You are on page 1of 8

A METHOD OF TRANSFORMER PARAMETERS DETERMINATION FOR

POWER ELECTRONICS APPLICATIONS


Thiago Costa Monteiro, Fernando Ortiz Martinz, Wilson Komatsu, Loureno Matakas Junior
Departamento de Energia e Automao Eltricas Escola Politcnica da Universidade de So Paulo
Av. Prof. Luciano Gualberto, Travessa 3, 158 Sala A2-07 - CEP-05508-900 So Paulo SP - Brasil
tcmonteiro@gmail.com, fernando_martinz@yahoo.com, wilsonk@usp.br, matakas@pea.usp.br

Abstract This paper proposes a model that to twice its nominal steady-state value [2]; or when fed by
adequately describes the operation of the transformer at DC biased voltages, that saturates it, or to evaluate the
deep saturation, suitable for power electronics performance of antisaturation strategies. The model in Figure
applications, and a method for determining its 1 describes the saturation effects by including the non linear
parameters. The magnetizing branch is represented by inductor LM [4] [5]. For convenience, and without loss of
the parallel association of a fixed resistance and a non generality, this paper locates the magnetizing branch at the
linear inductor that can be implemented by using a secondary side.
current source or switched linear inductors. The
parameters with linear behavior are obtained by the
traditional short circuit and open circuit tests. The non
linear behavior of the magnetizing inductance is obtained
by processing the voltage and current waveforms
measured in the open circuit test. Deep saturation
condition is achieved by acquiring these waveforms
during the transformer energization at nominal voltage,
instead of using steady state data, which would require
an oversized power supply and cause excessive thermal
stress at the windings. Pre-demagnetization procedures Fig. 1. Power transformer single phase equivalent circuit.
are developed to cope with the remanent flux. Simulation
and experimental results are presented to confirm the This paper initially presents an alternative method for
validity of the model and the method. obtaining the linear parameters of the transformer, in which
the meters are substituted by an oscilloscope with voltage
Keywords transformer non linear model, transformer and current probes. Parameters are obtained by post
demagnetization, non linear inductor, transformer tests processing only the instantaneous voltage and current
waveforms.
I. NOMENCLATURE Non-linear behavior is then analyzed. Deep saturation is
experimentally achieved by two methods. The first one
L LP - primary winding leakage inductance consists in exceeding the transformer nominal voltage by
means of a variable voltage source while the second strategy
RLP - primary winding resistance consists in forcing several inrush transients with transformer
LLS - secondary winding leakage inductance nominal voltage. The current and voltage waveforms are
RLS - secondary winding resistance registered, the N i M curve is calculated, and a
- magnetization inductance referred to mathematical function is fitted to the curve [5]. Since
LM residual flux density dislocates the N i M curve [5], a
secondary
- magnetization resistance referred to pre-demagnetization procedure, by means of the pre-
RM
secondary insertion of a capacitor, is also proposed.
Numerical processing is again applied to the acquired data
Character indicates reference from primary to secondary
to evaluate magnetization parameters and to find a nonlinear
side or from secondary to primary side.
mathematical function that represents the N i M behavior.
II. INTRODUCTION Two strategies are used to model the non linear inductor. The
first one consists of a controlled current source plus an
Traditional short-circuit and open circuit tests, where integrator and a lookup table or a continuous function. The
voltmeter, ammeter and wattmeter root-mean squared (rms) second one consists of switched inductors.
values are used to obtain an equivalent model of power Finally, simulated and experimental results are compared
transformers leakage and magnetization parameters, are well to conclude about the validity of the derived model and the
known methods [1]. Despite the fact that these tests are parameter estimation method. Simulations were carried out
consolidated for steady state and linear operation, it must be using the PSIM software.
emphasized that power electronics applications like UPS, This method is suitable for computer simulated
active filters, etc., require simulation and analysis of the transformer models excited by PWM modulated inverters
behavior of a power transformer during inrush transients, and transformer transient studies in power systems.
when the transformer can experience a flux that can reach up

