Analysis and Modeling of Dielectric Responses

of Power Transformer Insulation

Zheng Tong. Yao, Member, IEEE, and Tapan. Kumar Saha, Senior Member, IEEE

In recent time return voltage measurement has gained

Abstract—This paper reports the modeling on dielectric significant attention by utility engineers. However, accurate
response for power transformer condition assessment. A model interpretation of RV still remains a difficult task. Due to the
(a numerical tool) is implemented by C++ code based on uncertainty in the results obtained from return voltage
dielectric theory. The model can calculate return voltage and
measurement, modeling of the RV is essential which can help
polarization/depolarization currents. It can also calculate the
maximum return voltage to generate spectrum data. Through us to quantitatively analyze the dielectric response. A
the modeling, the parameters which contribute to the dielectric mathematical model of return voltage can give us great help
response in the real measurement can be seen much more in the interpretation of RVM test results. Through the model
clearly. A good numerical model is indispensable to remove the for return voltage the actual parameters involved or
uncertainty and to accurately interpret the measurement results. contributing to the results can be easily seen. These
The model can help us quantitatively analyze the measurement
parameters are: DC conductivity (σ), permittivity (εr),
results from dielectric responses of oil paper insulation.
response function (f(t)), geometrical capacitance (C0), etc. A
Index Terms--Modeling, condition monitoring, dielectric good numerical model is indispensable to accurately interpret
response measurements, polarization current, return voltage, the measurement results and thus helps to correlate with
transformer insulation. physical properties of insulation. Gafvert et al. [1] [2] have
done some significant work in this area.

I. INTRODUCTION Our model, a numerical tool is based on the dielectric

U TILITIES desire to know the accurate condition of their

equipment, including aged and new power transformers.
Condition monitoring techniques need to provide clear and
response--polarization/depolarization current measurement.
From the current measurement, we can determine the
response function. With the response function, we can model
correct information about the condition of the equipment to (calculate) the return voltages and the RV spectrum.
satisfy this requirement. The quality of the electrical
insulation is a key element for reliable operation of a power We have designed comprehensive C++ software to
transformer. The electrical insulation materials of a power implement the model. At this stage the model can estimate
transformer are mineral oil and cellulose paper. During its return voltage (with selected charging/discharging time). It
operation, the oil/paper insulation can be aged due to thermal, can also find out the maximum return voltage to generate
oxidative and hydrolytic degradation. The service ageing can spectrum data.
be monitored by chemical techniques, such as dissolved gas
analysis (DGA). The ageing also can be assessed by electrical
diagnostic techniques, such as dielectric response II. BACKGROUND KNOWLEDGE
measurements. Dielectric response measurements are non- This section provides the background knowledge of the
destructive and relatively new diagnostic techniques used to dielectric response measurement (in the time domain). Fig. 1
assess the condition of oil/cellulose insulated equipment, such shows a simplified diagram of dielectric response
as power transformers. The methods of dielectric response measurement (time domain) for a power transformer.
measurements in the time domain are Transformer HV and LV terminals are connected together to
polarization/depolarization current measurement and return form a two terminal test object. In the diagram, polarization
voltage (RV) measurement. current measurement is performed with s1 closed, s2 and s3
open. For the depolarization current measurement, s2 is
closed, s1 and s3 open. For the return voltage measurement,
s3 closed and s1, s2 open.
Z. T. Yao and T. K. Saha are with the School of Information Technology
and Electrical Engineering, The University of Queensland, Brisbane, QLD
4072, Australia. (Emails: zheng@[Link], and
saha@[Link]).

0-7803-7519-X/02/$17.00 (C) 2002 IEEE

Fig. 1. Simplified Diagram of Dielectric Response Measurement (time
domain) for Power Transformer

