XVth International Symposium on High Voltage Engineering

University of Ljubljana, Elektroinštitut Milan Vidmar, Ljubljana, Slovenia, August 27-31, 2007 T2-61.pdf

Investigation of Fast Transients Propagation in Layer-type Transformer Windings –

Measurements and Modelling
M. Popov1*, L. van der Sluis1 and R.P.P. Smeets2,3, V.V. Terzija4
1
Delft University of Technology, Faculty of Electrical Engineering, Mathematics and Computer Science,
Mekelweg 4, 2628 CD, Delft, the Netherlands
2
KEMA T&D Testing, Utrechtseweg 310, 6812 AR Arnhem, the Netherlands
3
Eindhoven University of Technology, Faculty of Electrical Engineering,
Eindhoven, the Netherlands
4
The School of Electrical and Electronic Engineering, The University of Manchester,
Manchester M601QD, U.K.
*Email: M.Popov@tudelft.nl

Abstract: When specifying the winding insulation of Traditionally we can distinguish two types of models:
transformers, it is important to know the electrical the first one is based on lumped parameters [4], while
stresses to which the winding can be exposed during the others are based on the transmission line theory
fast transient oscillations. These oscillations occur [5,6]. The advantage of the first type is that it can be
during switching operations by circuit breakers, or when easily implemented in simulation software, whilst the
using gas insulated substations. Therefore, one of the second model, which is purely mathematical, takes into
priorities is to use a high-frequency transformer model account the frequency-dependent losses, but is not very
capable to simulate fast transient oscillations in the easy to be implemented in a simulation software
windings. The model presented in this paper requires environment.
only information about the geometry of the winding and So far it is shown that the transmission line theory
the core, as well the electrical and magnetic parameters can be successfully applied to evaluate voltages along
for the used materials. In the model, frequency- the windings of electrical machines and transformers [5-
dependent core and copper losses are included. 8]. Most of the time, it is considered that for high
Numerical computations are done with and without frequencies, the flux does not penetrate in the core and
taking into account the core losses. Two types of the iron core losses can be neglected accordingly.
measurements are done to verify the validity of the However, in [9] by measuring transformer impedance
model. First, the voltage transients are measured and frequency characteristics for a short-circuited
computed by applying a step impulse with a rise time of transformer and transformer under no load, it was
50 ns. Then, the transformer is switched by a vacuum demonstrated that the iron core does have significant
circuit breaker and the multiple reignitions are analysed. influence up to approximately 100 kHz. In [10], it was
The results verify that the model is applicable to reported that even up to 1 MHz, the iron core losses
simulate the voltage distribution in transformer influence the frequency transients.
windings in a wide frequency range.
The present paper deals with the problem of
1 INTRODUCTION evaluation of fast transient voltages in distribution
transformer windings. The analysis is done for a 15
Experience shows that very fast transient kVA single-phase test transformer with layer-type
overvoltages (VFTO) are not only dangerous with their windings. The transformer has special measuring points
amplitude, but also with their rate of rise. When at the 200th and 400th turn, which are accessible outside
transformer or motor insulation is frequently exposed to the transformer tank. To get insight of how the losses
fast transients like switching surges, it deteriorates and influence the distribution of fast transients, two types of
finally it may fail. The fact [1,2] is that overvoltages computations have been performed. The first case takes
with lower amplitude and higher rate of rise can be as into account only copper losses, whilst in the second
dangerous as overvoltages with higher amplitude. case, voltage transients are computed by taking into
Transformers are normally protected by surge arresters. account the frequency-dependent core losses. In both
However, these devices limit only the amplitude of the cases, the proximity effects are taken into account. The
overvoltages. In practice [2,3], zinc-oxide arresters with transformer is switched by a vacuum circuit breaker
R-C snubbers (ZORC) are found to be an appropriate (VCB) and the voltage transients are measured at the
protection for medium voltage circuits. Because of the same measuring points. The present model is able to
high price of these devices compared to the price of simulate voltage distributions along the winding for a
distribution transformers, utilities want to know how longer period of time. Within the observed time, voltage
probable and how dangerous the occurrence of fast oscillations have a broad frequency range, from a few
transients is. The study of very fast transients is also tens of kHz up to a few MHz.
important for the design of the transformer insulation.