978-1-4244-3370-4/09/$25.00 2009 IEEE 1019


III. SHORT CIRCUIT TEST '
X LP = X LS = XL 2 (6)
Figure 2 shows the setup of short-circuit test, in which LLP = X LP 2f (7)
'
usually the low-voltage side is short circuited. By doing that, where X LP is the leakage primary reactance and X is theLS
the magnetization parameters can be neglected and thus the leakage secondary reactance referred to the primary side.
results of this test will represent the leakage impedance z L . The transformer turns ratio is defined as TR = N S N P ,
For convenience, and without loss of generality, this paper where N P and N S are the primary and secondary windings
considers the low voltage side as secondary side.
number of turns, respectively.
A variable voltage source is required to adjust the
The remaining parameters are calculated by:
instantaneous short-circuit voltage v SC (t ) (SC subscript
X LS = X LP (TR )
2
(8)
means short circuit) in a way that the instantaneous short-
circuit current i SC (t ) will be equal to the secondary nominal LLS = X LS 2f (9)
current. An oscilloscope is responsible for registering of where L LS is the secondary side leakage reactance, L Lp is
sampled variables i SC (k ) and v SC (k ) . the primary side leakage reactance and f is the fundamental
The DC resistance of the transformer is also computed by frequency.
applying DC voltage to the transformer, but in general this If the current densities at the primary and secondary are
value is provided by the manufacturer. equal, the ratio of primary to secondary DC resistances is
approximately equal to the square of the turns ratio (10),
where RLSDC is the secondary DC resistance and R LPDC is the
primary DC resistance.
R
TR 2 LSDC (10)
RLPDC
For the sake of simplicity, the AC resistances are
considered equal to DC resistances and skin and proximity
effects are neglected.
If the secondary is short circuited, the winding resistances
are given by (11) and (12):
Fig. 2. Short circuit test setup RLP RLP + RLS TR 2 RL
= 2
=
RLPDC RLPDC + RLSDC TR RLPDC + RLSDC TR 2
A. Leakage parameters derivation RL RLPDC
RLP = (11)
A program is implemented in Matlab to calculate the RLPDC + RLSDC TR 2
leakage impedance in short-circuit test. Initially voltage and R LS ' = RL R LP
current offsets are evaluated and subtracted from the original RLS = (RL RLP ) TR 2 (12)
signals. Since sampling and fundamental frequencies are
known, the number of samples per cycle (N) is determined r
where LS ' is the secondary winding resistance, referred to
and the offset is given by: the primary side. Finally, the primary (13) and secondary
1 N (14) leakage impedances are, respectively:
offset = (k ) (1), Z LP = R LP + jX LP (13)
N k =1
where (k ) are the samples of the voltage or the current. Z LS = R LS + jX LS (14)
Leakage parameters must be obtained at the fundamental
frequency. However, since harmonics are usually present at IV. OPEN CIRCUIT TEST
the mains, v SC (t ) will have other components besides 60Hz. Figure 3 shows the setup for open-circuit test, in which the
Fourier analysis is applied to obtain the amplitude of the primary side is left open and the secondary is fed at nominal
fundamental frequency components of the current and voltage. An oscilloscope is responsible for registering
voltage ( I1 , V1 ) and their respective angles i1, v1 . Thus, sampled magnetizing current i 0 (k ) and v 0 (k ) .
the total leakage impedance (2), the displacement factor (3),
the total leakage resistance (4) and the total leakage
reactance (5) are respectively given by:
Z L = V1 I 1 (2)
cos = cos ( V 1 I1 ) (3)
RL = Z L cos( ) (4)
2 2
X L = Z L RL (5)
Considering that the reflected leakage impedance is
equally divided between primary and secondary windings:
Fig. 3. Open circuit test setup