The polarization currents measurement is performed by Fig. 4. A Demonstration of the Return Voltage Measurement (RVM)
applying a step excitation voltage (Vc) on the dielectric Method (One of the 12 Cycles)
materials and depolarization current is measured with power
supply removed as shown in Fig. 2.
Return voltage measurement is shown in Fig. 4. In RV
measurement, a transformer is charged initially for 0.5
second, then in every next cycle the charging time is doubled,
up to 1024 seconds. The charging to discharging time ratio is
always kept as 2. Charging and discharging current and return
voltage data are recorded for every test cycle. Summarize the
maximum return voltage and the central time constant from
each test cycle, a return voltage spectrum can be formed. The
return voltage spectrum can give an indication to the
insulation condition. Some typical return voltage spectra are
given in Fig. 5. The operation times of these transformers are
as follows: T4 (3.15 MVA) was 3 years, T5 (5 MVA) was 38
years, and T6 (5 MVA) was 33 years. Take the T4’s spectrum
Fig. 2. A Demonstration of the Polarization and Depolarization Currents (the right hand side one in Fig. 5) as an example, the peak in
the spectrum can provide indication of the insulation
Fig. 3 shows measured polarization currents from some
condition. The spectrum also provides the range of the
moisture-conditioned samples we prepared in our laboratory.
response in the time domain [4] [5] [6] [7].
It can be seen that the amplitude of long term DC polarization
current is very sensitive to the moisture content in paper
insulation. This demonstrates that polarization current
measurement is capable of assessing the paper insulation
moisture level [3].

Fig. 5. Some Typical Return Voltage Spectra

Fig. 3. The Measured Polarization Current from Moisture Conditioned
Samples

0-7803-7519-X/02/$17.00 (C) 2002 IEEE

 III. MODELING the f(t) calculated from (3), we can find the parameters of f(t)
This section describes the theory and practice of the for (4). TABLE I provides some f(t) parameters.
modeling. The modeling includes three parts: Modeling of
polarization and depolarization currents, modeling response
TABLE I
function f(t), and modeling return voltage, where the return RESPONSE FUNCTION’S PARAMETERS
voltage modeling is the most complex part.
Test Object A t0 n m
A. Modeling Polarization and Depolarization Current
The polarization and depolarization current can be 100kva 0.1469 1710.9 0.2823 1.4657
modeled with the following two equations [2]. Transformer
2% H2O 0.01883 800.16 0.39136 1.439
σ
i pol (t ) = C oVc + C 0Vc f (t ) (1) Sample
εo 3% H2O 0.02544 459.51 0.44221 1.8595
Sample
i depol (t ) = − C oVc f (t ) + C 0Vc f (t + t pol ) ( 2)
The capacitance C of the test object can be obtained by
using a LCR meter and then dividing by the relative
where:
permittivity εr we can obtain C0. C0=C/εr.
ipol is the polarization current,
The DC conductivity σ of the test object can be obtained
idepol is the depolarization current,
C0 is the geometrical capacitance of the test object, by adding equation (1) and (2) together, assume f(t+tpol) ≅ 0.
Vc is the step voltage (charging voltage), An approximation of the DC conductivity of cellulose paper
σ is the DC conductivity of the dielectric material, can be made from polarization and depolarization currents,
using the following relation:
ε0 is the vacuum permittivity,
f(t) is the response function of the dielectric material, ε0
σ ≅ [i pol (t m ) + idepol (t m )] (5)
tpol is the length of the polarization period. Vc C 0
where:
On the right hand side of (1), the first term relates to the C0 is geometrical capacitance, C0 = C/εr, (C obtained from
intrinsic conductivity of the test object and the second term
measurement, εr is relative permittivity, in the range of 3~5),
represents the polarization processes during the time of
tm represents the largest value of time, for which the currents
charging voltage application [8]. If we know f(t), C0 and σ,
have been measured.
the currents can be modeled with the above two equations.
Once we find the f(t)’s parameters and obtain the
B. Modeling the Response Function f(t) geometrical capacitance of the test object as well as find out
When the polarization time is sufficiently long, f(t+tpol)≅0, the dc conductivity of the test object, we can model
from (2) the response function f(t) is equal to (calculate) return voltage curves. σ can be directly related to
moisture and ageing. So by estimating the conductivity, we
− idepol (t )
f (t ) = (3) can also estimate the condition of insulation. The main
C 0 Vc questions of separation of moisture and ageing still remains
The f(t) is calculated from the depolarization current. If we unanswered. We have some work in progress to investigate
want to model the currents (and the return voltage), we need this phenomenon [3].
to model the f(t) from (3) into parametric form, in which the
response function can be expressed as follows: C. Modeling the Return Voltage
A The modeling of the return voltage starts with the current
f (t ) = n m
(4)
t  t  equation. The current density is the sum of conduction and
  +   displacement currents [10] [11].
 t0   t0  dD
The equation comes from the Curie-von Schweidler law: i(t) i (t ) = σ E (t ) + (6)
dt
≈ At-n, a “universal” law which states the depolarization where:
currents of a wide range of dielectric materials [9]. σ is the DC conductivity of the dielectric material,
Using the KaleidaGraph software curve fitting function
E(t) is an external electric field,
(which is based on the Levengerg-Marquardt algorithm), and
D(t) is electric displacement (or dielectric induction).

