found in subcase of BENCHMARK DC-7.

As reported in [2] users may be able to implement their own user supplied FORTRAN model,
though this reqiures ATP executable to be recompiled. For this reason the MODELS language
has been chosen as a development platform for a hysteretic model based on the Jiles-Atherton
theory. Among the numerous versions of the Jiles-Atherton model present in literature, the
variant proposed by Chandrasena [10] has been preferred due to the compatibility with the
ATP solution method (trapezoidal rule of integration) and the handling of eddy current losses.
The model proposed by Chandrasena has been slightly modified and the set of equations has
been described as function of electrical parameter λ-i instead of magnetic parameter M-H; the
new formulation fits better with ATP implementation and with the estimation of parameters
from test report data.

X555

LIB

C User-supplied fortran follows. This is a regular Type-93 NL inductor until

C the time-step loop. Note 3-card characteristic, followed by "9999" bound.
C The characteristic parameter usage is: psi = a * i**B + c * i
93X555 .08 35.0 1
-555777. 55.FORTRAN { -555777 = flag; a = 1st param
.07 0.0FORTRAN { b = 2nd of 3; c = 3rd of 3 params
1.0 1.0FORTRAN { Dummy third card to protect card 2
9999 { End of user-supplied fortran (see request in cols. 33-39)
/INITIAL { initialization card, residual flux can be sprcified
4X555 0.0 0.

Fig. 3. Use of User supplied FORTRAN in ATPDraw.

4 Capabilities and limitations
The behavior of the nonlinear and hysteretic inductor models presented in the previous

• no-load current;
session is compared with focus on:

• Active losses;
• shape of the hysteresis loop;
• residual flux initialization;
• response to a deenergizarion-reenergization transient.
Fig. 4 shows the basic core models investigated in this paper. The models can be divided in
R-L and hysteretic. In R-L models the losses can be represented by either a linear or a
nonlinear (type-99 or -92) resistor, while the saturation curve is modeled with a nonlinear
inductor (type98, -93 or user-defined FORTRAN). Hysteretic models include losses and
saturation in the same model; the only available hysteretic model in ATP is the type-96, but
additional models can be defined as type-94 elements.
Lin.Res. or type-99 Lin.Res. or type-92 Lin.Res. or type-92 type-96 type-94 elem.
and type-98 and type-93 and user-defined FORTRAN Hevia Jiles-Atherton
LIB
TYPE
R(i)

R(i)

or or or R(i) 94 ss

H
Fig. 4. Nonlinear and hysteretic inductor models in ATPDraw.

4.1 No-load current, losses and shape of the hysteresis loop

The no-load current is the sum of the magnetization current and the current due to the core
losses. Core losses are an important aspect to take into accout; their nonlinear and frequency
dependent nature makes the implementation in a transient simulation program not straight
forward. Core losses can be devided in two parts: hysteresis losses and losses due to eddy
current in laminations. Hysteresis losses per cycle are frequency indpendent, but they are
affected by the excitation level. Eddy current losses vary non linearly with frequency.

The most used and straight forward way to take losses into account it to use a linear resistor
that match the losses at rated condition. A nonlinear resistor may also be used to represent the
nonlinear behaviour. Finally, hysteretic models directly include hysteresis losses, but
eventually require additional handling of eddy current losses.

Fig. 5 contains the comparison between simulated and test report active losses. When a linear
resistor is used, it is common to fix the resistance value to match the rated losses. This gives
overestimation of losses for lower level of excitation and underestimation for higher level of
exitation. Nonlinear resistors (type-99 and type-92) can be used to have an accurate matching
of the losses at the terminal. Type-96 hysteretic inductor gives a result similar to the linear
resistor up to the second last point; for value higher than this exitation level, the characteristic
became single valued and the losses remain constant (Fig. 7c). The accuracy of the Jiles-
Atherton model in fitting the losses is quite poor; this is probably due to a lack of data for the
estimation of the correct model parameters. However, also [10] shows some difficulties in a
proper matching of core losses with this model; the Jiles-Atherton theory is quite mature and
widely axepted, but has room for improvement in losses fitting.

Fig. 6 provides the comparison between simulated and test report rms no-load current. The
overall agreement is quite satisfatory. Both nonlinear inducance and losses representation are
taken into account here; better agreement can be achieved with a better curve fitting and
losses modeling.
 80 60
Test Report Test Report
Linear Resistor Type-98/93
70 Type-96 50 Type-96
Type-99/92 -333777
Jiles-Atherton -444777
60 -555777
Active Loss [kW]

40

Rms Current [A]

Jiles-Atherton

50 30

40 20

30 10

20 0
11 12 13 14 15 16 17 18 11 12 13 14 15 16 17 18

Rms Voltage [kV] Rms Voltage [kV]

Fig. 5. Active losses comparison. Fig. 6. No-load current comparison.