1
2 THE MODEL one can conclude that hybrid modelling is equivalent to
modelling on a turn-to-turn basis.
The origin of Multi Transmission Line Modelling is For the observed case, the transformer winding is
described through the theory of natural modes [11]. formed by layers, and therefore, the easiest way is to
When a network of N coupled lines exists, and when Z model the transformer on a layer-to-layer basis. When
and Y are the impedance and admittance matrices, the voltages in each layer are known, the procedure can
which are the self and mutual impedances and be applied for each layer and the voltages in the turns
admittances between the lines, then: can be determined. The disadvantage of the terminal
admittance model is that it is very difficult to provide
the terminal admittance over a wide frequency range.
d 2V
= − ZYV Also, when a group of turns are terminated by the
dx 2 (1) terminal admittance that represents the rest of the
d2I windings, the computed results are very sensitive to the
= −Y Z I variation of this admittance.
dx 2 Fig. 1 shows the representation of the windings by
transmission lines.
where V and I are incident voltage and current vectors
-1
of the line. Note that Z ≠ Y . Applying the modal
analysis, the system can be represented by the following
two-port network:

⎡ I S ⎤ ⎡ A − B ⎤ ⎡V S ⎤
⎢ I ⎥ = ⎢ − B A ⎥ ⎢V ⎥ (2)
⎣ R⎦ ⎣ ⎦⎣ R⎦
where:
( )
A = Y 0 S γ coth γ l S
−1
Fig. 1 Windings or turns represented by transmission lines.
(3)
S γ cos ech (γ l ) S
−1
B =Y0 In the present case, the high-voltage winding is
In (2) and (3): grounded, at the end. Therefore, to provide convergence
during the computation process, the line is terminated
IS, IR – are current vectors at the sending and the
by impedance Z=1.e-9 Ω. Applying equation (2) to Fig.
receiving end of the line,
1 results in the following equation:
VS, VR – are voltage vectors at the sending and the
receiving end of the line. ⎡ I S1 ⎤ ⎡V S 1 ⎤
⎢I ⎥ ⎢V ⎥
−1
⎢ S2 ⎥ ⎢ S2 ⎥
Y 0 = Z - characteristic admittance matrix, ⎢ ⎥ ⎢ ⎥
⎢ ⎥ ⎡ A' − B '⎤ ⎢ ⎥
S, γ2 – eigen vectors and values of matrix Z Y, ⎢ I Sn ⎥ = ⎢ ⎥ ⎢V Sn ⎥ (4)
and ⎢ I R1 ⎥ ⎢ ⎥ ⎢V ⎥
l - length of the line. ⎢ ⎥ ⎢⎣ − B ' A ' ⎥⎦ ⎢ R1 ⎥
⎢ I R2 ⎥ ⎢V R 2 ⎥
Transformers are normally constructed with many ⎢ ⎥ ⎢ ⎥
turns, and therefore the above theory can be applied for ⎢ ⎥ ⎢ ⎥
a group of turns in order to avoid storing a large number ⎢⎣ I Rn ⎥⎦ ⎢⎣V Rn ⎥⎦
of elements in the computer memory. One way is to In (4), A’ and B’ are square matrices of n-th order
observe some number of turns and the rest of the calculated by (3). The following equations hold for Fig.
transformer to be represented by a terminating 1:
admittance [5,6]. Another way is to apply hybrid V Rn
modelling [10]. This means that the total number of I R1 = − I S 2 , I R 2 = − I S 3 ,… , − I Rn =
turns is split in a number of groups so that the number Z (5)
of a group of turns is far less than the number of turns in V R1 = V S 2 ,V R 2 = V S 3 , ,V Rn −1 = V Sn
the transformer. The grouping should be done in such a
way that the group of turns retains the information for By using these equations and making some matrix
the voltages and currents at the beginning and at the end operations, equation (4) can be expressed as:
of the observed group. If these parameters have the
same value as they have on a turn-to-turn basis, than