0-7803-7519-X/02/$17.00 (C) 2002 IEEE

 (Vr = Vreturn.) Also let Vr = V to simplify the equation. The
The dielectric induction is following (13) is the equation we use to model (calculate) the
D(t ) = ε 0ε r E (t ) + P(t ) (7 ) return voltage
where: dV
σ V (t ) + ε 0 ε r + ε 0 f (0)V (t ) + ε 0V0[ f (t − t1 ) − f (t )]
εr is the relative permittivity, dt
t df (t − τ )
P(t) is polarization. + ε0 ∫ V (τ ) dτ = 0 (13)
t2 dt
The polarization is given by the convolution integral Integrate equation (13) between ti and ti+1, where ti+1 > ti >
t t2 giving.
P(t ) = ε 0 ∫ f (t − τ ) E (τ ) dτ (8) t t dV t
0
σ ∫ V (t )dt + ε 0ε r ∫ dt + ε 0 f (0)∫ V (t )dt
i +1 i +1 i +1

t
i t dti t i

If we substitute (7) and (8) into (6), then the electric t t t d f (t − τ )

+ ε 0V0 ∫ [ f (t − t1 ) − f (t )]dt + ε 0 ∫ {∫
i +1 i +1
V (τ ) dτ }dt
current becomes t t t dt

{∫ f (t − τ ) ⋅ E (τ )dτ } (9)
i i 2

dE d t =0 (14)
i (t ) = σE (t ) + ε 0 ε r + ε0
dt dt 0 where V in (14) is return voltage.

If we use Leibnitz’s rule for differentiation of integrals The DC conductivity σ in the return voltage modeling
should be the composite conductivity which considers the oil
d φ 2 (α ) φ 2 (α ) ∂F dφ dφ and paper layers. The oil conductivity can also be calculated
dα ∫ φ1 (α )
F ( x, α )dx = ∫
φ1 (α ) ∂α
dx −F (φ1 , α ) 1 + F (φ 2 , α ) 2
dα dα with (5) by replacing the tm to tb. Where the ipol(tb) and
idepol(tb) are the initial values in polarization and
then equation (9) can be modified to (10) depolarization currents. The conductance of the composite
dE material can be calculate as follows [1]:
i (t ) = σE (t ) + ε 0ε r +
dt
Goil ∗ G paper
 t d f (t − τ )  Gcomposite = (15)
ε 0 f (0) E (t ) + ε 0 ∫ ⋅ E (τ ) dτ  (10) G paper + Goil
 0 dt 

Referring to Fig 4, during the return voltage measurement Which means the composite conductivity can be calculated
(of each cycle) a charging voltage (which generates the as (16).
external field) is applied to a dielectric test object for times 0
< t < t1. When t >t1, charging voltage Vc=0, external field σ oil ∗ σ paper
E=0, and discharging takes place during t1 < t < t2. When σ composite = (16)
σ paper + σ oil
t>t2, discharging path is removed, therefore the test object in
Fig. 1 is in open circuit condition, that is i(t)=0, for t>t2.
The relative permittivity εr also needs to be considered for
The 0—t2 region of the integral in (10) may be treated as the composite system when a value is chosen.
follows (refer to Fig. 4)
The calculated and measured return voltage curves from a
d f (t − τ )
t2 t1 d f (t − τ ) t1 d f (t − τ ) 100KVA transformer are given in Fig. 6.
∫0 dt E (τ )dτ = ∫0 dt E (τ )dτ = E0 ∫0 dt dτ
= E0 [ f (t − t1 ) − f (t )] (11)

With the above result and the open circuit condition (i(t)=0
for t≥t2), (10) becomes (12)
dE
σ E (t ) + ε 0ε r + ε 0 f (0)E (t ) + ε 0 E0 [ f (t − t1 ) − f (t )] +
dt
t d f (t − τ )
ε0∫ E (τ )dτ = 0 (12)
t2 dt
Use E0 = V0/d (where V0 = Vc = Vcharging) and E = Er = Vr/d
to convert equation (12) to (13) for return voltage calculation.
Fig. 6. Calculated and Measured Return Voltage Curves