Active losses and rms values only give an average indication of the behavior over one period.
The analysis of instantaneous quantities is required in order to better understand the
differences of the various models. The shape of the hysteresis loops as shown in Fig. 7 is
suitable for this purpose. The hysteresis loops are found by stepping up the excitation voltage
according to the test report and calculation of the corresponding flux linkage in TACS. When
an equation is used as a replacement for a piecewise characteristic, the hysteresis loops
become smoother (compare Fig. 7a with 7b, and Fig. 7c with 7d). True hysteretic models
(Fig. 7c and 7d) have a better relocation of the losses in the full area of the hysteresis loop,
while models that represent losses with a resistor in parallel to a nonlinear inductor (Fig. 7a
and 7b) have a tendency to underestimate the losses in the vicinity of the knee area (this can
be seen from the narrow width of the hysteresis loops in this area of the curve).
80 80

70 70

60 60
Flux-linkage [Wb-t]
Flux-linkage [Wb-t]

50 50

40 40

30 30

20 20

10 10

0 0
0 20 40 60 80 100 120 0 20 40 60 80 100 120
Current peak [A] Current peak [A]

a. Type-93 & Type-92 b. -555777 & Type-92

80 80

70 70

60 60
Flux-linkage [Wb-t]
Flux-linkage [Wb-t]

50 50

40 40

30 30

20 20

10 10

0 0
0 20 40 60 80 100 120 0 20 40 60 80 100 120
Current peak [A] Current eak [A]

c. Type-96 d. Jiles-Atherton

Fig. 7. Hysteresis loop function of the excitation level.

 80 80

70 70

60 60
Flux-linkage [Wb-t]

Flux-linkage [Wb-t]
50 50

40 40

30 30

20 20

10 10

0 0
-10 0 10 20 30 40 50 -10 0 10 20 30 40 50
Current peak [A] Current peak [A]
a. Linear resistor b. Nonlinear resistor (Type-92)

Fig. 8. Effect of the nonlinear resistor.

The area of the hysteresis loop represents the losses per cycle and provides information on the
losses distribution along a cycle. Fig. 8 shows how the losses distribution changes when a
nonlinear resistor is used instead of a linear one. While a nonlinear resistor models the losses
at the terminal more accurately (Fig. 5), the losses accumulate at the base of the hysteresis
loop. A nonlinear resistor characteristic is defined as a v-i curve: the differential resistance
became lower (higher losses) as the voltage level increases. The losses reach their highest
value at voltage peak, which correspond to the zero of the flux-linkage and to the base of the
hysteresis loop.

Fig. 9 presents how the hysteresis loops behave as function of frequency. Each model has
been tested at values of frequency 25, 50, 200 and 400 Hz. As the frequency increases, also
the applied voltage has to increase to obtain a constant flux density in the core (V/f constant).
Therefore, when a parallel linear or nonlinear resistor is used, it results in an over estimation
of the losses. A frequency dependent resistor may be used to compensate this undesired
behavior. The type-96 hysteretic inductor does not show modification of the hysteresis loop
as the frequency changes. This is the expected behavior as it only takes into account
hysteresis losses (hysteresis losses per cycle are frequency independent). The original Jiles-
Atherton model is also frequency independent. However, the implemented model allows to
take into account classical and excessive eddy current losses in addition to the hysteresis
losses. The curves illustrated in Fig. 9d have only a qualitative meaning since no information
was available to accurately calculate the parameters that characterize the frequency
dependency behavior of the model.

4.2 Residual flux initialization

The residual flux is the flux trapped in a magnetic core after a transformer is disconnected
from the network. The value of the residual flux greatly affects the peak of inrush current that
develop at the subsequent re-energization. The ability to initialize an inductor model is of
great value to test different re-energization scenario. The type-93 element that has a special
initialization card; user-defined FORTRAN elements can advantage from the same
initialization card as shown in Fig. 3. The type-96 hysteretic inductor and the Jiles-Atherton
based model provide the option to set a residual flux value. The type-98 nonlinear inductor
model lacks of this feature. A possible workaround to create an initial step in the flux value
(in order to resemble a residual flux) is to use a DC source to create a voltage impulse. The