2
 ⎡ I S1 ⎤ ⎡V Si ⎤ normally designed with interleaved, concentric or layer
⎢ 0 ⎥ = [ F ] ⎢V ⎥ (6) windings.
⎣ ⎦ ⎣ Rn ⎦ The studied transformer is a single-phase 15 kVA,
6.6/0.07 kV/kV transformer with layer-type windings,
In (6), i = 1,2,3,..,n and 0 is a zero vector of order nx1.
which was specially built for the investigation of
When we observe the model on a layer-to-layer basis,
transient’s propagation. The transformer is equipped
then VRn=0, hence (6) can be rewritten as: with special measuring points in the middle and at the
⎡V S 1 ⎤ end of the first layer of the transformer high-voltage
[ V Si ] = ⎡⎣ Hi−1
H χ ⎤⎦ ⎢ ⎥ (7) side, and also at the end of the second layer. All
⎣0 ⎦ measuring points can be reached from the outside of the
where subscript i = 2,3,..,n, and Hi-1 is a vector of order transformer and measurements can be performed
(n-1)x1. Hi-1 represents the transfer function, the directly at the layers.
elements of which can be calculated from the elements Fig. 2 shows the winding design of the studied
transformer.
of the F matrix as:
Fi,1−1 core
H i-1 = −1
(8)
F1,1 CHg
r11 layers
The other sub-matrix H χ does not have any physical r22
meaning because a zero vector always multiplies it.
The voltages at the end of each layer can be LV
winding
calculated when the voltage at the input is known and
the corresponding transfer functions are calculated. The CHL CHH
time-domain solution results from the inverse Fourier HV winding
transform:
CHg

sin(πω / W )
W
1
VSi (t ) =
2π ∫
−W
πω / W
VSi (b + jω )e(b + jω )t dω (9)
Fig. 2 Description of the layers in a layer-type transformer
winding;
r11, r22, – inner and outer layer radius, CHH-capacitance
for i=2,3,…n and the VSi (b+jω)=Hi-1(b+jω)VS1 (b+jω). between neighboring layers, CHg-capacitance between the
If we divide the real and imaginary part of the integral layer and the core, CHL- HV-LV capacitance.
function, and if we apply the property of evenness of the
real part and oddness of the imaginary part with respect The transformer is modelled on the layer-to-layer basis.
to ω, the following expression can be used [11]: The capacitances are calculated by assuming that the
layers form cylindrical capacitors. So the capacitance
2ebt
W
sin(πω / W ) matrix C is formed in the following way:
VSi (t ) =
π ∫
0
πω / W
real{VSi (b + jω )}cos(ωt )dω (10) Ci,i – is the capacitance of the layer i to ground and the
sum of all other capacitances connected to the layer i,
Ci,j – is the capacitance between layers i and j taken
In (10), the interval [ 0, W ] , the smoothing constant b
with the negative sign (i≠j).
and the step frequency length dω must be chosen The investigation so far [12] has shown that the
properly in order to arrive at an accurate time-domain capacitances hardly change with the frequency. The
response. The modified transformation requires the capacitances to ground in the observed case are actually
input function VS 1 (t ) to be filtered by an exp( −bt ) capacitances between the layers and the core. Because
of the small surface of the layers these capacitances are
window function. To compute the voltages in separate very low and their value is around 1 pF. The static
turns the same procedure can be applied. voltage distribution is strongly affected by the ground
capacitances. However, for values less than 100 pF, the
3 TRANSFORMER AND DATA static voltage distribution can be considered as linear.
For very fast transients, especially for the first a few
The propagation of fast surges in transformers microseconds, it is a common practice to ignore the core
depends on the geometrical design of the windings and and eddy current losses. The inductance matrix L is
the physical parameters of the insulation, like dielectric evaluated as:
C−1
permittivity and magnetic permeability. Electrical and
magnetic properties of the windings and core must also L= (11)
be taken into account. High-voltage transformers are v s2