978-1-4244-3370-4/09/$25.00 2009 IEEE 1020


A. Magnetization parameters derivation 1 N
v(k )
2
VMRMS = (21)
In Figure 3, the magnetization voltage v M (t ) is: N k =1
di 0 (t ) The linkage flux is given by Faradays Law:
v M (t ) = v 0 (t ) rLS .i 0 (t ) L LS .
(15)
dt d(t )
v M (t ) = N (22)
where rLS and L LS are known from the short circuit test. dt
Similarly to short circuit test, offsets are computed and Discretization of (22) by means of bilinear transform leads
subtracted from the measured current and voltages (1). to:
Since the derivative of magnetization current in (15) TS
N (k ) = N (k 1) + [v M (k 1) + v M (k )]
(23)
results in amplification of high order harmonics, a pole is 2
included to the derivative transfer function D(s ) to obtain where TS is the sampling period. The N i M curve can be
the magnitude Bode diagram of Figure 4. plotted by using flux linkage (23) and magnetization current
(18). The curve N (k ) i m (k ) will be used to evaluate the
parameters of the non linear inductor LM .

V. MODELING MAGNETIZATION INDUCTANCE

The non linear characteristic of the magnetization


inductance reflected to the secondary side is modeled in this
paper either by a piecewise linear fit or by the current source
method. In the first case, the N i M curve is subdivided
into two or more parts and a linear fit is made to determine
the slope of these parts. The slopes of these lines represent
Fig. 4. Modified derivative transfer function
the magnetization inductances in the many regions of the
curve. A set of comparators decide the inclusion of a new
In fact, the modified derivative transfer function D ' (s )
inductor in parallel to the previous one, changing the value of
represents a low-order filter expressed by: the total inductance at presetted currents [3]. Figure 5 shows
k s the model for a two slope case.
D ' (s ) = (16)
s + d
where d is the cut-off frequency in rad/s. The condition
which assures unity gain at the fundamental frequency is:
02 + d2
k= (17)
0
where 0 is the fundamental frequency in rad/s. This
transfer function is discretized in Matlab by means of a
bilinear transform. The magnetization current in Figure 3 is: Fig. 5. PSIM saturable inductor model
i M (t ) = i 0 (t ) i R (t ) (18)
In the adopted current source method (Figure 6), the
The term i R (t ) , which represents the hysteresis losses of magnetization voltage is integrated in the circuit simulator
i M (t ) , is equal to: environment to obtain the linkage flux. Since the sampled
v M (t ) values of iM (k ) x N (k ) are computed in topic VI-A and VI-
i R (t ) = (19)
RM B, the magnetization current can be determined by data
search and interpolation on a lookup table, or by evaluating a
where R M is a resistive component that models hysteresis
continuous function, if possible. In fact, the data is processed
losses. The value of R M is: by one of the curve fitting alternatives listed in topic VI-B2.
2
vMRMS 2
vMRMS (20) The elements inside the dotted box of Figure 6 represent
RM = =
PM 1 the model of the magnetization inductance.
vM (t ) iR (t )dt
T

T
0

where v MRMS is the RMS magnetization voltage and PM is


the average active power delivered to R M . If all the losses in
the magnetizing branch are due to R M ( LM is lossless), PM
1 N
is evaluated by ( vM (k )i0 (k )) , using the current io (t )
N k =1
instead of i R (t ) . The RMS value of the magnetization
Fig. 6. Current source magnetization inductance model
voltage is calculated by:

978-1-4244-3370-4/09/$25.00 2009 IEEE 1021


VI. EXPERIMENTAL RESULTS FOR OPEN AND 0.1050A, and (25) for the polynomial region. The resulting
SHORT CIRCUIT TESTS curve is shown in Figure 10.
i M = N (24)
Short circuit and open circuit tests were made for Table I
i M = 10 6 ( 6,4201( N ) 9 + 0.0574( N ) 8 + 0.5418( N ) 7
parameters. An Agilent DSO6014A Oscilloscope with (25)
Agilent N2775A Current Probe and Tektronix P5200 0.0052( N ) 6 + 0.0118.( N ) 5 + 0.0002( N ) 4 0.0001.( N ) 3 )
Voltage Probe were used to acquire data. Current probes
were demagnetized after each measure to reduce measuring
errors.
Table I Experimental setup data
Parameter Value
Primary nominal voltage 127V
Secondary nominal voltage 31V
Apparent Power 100VA
Primary DC resistance 0.1461
Secondary DC resistance 2.4357
Derivative cut-off frequency (see Fig 4) 600Hz
Mains frequency 60Hz
Demagnetization Capacitor 246F

A. Steady State Tests


1) Short circuit test
Fig. 8. N iM curve for nominal voltage condition (1 pu)
The leakage impedances, obtained in short-circuit test experimental results.
were z LS = 0.1539 + j 0.025 and z LP = 2.5655 + j 0.42 , that
is, L LS = 66.4 H and LLP = 1.11mH .