0-7803-7519-X/02/$17.00 (C) 2002 IEEE

 [3] Z. T. Yao and T. Saha, "Separation Of Ageing And Moisture Impacts
OnTransformer Insulation Degradation By Polarization
Comprehensive C++ software has been designed to Measurements," accepted by CIGRE, Paris, 2002.
implement the model. The user-friendly interface of the [4] T. K. Saha, M. Darveniza, D. J. T. Hill, and T. T. Le, "Electrical and
software is shown in Fig. 7. At this stage the model can Chemical Diagnostics of Transformers Insulation-Part A: Aged
calculate return voltage (with selected charging time, as many Transformer Samples," IEEE Transactions on Power Delivery, vol. 12,
pp. 1547-1554, 1997.
cycles as we want), and polarization & depolarization current. [5] T. K. Saha, M. Darveniza, D. J. T. Hill, and T. T. Le, "Electrical and
It can also determine the maximum return voltage and the Chemical Diagnostics of Transformers Insulation-Part B: Accelerated
corresponding central time constant to generate spectrum Aged Insulation Samples," IEEE Transactions on Power Delivery, vol.
12, pp. 1555-1561, 1997.
data. Once the models are fully implemented, thorough [6] T. K. Saha, Z. T. Yao, T. T. Le, M. Darveniza, and D. J. T. Hill,
understanding of RV spectra would be possible for different "Investigation of Interfacial Polarization Spectra Parameters for
moisture and ageing condition. To support this modeling Accelerated Aged Oil-Paper Insulation and Its Correlation with
Molecular Weights and Furan Compounds," presented at CIGRE,
accelerated ageing experiments of different moisture and Paris, 2000.
ageing condition are vital. The models also need to be [7] T. K. Saha, M. Darveniza, D. J. T. Hill, Z. T. Yao, and G. Yeung,
validated by field transformer measurements and must be "Investigating the Effects of Oxidation and Thermal Degradation on
Electrical and Chemical Properties of Power Transformers Insulation,"
correlated with other diagnostic methods. This work is in
IEEE Transactions on Power Delivery, vol. 14, pp. 1359-1367, 1999.
progress and finding will be reported in future paper. [8] W. S. Zaengl, "Dielectric Spectroscopy in Time and Frequency
Domain for HV Power Equipment (Transformers, Cables etc.),"
presented at International Symposium on High Voltage Engineering,
Bangalore, India, 2001.
[9] A. K. Jonscher, Dielectric Relaxation in Solids. London: Chelsea
Dielectrics Press, 1983.
[10] A. K. Jonscher, Universal Relaxation Law. London: Chelsea
Dielectrics Press, 1996.
[11] E. Ildstad, U. Gafvert, and P. Tharning, "Relation Between Return
Voltage and Other Methods for Measurements of Dielectric
Response," presented at IEEE International Symposium on Electrical
Insulation, Pittsburgh, PA, USA, 1994.

VI. BIOGRAPHIES

Zheng Tong Yao received his BEE from the

Fig. 7. The User Interface of the Model Department of Electrical and Computer System
Engineering, James Cook University, Australia in
1995. From 1996 to 1998, he worked as an
Electrical and Electronic Engineer at The
University of Queensland (UQ), Australia. In 1999,
Tong gained his Master Degree (by research) in the
IV. CONCLUSION
field of Computer Science and Electrical
An attempt has been made to model dielectric response of Engineering from UQ. From 1999, Tong is
transformer insulation. The modeling work shows promising pursuing his PhD in the School of Information Technology and Electrical
Engineering, UQ. He is a Member of IEEE.
result and this is still in progress. The dielectric modeling
results will be correlated to chemical results to predict the
condition of transformer. The task of RV results
Dr. Tapan Kumar Saha is Senior Lecturer in the
interpretation will be much more easier with this tool. School of Information Technology and Electrical
The model will help us quantitatively analyze the results Engineering, The University of Queensland,
from dielectric response measurement. Through the modeling Australia. He is a Senior Member of the IEEE and a
Chartered Professional Engineer of the Institute of
of the dielectric response, the parameters which significantly Engineers, Australia. His research interests include
contribute to the response can be seen much more clearly. power systems and condition monitoring of
This greater clarity will allow us to predict more accurately electrical equipment.
the condition of the oil paper insulation in power transformer.

V. REFERENCES
[1] U. Gafvert, "Condition Assessment of Insulation Systems--Analysis
of Dielectric Response Methods," presented at Nordic Insulation
Symposium, Bergen, Norway, 1996.
[2] U. Gafvert, G. Frimpong, and J. Fuhr, "Modelling of Dielectric
Measurements on Power Transformers," presented at CIGRE, 1998.

0-7803-7519-X/02/$17.00 (C) 2002 IEEE