3
where the velocity of the wave propagation v s is 4 RESULTS AND VALIDATION
calculated by
c The measurements of fast transients for the particular
vs = . (12) measuring points are done in the following way. A pulse
εr generator connected to the high-voltage transformer
In (12), c is the speed of light in vacuum and ε r is the terminal provides maximum 500 V with a very short
relative dielectric permittivity of the insulation. The rate of rise. The full voltage is achieved within 50 ns.
losses were calculated from the inductance matrix L and The transformer has special measuring points so that
the capacitance matrix C [4,5]. The impedance and apart from the terminal voltage, measurements can be
admittance matrices Z and Y are then: done at the end of the first layer, and at the end of the
second layer. In Fig. 4, the measurements and the
⎛ 2ω ⎞
Z = ⎜⎜ jω + 2 ⎟
⎟L
simulations of the voltage oscillations with and without
⎝ σμ 0 d ⎠ (13) iron core losses are presented.

Y = ( jω + ω tan δ ) C 70
Measured voltges: Source, 200th turn and 400th turn
Computed (with iron core losses): 200th turn and 400th turn
60 Computed (no iron core losses): 200th turn and 400th turn
In (13), the second term in first equation corresponds
to the Joule losses because of the skin effect in the 50
copper conductor and the proximity effect. The second
term in the second equation represents the dielectric Voltage (V) 40
losses. In (13), d is the distance between layers, σ is the
conductor conductivity and tan δ is the loss tangent of 30

the insulation.
20
5
100 10
L1,4 10
L1,7 4

80 L1,10 10 0
L1,14 0 0.5 1 1.5 2 2.5 3 3.5 4
3 Time (μs)
10
Inductance (mH)

Resistance (Ω)

60 Fig. 4 Measured and computed voltage oscillations with and

2 without taking into account the frequency-dependent core
10
losses.
40
1
10
70
Computed voltages(with iron core losses)
20 0 Computed voltages(with no iron core losses)
10 60
200th turn
400th turn
0 2 50 600th turn
3 4 5 6 7
10 10 10 10 10 10 800th turn

Frequency (Hz)
Voltage (V)

40
Fig. 3 Mutual inductance and resistance between the 1st and
other coils of the studied transformer.
30

In this way, the impedance matrix takes into account 20

the frequency-dependent copper losses and the 1000th turn
proximity effects. More general, the frequency- 10 1200th turn
1400th turn
dependent core losses should be taken into account.
Applying the Maxwell equations, explicit formulas for 0
0 1 2 3 4 5
the self- and mutual impedances of coils have been Time (μs)
derived [13,14]. In this way, the Z matrix takes into Fig. 5 Computed voltage oscillations with and without
account the frequency-dependent core and copper consideration of the frequency-dependent core losses at
losses. For the studied case, the order of the Z matrix is specific points.
equal to the number of transformer layers. The tested
transformer contains 15 layers with 200 turns/layer. The results, which do not take into account the
Fig. 3 shows the computed mutual inductance frequency-dependent core losses are calculated by
between the 1st and other coils of the studied applying the impedance matrix according to (13). At the
transformer. first local maximum, around 1 μs, the model with the
core losses shows better matching with the measured

4
results. Since it was not possible to measure the voltages To show the validity of the transformer model for a
at each layer, and based on the validation of the model wide frequency range, measurements of the multiple
with the presented measurements, the computed reignition overvoltages at the transformer high-voltage
voltages in the rest of the layers are compared with terminal, at the 200th and the 400th turn during switching
those computed by taking into account the frequency- of the transformer with a VCB have been done. The
dependent losses. In Fig. 5, computed voltage experimental test set-up is shown in Fig. 6. The
oscillations at specific layers are presented. For the first capacitor applied at the source side keeps the voltage
microsecond, the rate of rise of the voltages in both after switching stable. The voltage at the source side is
cases is almost similar. measured by using a voltage transformer. An inductive
Between 1 μs and 3 μs both lines show good load is used to provoke reignitions. The transformer is
agreement of the oscillation frequency and amplitude. switched under a source voltage of 5 kV. The voltages
Beyond 3 μs the difference becomes greater. Therefore, are recorded at the measuring points simultaneously.
when the input excitation to the transformer is a
complex surge with oscillations with different 50
Measured
frequencies, and the surges are observed over a longer 40
Calculated
period of time, it is desirable to use a high-frequency
transformer model that takes into account the frequency 30
dependency of its parameters [12,15].
20