λRES
DC source has to be active for one time step and have amplitude:
VPEAK = (6)
timestep
 80 80

60 60

40 40

Flux-linkage [Wb-t]
Flux-linkage [Wb-t]

20 20

0 0

-20 -20

-40 -40

-60 -60

-80 -80
-100 -80 -60 -40 -20 0 20 40 60 80 100 -100 -80 -60 -40 -20 0 20 40 60 80 100
Current peak [A] Current peak [A]

a. Linear resistor b. Nonlinear resistor (Type-92)

80 80

60 60

40 40
Flux-linkage [Wb-t]

Flux-linkage [Wb-t]
20 20

0 0

-20 -20

-40 -40

-60 -60

-80 -80
-60 -40 -20 0 20 40 60 -60 -40 -20 0 20 40 60
Current peak [A] Current peak [A]

c. Type-96 d. Jiles-Atherton

Fig. 9. Hysteresis loops function of frequency (V/f constant).

80 80

60 60

40 40
Flux-linkage [Wb-t]

Flux-linkage [Wb-t]

20 20

0 0

-20 -20

-40
-40

-60
-60
-50 0 50 100 150 200 250
-50 0 50 100 150 200 250 300 350
Current peak [A]
Current peak [A]
a. User-defined FORTRAN (and similarly type-93 and -98), b. Type-93 (and similarly user-defined FORTRAN), internal
initialized with DC source initialization
80 80

60
60

40 80
40
Flux-linkage [Wb-t]

Flux-linkage [Wb-t]

60
20
20 40
Flux-linkage [Wb-t]

20
0
0
0

-20 -20
-20
-40
-40 -60
-40
-80
-60 -1 -0.5 0 0.5 1
-60
Current peak [A] 4
x 10
-50 0 50 100 150 200 250 300
-20 0 20 40 60 80 100 120 140 160 180
Current peak [A]
Current peak [A]
c. Jiles-Atherton, internal initialization d. Type-96, initialized with DC source (sub-figure: internal
initialization)

Fig. 10. Hysteresis loops alteration due to residual flux initialization.

The residual flux acts as a DC offset of the flux and the expected behavior is the saturation of
the core for only either the positive or negative half period, depending on the sign of the
residual flux. Fig. 10 shows the effect on the hysteresis loops when the models are initialized
with a residual flux 10% of the rated flux. All the single-valued nonlinear inductors model of
ATP can be initialized either with an initialization card or with a DC source and perform as
expected and illustrated in Fig. 10a and 10b. The Jiles-Atherton model also reproduces the
expected behavior both if initialized with its internal parameter (Fig. 10c) or by a DC source.
Finally, the type-96 does not operate correctly if internally initialized; better result is obtained
employing the DC source initialization technique (Fig. 10d).

4.3 Deenergizarion-reenergization transient

Each model is tested with the ATPDraw circuit presented in Fig. 11 to verify its capability to
automatically initialize the residual flux after a disconnection transient. The simulation first
generates a ringdown transient followed by an inrush transient. A ringdown transient is a
natural LC response that appears as the stored energy dissipates whenever a transformer is
deenergized, [11]. Residual fluxes are due to the remnant magnetization of the core, after a
transformer has been deenergized. The residual flux pattern is mostly unknown or not known
precisely due to the complexity of the ringdown transient itself: an advance and accurate
model is required to correctly simulate a ringdown transient.
top = 13 ms 0.15 Ohm

16 kV tcl = 35 ms
50 Hz Core Model
100 nF

Fig. 11. Deenergizarion-reenergization circuit.

Fig. 12 presents the results of the performed simulations. Fig. 12a shows the common
behavior of most of the nonlinear ATP inductor models: the whole energy dissipates in the
parallel R-L elements so that no energy remains trapped in the core and no residual flux is
generated. The result of any simulation performed with such models is independent of the
disconnection instant since the remanent magnetization of the core will rapidly go to zero.
The user-defined FORTRAN model -555777 gives the impression to reach some value of
residual flux even if it is part of the parallel RL category. A longer duration of the
disconnection transient has confirmed that this residual flux is not a characteristic of the
model, but it is artificially generated by the simulation algorithm: it will eventually decay to
zero if the reenergization is postponed for a sufficient time. It is difficult to explain the reason
of this behavior without information about the implementation of this model.
True hysteretic models should be used to obtain a better representation of the ringdown
transient, as demonstrated in Fig. 12c and 12d. In this case the residual flux stabilizes at a
constant value after few milliseconds from the disconnection operation.