2) Open circuit test


The open circuit test was made in steady state for 1 pu and
1.45 pu secondary voltages. Figure 7 shows the waveform of
v 0 (t ) i (t )
and 0 N i M , for 1 pu. Figure 8 shows the
N iM curve for the same data. The calculated
magnetization resistance is R M = 618.71 . It can be noted
that the transformer is not in a deep saturation condition.

Fig. 9. N iM curve for 1.45 pu voltage condition


experimental results.

0.25

0.2

0.15

0.1
LINKAGE FLUX (V.s)

0.05

Fig. 7. Open Circuit test for nominal condition (1 pu) 0


experimental results. Voltage (top), current (bottom).
-0.05

Since the transformer ratings are small, it is possible to -0.1


proceed with deep saturation test in steady state. Thus, -0.15
Figure 9 presents the N i M curve for 1.45 pu deep -0.2
saturation, when the magnetization resistance is -0.25
R M = 703 . -8 -6 -4 -2 0 2 4 6 8
MAGNETIZATION CURRENT (A)
Polynomial curve fitting was applied to the acquired data
with a linearized region. The results are expressed by (24), Fig. 10. Polynomial fit applied to N iM curve for steady state
which expresses the linear region for currents lower than deep saturation (1.45 pu voltage).

978-1-4244-3370-4/09/$25.00 2009 IEEE 1022


B. Transient Tests N i M curve several times, so that in the end the
In the last item, deep saturation was reached by feeding remaining flux is close to zero.
the transformer with a 1.45 pu input voltage. It is not
recommended to perform steady state tests to obtain deep
saturation data because it would be necessary to operate the
transformer in very severe overvoltage and over-current
conditions and besides these factors, this is an operation
characterized by a highly distorted current. A transformer
with nominal power of more than some kVAs would
probably be damaged. Fig. 12. RLC circuit obtained by insertion of capacitor.
An additional issue to do this trial is the requirement of a
voltage source with higher capacity than the transformer It is reasonable to admit that the maximum voltage
itself. available is the peak of mains voltage V0 .
A good opportunity to obtain deep saturation data occurs
during the transformers energization, once it is known that V0 = VNOM 2 (26)
the transient inrush current can achieve tens of the steady Capacitance C can be estimated by neglecting the losses,
state magnetization current [5], depending on the switching and considering that all the capacitor energy at t=0 is
instant, and on the remanent flux. supplied to the inductor after a quarter of the resonant period,
A simple procedure would consist in acquiring voltage producing a peak current I 0 .
and current data during consecutive energizations. This 2
means: energize and de-energize the transformer for a CV02 LM I 0
= (27)
number of times, with nominal voltage. 2 2
It is difficult to determine the parameter C because of LM
50
non linearity. At this point of the procedure, only the open
VOLTAGE (V)

test with nominal voltage will be available. As the hysteresis


0 curves slope represents inductance, it is possible to find the
magnetization inductance at nominal voltage ( LSAT ), as it is
shown in Figure 13.
-50 The saturated magnetization inductance represents a better
-0.05 0 0.05 0.1 0.15 0.2
parameter to estimate C:
50
L i2
C = SAT 2 0 (28)
CURRENT (A)

0 V0

-50

-100
-0.05 0 0.05 0.1 0.15 0.2
TIME ( )
Fig. 11. Transformer energization. Voltage and inrush current.

Figure 11 shows the tested transformer behavior during its


magnetization. The inrush current is more than twenty times
the nominal current.
Residual magnetic flow present in the transformer during
magnetization causes high currents, and vertical
displacement of the hysteresis curves. This displacement can
be manually compensated for each experimental data
acquisition [5]. Next item will propose a demagnetization
method.

