Understanding Moisture Dynamics and Its Effect on the

Dielectric Response of Transformer Insulation

Author
Cui, Yi, Ma, Hui, Saha, Tapan, Ekanayake, Chandima

Published
2015

Journal Title
IEEE Transactions on Power Delivery

Version
Post-print

DOI
https://doi.org/10.1109/TPWRD.2015.2426199

Copyright Statement
© 2015 IEEE. Personal use of this material is permitted. Permission from IEEE must be
obtained for all other uses, in any current or future media, including reprinting/republishing this
material for advertising or promotional purposes, creating new collective works, for resale or
redistribution to servers or lists, or reuse of any copyrighted component of this work in other
works.

Downloaded from
http://hdl.handle.net/10072/99103

Griffith Research Online

https://research-repository.griffith.edu.au
 1

Understanding Moisture Dynamics and Its Effect

on Dielectric Response of Transformer Insulation
Yi Cui, St. Member, IEEE, Hui Ma, Member, IEEE, Tapan Saha, Senior Member, IEEE, Chandima
Ekanayake, Member, IEEE

variation in temperature tends to change the moisture

Abstract— Dielectric response measurement has recently been distribution in cellulose and oil [5]. This can lead to an
adopted by utilities for evaluating moisture content in cellulose inaccurate estimation of moisture in the cellulose [6].
insulation (paper and pressboard) of transformers. Moisture The utilities have widely adopted the dielectric response
distribution is highly dependent on temperature. Since the
method for estimating moisture content in transformers’
temperature inside a transformer may change during the
dielectric response measurement, the moisture in the cellulose insulation. During dielectric response measurement,
transformer’s cellulose and oil insulation can hardly attain an the transformer is removed from the grid and thus the
equilibrium state. Instead, moisture dynamics exist inside the temperature inside the transformer continuously drops.
transformer: (1) cellulose absorbs (desorbs) moisture from (to) oil Consequently, during the course of the dielectric response
with the changes in temperature; and (2) moisture migrates inside measurement, moisture migrates between cellulose and oil.
cellulose due to a moisture gradient. This paper investigates
moisture dynamics and its effect on dielectric response of a
This will influence the interpretation of the results of dielectric
transformer’s cellulose insulation. It proposes a distributed response measurement and affect the accuracy of moisture
parameter model to reveal the correlation between moisture estimation in the transformer’s cellulose insulation. As an
distribution (under non-equilibrium conditions due to thermal example, the authors made two dielectric measurements on one
transients) and dielectric response parameters (dielectric losses utility’s transformer. During the first measurement a moisture
and permittivity) of cellulose insulation. It then estimates these content of 2.9 % at 30 °C top-oil temperature was recorded.
parameters under moisture non-equilibrium conditions. The
accelerated ageing and moisture diffusion experiments are
The second measurement commenced immediately after, and
conducted on a prototype transformer to verify the proposed recorded 3.6 % at 26.5 °C. Therefore, for an accurate
model. The methodology developed in this paper can help the estimation of moisture in cellulose insulation based on
proper interpretation of dielectric response measurement of field dielectric response measurement, it is necessary to investigate
transformers under thermal transients. the correlations between moisture dynamics and dielectric
response parameters (i.e. dielectric loss and permittivity) of a
Index Terms—Cellulose, dielectric response, insulation,
transformers’ insulation.
moisture diffusion, moisture dynamics, oil, transformer.
This paper is aimed at understanding temperature dependent
I. INTRODUCTION moisture dynamics and its effect on the dielectric response of the
cellulose insulation of transformers. A distributed parameter
Life expectancy of a power transformer is largely determined model is proposed to explore correlations between moisture
by the ageing condition of its cellulose insulation. Moisture is distribution under non-equilibrium conditions and dielectric
one of the most harmful agents for cellulose insulation. It can response of cellulose insulation. By using this model, dielectric
accelerate the cellulose ageing rate and reduce both dielectric losses and permittivity of cellulose insulation are estimated when a
and mechanical strength of cellulose insulation [1]. Therefore, it transformer is under temperature variation and consequent
is of great interest for utilities to estimate moisture content in moisture non-equilibrium. To verify the methodology developed in
cellulose insulation of transformers. this paper, extensive ageing and moisture diffusion experiments
Moisture content in cellulose can be directly measured by the are performed on a prototype transformer.
Karl Fischer Titration (KFT) method. However, it requires
collecting paper samples from a transformer’s winding, which is II. MOISTURE IN TRANSFORMER’S INSULATION SYSTEM
difficult in practice. An alternative approach is through
measuring the moisture content of an oil sample and then A. Moisture Dynamics in Cellulose and Oil Insulation
determining the moisture content in the cellulose from Due to moisture ingress from the environment and cellulose
equilibrium charts [2-4]. This approach assumes an equilibrium degradation, moisture is present in a transformer. During the
state of moisture distribution in the transformer. However, it is normal operation of a transformer, most moisture is affiliated
hard to attain an equilibrium state in the transformer and any with its cellulose insulation. Water generally has low solubility
in transformer oil. With an increase in temperature, the water
Y. Cui, H. Ma, T. Saha and C. Ekanayake are with School of Information solubility in oil can be significantly increased. On the other
Technology and Electrical Engineering, The University of Queensland ，
hand, free water can be formed if the moisture in oil exceeds
Brisbane, Australia (e-mail: y.cui3@uq.edu.au, huima@itee.uq.edu.au,
saha@itee.uq.edu.au, chandima@uq.edu.au). the saturation level.
 2