Voltage (kV)
10

-10

-20

-30
0 500 1000 1500 2000
Time (μs)
Fig. 8. Measured and computed voltage at the 200th turn
according to the experimental set-up from Fig. 6.

50
Measured
Calculated
40

Fig. 6 Experimental setup: 30

1. Step-up transformer; 2. AC voltage transformer; 3.
20
Voltage (kV)

Shunt capacitor; 4. VCB; 5. Cable; 6. HF current probe; 7.

Test transformer; 8. HF voltage probe; 9. Inductive load; 10.
10
HF earthing.
0
50
Measured
40 Calculated -10

30 -20

20 -30
0 500 1000 1500 2000
Voltage (kV)

10 Time (μs)
Fig. 9. Measured and computed voltage at the 400th turn
0
according to the experimental set-up from Fig. 6.
-10
Fig. 7 shows the measured voltage at the terminal of the
-20
transformer. This voltage is used as an input voltage that
-30 is needed to determine the other line-end coil voltages.
-40
The calculated terminal voltage is produced by applying
0 500 1000 1500 2000 the inverse Fourier transform of the measured voltage,
Time (μs)
which is also a test that the inversion is correctly done.
Fig. 7. Measured and computed voltage at the transformer Fig. 8 and Fig. 9 represent the measured and computed
terminal according to the experimental set-up from Fig. 6.
voltage at the 200th turn and the 400th turn respectively.

5
 The validity of the model is verified by measurements
10 of fast transients produced during switching the
Measured
transformer with a VCB. The method introduced here
Voltage (kV)

5 Calculated
offers a possibility to predict the voltage distribution
0
along the transformer winding when the transformer is
-5 excited by fast transient surges, which can be measured
-10 at the transformer terminal.
600 620 640 660 680 700
Time (μs)
a)
6 ACKNOWLEDGMENT
20 Measured
Calculated with iron losses This work was supported by the Dutch Scientific
Voltage (kV)