1) Core demagnetization method


This section proposes a demagnetization procedure, in
Fig. 13. Finding saturated magnetization inductance near nominal
order to eliminate residual flux and obtain reliable hysteresis voltage.
data. This procedure is based on the insertion of an initially
charged capacitor, creating a RLC oscillatory circuit, as can By this way it is possible to easily demagnetize the
be seen in Figure 12. The oscillatory transient currents in the transformer before each inrush transient analysis.
circuit should be considerably higher than the nominal Demagnetization voltage and current are shown in Figure 14.
current, forcing the core to reach the deep saturation region This method was tested and the results are still under
both in positive and negative sections. The desirable under evaluation. It proved to substantially reduce residual flux in
damped oscillation guarantees that the current travels the the transformer magnetic core in most cases, but it fails in a

978-1-4244-3370-4/09/$25.00 2009 IEEE 1023


few ones. Curves vertically displaced comparing to steady To obtain more reliable data and better results, the
state curve in fig. 7, were discarded. Additional tests using generated data was mirrored and plotted altogether, as shown
other capacitance values, to achieve different resonant in Figure 16.
frequencies and damping rates are necessary. It is also
possible that the used current probe add some residual flux, 0.25
affecting the measures. 0.2

100 0.15
VOLTAGE (V)

0.1

LINKAGE FLUX (V.s)


50

0 0.05

-50 0

-100 -0.05
-0.05 0 0.05 0.1 0.15 0.2
-0.1
20
-0.15
CURRENT (A)

0 -0.2

-0.25
-20 -15 -10 -5 0 5 10 15
MAGNETIZATION CURRENT (A)
-40
-0.05 0 0.05 0.1 0.15 0.2 N iM
Fig. 16. Superimposed Transient deep saturation curves.
TIME (s)

Fig. 14. Transformer demagnetization. It is important to highlight that there were found peak
currents near 15A, when nominal magnetization current is in
It is also possible to dismiss the demagnetization the range of 0.3A. This means almost 50 times the nominal
procedure, but in this case manual data adjustment will have values.
to be done [5], decreasing the method accuracy. One possible Since simplified modeling of RM as a fixed resistance
merit of not demagnetizing, is the possibility of reaching forces a wider loop, it is necessary to perform a mathematical
deeper saturation for nominal voltages. curve fitting method to find a representative function for the
lossless N i M curve. This function I x N can be used
2) Obtaining the N i M curve
directly or used to generate a table, in the simulation model
The deep saturation N i M curve is obtained from the presented in item V. The lossless curve eases the
acquired data using similar method as used in open circuit determination of the many inductances of the switched
test (shown is section IV). inductor model (topic V). Curve fitting tools require previous
Since inrush currents have strong DC average values, as sorting of the data vectors.
Some groups of curves were studied to perform the curve
shown in previous section, the N i M curves found in this
fitting method. Good results were obtained from polynomial
transient analysis present the asymmetrical patterns as shown and sigmoidal fitting, but the best one was achieved from a
in Figure 15. sum of exponentials. This result is shown in Figure 17.
In the transient tests, instrumentation offset cannot be
compensated, as done in the steady state tests. 0.25

0.2
LINKAGE FLUX (V.s)

LINKAGE FLUX (V.s)

0.2 0.2 0.15


0.1 0.1
0.1
LINKAGE FLUX (V.s)

0 0
-0.1 -0.1 0.05
-0.2 -0.2
0
-10 0 10 -10 0 10
MAGNETIZATION CURRENT (A) MAGNETIZATION CURRENT (A) -0.05

-0.1
LINKAGE FLUX (V.s)

LINKAGE FLUX (V.s)

0.2 0.2
-0.15
0.1 0.1
0 0 -0.2
-0.1 -0.1
-0.2 -0.2 -0.25
-15 -10 -5 0 5 10 15
-10 0 10 -10 0 10 MAGNETIZATION CURRENT (A)
MAGNETIZATION CURRENT (A) MAGNETIZATION CURRENT (A) Fig. 17. Sum of exponentials fit applied to transient deep saturation
N iM
N iM curve.
Fig. 15. Transient curves.