When the switch closes, the response of the four models to a reenergization transient is
compared in Fig. 12. In order to have a more realistic comparison of the inrush transient a
residual flux is artificially created by applying a voltage impulse to the Type-93 model. In this
way a similar initial condition of the flux is ensured for all the models. The response of the
model to the reenegization is quite dissimilar due to the different characterization of the
saturation curves in complete saturation. Type-93 and type-96 model uses linear extrapolation
of the last defined segment of the characteristic, while user-defined FORTRAN relay on
curve fitting. Thanks to the test report extension procedure presented in the first part of this
paper, the type-93 model performs similarly to the -555777 model; without a preliminary
curve fitting the type-93 model would have given much lower current values. In addition to
curve fitting, the Jiles-Atherton model defines an air-core inductance (proportional to ȝ0) for
excitation above the saturation level. This results in a much higher inrush current. The decay
of the flux DC offset and the inrush current are influenced by the voltage drop on the series
resistor, thus a faster decrease occurs for higher inrush currents.

6000 100 6000 100

3000 50 3000 50

Flux-linkage [Wb-t]

Flux-linkage [Wb-t]
Current [A]

Current [A]
0 0 0 0

-3000 -50 -3000 -50

0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
Time [s] Time [s]

a. Type-93 (and similarly type-98 and user-defined b. User-defined FORTRAN -555777

FORTRAN)
4
x 10
2500 100 2 100

1250 50 1 50
Flux-linkage [Wb-t]

Flux-linkage [Wb-t]
Current [A]

Current [A]

0 0 0 0

-1250 -50 -1 -50

0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
Time [s]
Time [s]

c. Type-96 d. Jiles-Atherton

Fig. 12. Ringdown and inrush transient simulation.

5 Conclusions

The purpose of this paper has been to compare the features and the capability of available
inductor models and a possible way to implement advanced hysteretic models in ATP.
Methods for overcoming limitations due to limited amount of data and flux initialization have
been presented. An optimum model for each circumstance cannot be selected due to the lack
of comparison with measurements and several weaknesses of each model. The two major
problems that need to be recalled are the faulty flux initialization of the type-96 hysteretic
inductor and the non suitable R-L representation for deenergization transient simulations.
Table I summarizes the most important quality of each model and may be helpful for
choosing the correct model to use in a simulation.

The outcome of this investigation suggests that ATP lacks of an advanced hysteretic model.
The Jiles-Atherton model tested has good potential, but more work has to be done to refine
the losses model and the estimation of parameters from standardly available data. In addition
to the Jiles-Atherton model, another widely accepted model that may be worth to investigate
and test in ATP is the Preisach model, [12]. Future work will include the extension from a
single-phase equivalent model to a topologically correct three phase core model where several
hysteretic inductors need to interact together.
 TABLE I
MODELS COMPARISON.

Parallel R-L description True hysteretic model

User-defined
Nonlinear inductance: Type-98 or -93 Type-96 Type-94, Jiles-Atherton
FORTRAN
Not compatible with
Any, only required for HEVIA & HYSDAT
(5), but any can be used
Fitting equation: test report extension (2-4) & possibly other. routines (possible tuning
by the J-A model.
procedure. of the slope of the last
segment).
Losses model: Parallel linear or nonlinear resistor. Included in the model. Included in the model.
Poor, constant losses Poor, difficult parameters
Linear resistor is commonly used; very good with
No-load losses: after the second last estimation. Requires
nonlinear resistor.
point. model improvement.
Piecewise, bad losses Smooth, bad losses Piecewise, ok losses Smooth, good losses
Shape of hysteresis loop:
distribution. distribution. distribution. distribution.
Hysteresis. Also eddy
Core losses frequency current losses but require
Bad, require frequency dependent resistor. Only hysteresis losses.
dependency: estimation of extra
parameters.
Built-in initialization Ok with both built-in
Internal initialization for type-93 and user-defined
Flux initialization: does not work. Requires initialization and DC
FORTRAN (type-98 requires DC source workaround).
DC source workaround. source workaround.
Flux goes to zero (very
Residual flux after OK, accuracy to be tested
Flux goes to zero slow decay for -555777, OK, but not accurate.
deenergization: with measurement.
“artificial” residual flux).
Poor, linear extrapolation
Poor, linear extrapolation
after the last segment
after the last segment.
(Can be improved with
Complete saturation: Can be improved with OK, curve fitting. OK, curve fitting.
manual tuning of the
test report extension
slope of the last
procedure.
segment).
Possible but nor
Test report, CONVERT straightforward from test
Test report & Test report & Curve
Parameters estimation: & HEVIA (or HYSDAT) report. (Better from
CONVERT routine. fitting.
routines. measured hysteresis
loops).
+++++ (uses MODELS,
Computational time: + ++ (+++ for -555777) ++
not hard-coded).