10 Calculated with no iron losses Foundation NWO/STW under Grant Veni, Det.6526.
0
-10
7 REFERENCES
-20
1379 1379.5 1380 1380.5 1381 1381.5 1382
Time (μs) [1] R.E. Pretorius and A.J. Eriksson: “Field studies of Switching
b)
Surge Generators in High-voltage Vacuum Contactor Controlled
Fig. 10. Enlarged reignitions: a) between 600 μs and 700 μs Motors – Results of Extensive Practical Investigations”, IEE
from Fig. 8; b) between 1379 μs and 1382 μs from Fig. 9. Publication 210, 1982, pp. 59-65.
[2] R.E. Pretorius: “Optimal Surge Suppression on High-voltage
Fig. 10a shows the enlarged values of the measured and Vacuum Contactor Controlled Motor”, IEE Publication 210,
1982, pp. 65-70.
calculated voltage of the 200th turn for reignitions [3] J. Lopez-Roldan, H. de Herdt, T. Sels, D. van Dommelen, M.
between 600 μs and 700 μs. There is a good matching Popov, L. van der Sluis and J. Declercq: “Analysis, Simulation
between the calculated and measured voltage values and Testing of Transformer Insulation Failures Related to
between two adjacent reignitions. Within this region, Switching Transients Overvoltages”, CIGRE, paper 12-116,
Paris, 2002.
the voltage oscillates with frequency between 10 kHz [4] Y. Shibuya and S. Fujita: “High-Frequency Model of
and 100 kHz. Transformer Winding”, Elect.Eng. Jpn., vol. 146, No.3, pp. 8-
Fig. 10b shows the voltage oscillation between 1379 15, 2004.
μs and 1382 μs from the time scale shown in Fig. 9. We [5] K. J. Cornick, B. Fillat, C. Kieny and W. Müller: “Distribution
of Very Fast Transient Overvoltages in Transformer Windings”,
observe only one reignition that occurs within 3 μs of CIGRE, paper 12-204, Paris, 1992.
this region. In the same figure, the variation of the [6] J.L. Guardado and K.J. Cornick: “A Computer Model for
voltage is shown for the case when the iron core losses Calculating Steep-Fronted Surge Distribution in Machine
Windings”, IEEE Trans. on Energy Convers, vol. 4, no. 1, pp.
are ignored. 95-101, Mar. 1989,.
For very fast transient oscillations like those shown in [7] R.E. Adjaye and K.J. Cornick: “Distribution of Switching Surges
Fig. 10b, the results can be improved by observation of in the Line-end Coils of Cable Connected Motors”, Proc. Inst.
a shorter time interval, for example an interval where Elec. Eng., Elect. Power Applic., vol..2, no. 1, pp.11-21, Feb.
1979.
only one reignition occurs. This allows us to apply a [8] P.G. McLaren and H. Oraee: “Multiconductor Transmission-line
wider frequency range W. Despite the good matching of Model for the Line-end Coil of large AC Machines”, Proc. Inst.
the voltages in the lower frequency region, in the high- Elec.Eng.B, vol. 132, no.3, pp. 149-156, May 1985.
frequency region, the voltage oscillates with [9] A.O. Soysal: “A Method for Wide Frequency Range Modelling
of Power Transformers and Rotating Machines”, IEEE Trans. on
approximately the same frequency and its amplitude is a Power Del., vol. 8, no. 4, pp. 1802-1810, Oct. 1993.
bit higher than that of the measured voltage. This [10] Y. Shibuya, S. Fujita and N. Hosokawa: “Analysis of Very Fast
implies that the frequency-dependent copper and core Transient Overvoltages in Transformer Windings”, Proc. Inst.
losses are very important for accurate determination of Elec. Eng., Gen. Trans.Dist., vol. 144, no. 5, pp.461-468, Sep.
1997.
very fast transients. [11] J.P. Bickford, N. Mullineux and J.R. Reed: Computation of
Power System Transients, London U.K.: Peregrinus, 1976, ISBN
5 CONCLUSION 0901223859.
[12] J.L. Guardado and K.J. Cornick: “Calculation of Machine
Winding Electrical Parameters at High-Frequencies for
A transformer model based on transmission line Switching Transients Studies”, IEEE Trans. on Energy Conver.,
approach by taking into account the complete iron core vol. 11, no. 1, pp. 33-40, Mar. 1996.
losses and the proximity effect between the layers has [13] D.J. Wilcox, M. Conlon and W.G. Hurley: “Calculation of self
and mutual impedances for coils on ferromagnetic cores”, Proc.
been presented. Careful attention has been paid to the Inst. Elec. Eng. A, vol. 135, no. 7, pp. 470-476, 1988.
more accurate representation of the transformer losses [14] D.J. Wilcox, W.G. Hurley and M. Conlon: “Calculation of self
and inductances depending on frequency. The present and mutual impedances between sections of transformer coils”,
modelling of transformer winding on a layer-to-layer Proc. Inst. Elec. Eng. C, vol. 136, no. 5, pp. 308-314, 1989.
[15] J.L. Guardado, J. A. Flores, V. Venegas and J.L. Naredo: “A
basis introduces the applicability at high frequencies machine Winding Model For Switching Transients Studies
and for longer time intervals. Using Network Synthesis”, IEEE Trans. on Energy Conver.,
2005, vol. 20, no. 2, pp. 322-328, Jun. 2006.