978-1-4244-3370-4/09/$25.00 2009 IEEE 1024


The group of functions used is given by (29):

CURRENT (A)
0.1
iM = f exp ( N ) = a.e b. N a.e b. N
(29) 0
Where a and b are coefficients to be determined by a -0.1
0.78 0.79 0.8 0.81 0.82 0.83 0.84 0.85
curve fitting tool applied to the superimposed transient deep
TRANSIENT METHOD - TIME (s)
saturation N i M curves data.

CURRENT (A)
0.1
These coefficients may be difficult to manipulate in a real 0
time state estimator algorithm of a digitally controlled
-0.1
converter, running in a fixed point processor, due to 0.78 0.79 0.8 0.81 0.82 0.83 0.84 0.85
numerical precision difficulties. In a computer simulation STEADY STATE METHOD - TIME (s)

CURRENT (A)
model their performance is not influenced by this issue. 0.1

In the particular case shown below, these coefficients had 0


to be manually adjusted because the fitting tool result was -0.1
not satisfactory. 0.78 0.79 0.8 0.81 0.82 0.83 0.84 0.85
TWO SLOPES METHOD - TIME (s)
A variable inductor that behaves according to this type of
function was implemented as a current source, according to Fig. 18. Open circuit test at 20V. Waveforms of magnetization
fig. 6, where i L is obtained by (29) and (30). currents, with experimental result (black) and simulated results
(red) of Model I (top), Model II (center) and Model III (bottom).
iL = f exp ( VL )
(30)

ERROR (%)
100
VII. TRANSFORMER MODEL VALIDATION
0

Simulation models were implemented using the PSIM -100


0.79 0.8 0.81 0.82 0.83 0.84
model of Figure 5 and the lookup table model of Figure 6, TRANSIENT METHOD - TIME (s)
with steady state and transient tests for the parameters
ERROR (%)

100
determination. Three models were verified: 0
I. Fig. 6 Nonlinear model based on the sum of
-100
exponentials fitting of the N i M curve obtained by the 0.79 0.8 0.81 0.82 0.83 0.84
STEADY STATE METHOD - TIME (s)
transient tests.
ERROR (%)

100
II. Fig. 6 Nonlinear model based on the polynomial
0
fitting of the N i M curve obtained by the steady state
-100
deep saturation tests. 0.79 0.8 0.81 0.82 0.83 0.84
TWO SLOPES METHOD - TIME (s)
III. Fig. 5 Modified linear model with two slopes
saturable inductor, fitting the N i M curve obtained by the Fig. 19. Open circuit test at 20V. Instantaneous percentage errors of
magnetization current for Model I (top), Model II (center) and
steady state deep saturation tests. Model III (bottom).
These models were implemented in PSIM simulation
software. In the next figures, simulated curves are shown in
red and experimental ones in black.
CURRENT (A)

5
Open circuit with 45V was also chosen for this
0
comparison because it provided data in high saturated
condition. Results are shown in Figures 18 to 21. -5
0.78 0.79 0.8 0.81 0.82 0.83 0.84 0.85
Model I was found by a PSIM model using the sum of TRANSIENT METHOD - TIME (s)
CURRENT (A)

exponentials continuous function: 5


iM = 0.0041.e 38.425. N 0.0041.e 38.425. N (31) 0
Model II was found by a lookup table generated from the
-5
polynomial fitting curve: 0.78 0.79 0.8 0.81 0.82 0.83 0.84 0.85
i M = 10 (6.4201(n ) + 0.0574(n ) + 0.5418(n )
6 9 8 7 STEADY STATE METHOD - TIME (s)
(32)
CURRENT (A)

5
0.0052(n ) 6 + 0.0118.(n ) 5 + 0.0002(n ) 4 0.0001.(n ) 3 )
For currents lower than 0.1050 A, the linear region was 0

experimentally obtained and is given in (33): -5


0.78 0.79 0.8 0.81 0.82 0.83 0.84 0.85
iM = 1.0 N (33) TWO SLOPES METHOD - TIME (s)
Model III was implemented with 2.018 H inductance for Fig. 20. Open circuit test at 45V (saturated). Waveforms of
currents lower than 50mA and a 70mH saturated inductance magnetization currents, with experimental result (black) and
for higher currents. simulated results (red) of Model I (top), Model II (center) and
Model III (bottom).