Appendix

TABLE II
GENERATOR STEP-UP TRANSFORMER TEST-REPORT.

Main Data [kV] [MVA] [A] Coupling

HS 432 290 338 YN
LS 16 290 10465 d5
Open-circuit E0 [kV,(%)] [MVA] I0[%] P0[kW]
LS 12(75) 290 0.05 83.1
14(87.5) 290 0.11 118.8
15(93.75) 290 0.17 143.6
16(100) 290 0.31 178.6
17(106.25) 290 0.67 226.5
Short-circuit [kV] [MVA] ek,er[%] Pk[kW]
HS/LS 432/16 290 14.6, 0.24 704.4
 TABLE III
LOSSES AND SATURATIN CURVES.

Test Report Extended Points

Current peak Voltage peak Current peak Flux-linkege Current peak Flux-linkage
[A] [kV] [A] [Wb-t] [A] [Wb-t]
3.264 16.970 2.756 54.018 3.540 49.210
4.240 19.799 14.403 63.022 22.755 66.720
5.039 21.213 24.107 67.523 53.846 72.491
6.350 22.627 50.171 72.025 135.244 77.033
8.054 24.042 118.773 76.526 348.344 80.260
- - - - 906.250 82.424

TABLE IV
PARAMETERS OF THE FITTING EQUATIONS.

Equation a b c λSAT L∞
Mod. Frolich 7.68e-4 1.16e-2 1.6e-2 - -
-333777 71 0.1 0.04 - -
-444777 3e-9 0.33 0.15 - -
-555777 55 0.7 0 - -
Mod. Annakkage 27.82 39.75 12.60 83 1.044e-3

References
[1] Dommel, H. W.; et.al.: Electromagnetic Transients Program Reference Manual (EMTP
Theory Book,. Portland, OR: Prepared for BPA, Aug. 1986.
[2] ATP Rule Book. Leuven EMTP Center, Jul. 1987.
[3] Chiesa, N.; Høidalen, H. K.: On the calculation of flux linkage/current-characteristic or
Δ-coupled transformer windings, EEUG Meeting 2005 European EMTP-ATP
Conference, Warsaw, Poland, Sep. 2005.
[4] Neves, W. L. A.; and Dommel, H. W.: Saturation curves of delta connected transformers
from measurements, IEEE Trans Power Delivery, vol. 10, no. 3, pp. 1432 – 1437, 1995.
[5] Chiesa, N.: Power Transformer Modelling: Advanced Core Model, M.SC. thesis,
Poltecnico di Milano, Italy, 2005.
[7] Annakkage, U.D.; McLaren, P.G.; Dirks, E.; Jayasinghe, R.P.; Parker, A.D.: A current
transformer model based on the Jiles-Atherton theory of ferromagnetic hysteresis, Power
Delivery, IEEE Transactions on , vol.15, no.1, pp.57-61, Jan 2000.
[8] Abdulsalam, S.G.; Wilsun Xu; Neves, W.L.A.; Xian Liu: Estimation of transformer
saturation characteristics from inrush current waveforms, IEEE Transactions on Power
Delivery, v 21, n 1, Jan. 2006, p 170-7
[9] Prikler, L.; Høidalen, H. K.; ATPDraw varsion 3.5 for Windows 3.1x/95/NT: user's
manual, SINTEF Energy Research, Trondheim, Oct. 2002
[10] Chandrasena, W.; McLaren, P.G.; Annakkage, U.D.; Jayasinghe, R.P.; Muthumuni, D.;
Dirks, E.: Simulation of hysteresis and eddy current effects in a power transformer,
Electric Power Systems Research, v 76, n 8, May, 2006, p 634-641.
[11] Chiesa, N.; Avendaño, A.; Høidalen, H. K.; Mork, B. A.; Ishchenko, D.; Kunze, A. P.:
On the ringdown transient of transformers, IPST’07 - International Conference on Power
System Transients, Lion, France, June 4-7, no. IPST-124, 2007.
[12] Liorzou, F.; Phelps, B.; Atherton, D. L.: Macroscopic models of magnetization, IEEE
Trans. on Magnetics, vol. 36, no. 2, pp. 418 – 428, Mar. 2000.

View publication stats