978-1-4244-3370-4/09/$25.00 2009 IEEE 1025


non linear inductor, were presented. The curve fitting of the
ERROR (%)

100

0 N i M curve was performed and discussed. Simulations


-100 for three different models were performed and compared
0.79 0.8 0.81 0.82 0.83 0.84
TRANSIENT METHOD - TIME (s)
with experimental results in order to validate these models.
It was shown that the deep saturation transient analysis is
ERROR (%)

100
very effective and produces results close to steady state deep
0
saturation analysis, with a wider saturation range and a more
-100
0.79 0.8 0.81 0.82 0.83 0.84
viable procedure, requiring no additional power supply.
STEADY STATE METHOD - TIME (s) Moreover, the nonlinear transformer model obtained from
sum of exponentials curve fitting of the N i M curve
ERROR (%)

100

0 generates more reliable results than traditional linear model


-100 and linear modified two slopes model.
0.79 0.8 0.81 0.82 0.83 0.84
TWO SLOPES METHOD - TIME (s)
This model is believed to be suitable for computer
simulated transformer models excited by PWM modulated
Fig. 21. Open circuit test at 45V (saturated). Instantaneous
inverters as well as in state estimation algorithms running in
percentage errors of magnetization current for Model I (top), Model
II (center) and Model III (bottom).
digital converter controllers.

From these results it is possible to notice that method I REFERENCES


provides a better results. Besides that, absolute error (one
[1] Massachussetts Institute of Technology, Magnetic
cycle average) is lower too, as it is seen on table II, which
circuits and transformers, John Wiley, 1stEdition, New
includes two other voltage source situations:
York, USA, 1943.
[2] C. Fitzer, A. Arulampalam, M. Barnes, R. Zurowski,
Table II - Absolute Error (one cycle average) for the
Mitigation of Saturation in Dynamic Voltage Restorer
Three Models in Different Source Voltage Conditions
Connection Transformers, IEEE Transactions on Power
Source Model I Model II Model III Electronics, vol. 17, no. 6, pp. 10581066, Nov 2002.
Voltage (%) (%) (%) [3] Powersim Inc., Saturable inductor test circuit, PSIM
(V) 5.01.
20 9.99 29.75 13.09 [4] V. Furlanetto, Proposta e validao experimental de
31 11.22 12.30 10.55 um modelo para mquina de solda a ponto CA,
45 4.12 3.45 15.74 Master Degree Dissertation, Escola Politcnica da
52 3.51 4.22 21.00 Universidade de So Paulo. Departamento de Engenharia
Total Eltrica. So Paulo, SP, 2004.
Average 7.21 12.43 15.09 [5] S. G. Abdulsalam, W. Xu, W.L.A. Neves, X. Liu,
Error Estimation of Transformer Saturation Characteristics
From Inrush Current Waveforms Power Engineering
It is possible to notice that model I has good accuracy for Society General Meeting, 2006.
all the tested situations, as model II has poor results for lower
source voltages and model III, for higher source voltages.
The modified linear model, with two inductance slopes is
adequate for low saturation cases, such as the nominal
voltage, but its error grows once into the deep saturation
area. This problem can be minimized by including additional
linear slopes.
Polynomial fit from steady state method has lower error
than two slopes method in both cases. However, it was found
that sum of exponentials fitting is a better approximation
curve and the data originated from transient analysis of deep
saturation can be considered as reliable as steady state
originated data.

VIII. CONCLUSION

This paper has presented some models and their obtaining


methods describing the operation of a transformer at deep
saturation, suitable for power electronics applications.
Strategies to achieve deep saturation in both steady state and
transient tests were analyzed and a core demagnetization
procedure was proposed. Data processing of the voltage and
current to obtain the transformer parameters, including the

978-1-4244-3370-4/09/$25.00 2009 IEEE 1026

You might also like