For an in-service transformer, it can absorb water from based on well-defined pressboard samples with different
environment and its oil can contain some amount of water. The temperatures and moisture contents, it is possible to estimate
transformer’s oil-cellulose insulation system is in an the moisture content in cellulose insulation in a transformer.
equilibrium state if the partial pressure of water in oil equals The next section will provide a brief review on dielectric
the partial pressure of water in cellulose. When the partial response technique for moisture content estimation.
pressures in oil and cellulose become different, cellulose
absorbs/desorbs water from/to oil to maintain the equilibrium. III. DIELECTRIC RESPONSE OF TRANSFORMER INSULATION
Moreover, moisture diffusion occurs inside cellulose insulation
due to a moisture gradient. A. Dielectric Response in Frequency Domain
When a dielectric material is imposed by an alternating

B. Estimation of Moisture Contents in Cellulose Insulation electric field E = Em eiωt , the complex dielectric displacement
Over the past twenty years, extensive studies have been in this material becomes (2)
performed for estimating moisture content in cellulose insulation.  t

Most of these studies are focused on moisture diffusion using D (t)= e 0e ∞ Em eiωt + e 0 ∫ f (t − t 0 )Em eiωt0 dt0 (2)
Fick’s second law [7] in one dimension as (1) −∞

where ε 0 denotes the permittivity of vacuum, ε ∞ denotes the

∂W ( x, t ) ∂  ∂W ( x, t ) 
= D  (1) high frequency relative permittivity of the material and f(t) is
∂t ∂x  ∂x  the response function of the material which monotonically
where W ( x, t ) is the moisture content of cellulose at position x decreases with time.
The Fourier transform of the response function f (t ) yields
and time t. D denotes the diffusion coefficient and is usually 
not a constant but depends on both moisture concentration and the complex susceptibility, χ (ω )
∞
temperature. By conducting laboratory experiments,  
(ω ) ∫ f (t )e = χ=
(ω ) χ ' (ω ) − j χ " (ω ) (3)
− iωt
researchers have derived diffusion coefficients for different f= dt
0
types of cellulose (i.e. paper or pressboard, non-impregnated
or impregnated, aged or non-aged, paper/pressboard with The real and the imaginary parts of the complex susceptibility
different thickness) under different local moisture are not independent from each other since they are both
concentration and temperatures [2-9]. However, certain generated by the same response function f (t ) . They can be
variations may exist in these derived diffusion coefficients regarded as the cosine and sine transforms of the response
since the experiments were conducted with different setups function respectively.

and conditions. Such variations can pose difficulties for The total current density j (ω ) in the dielectric material
utilities on selecting the correct coefficient for a moisture 
under E (ω ) excitation can therefore be expressed as
content estimation in their transformers.
 
García et al. proposed a moisture-in-oil model for power j (ω )= iωe 0 [e ∞ + c ' (ω ) − i (s dc / e 0ω + c " (ω ))]E (ω ) (4)
transformer monitoring [8]. The model considered the )')( )))'))) (
capacitive part resistive part
equilibrium relations between oil and cellulose, the moisture
dynamics before attaining the steady-state equilibrium and the where σ dc denotes dc conductivity of the material. This
increase of the amount of moisture due to cellulose ageing. expression shows that the current is composed of a resistive part
However, it is not clear whether moisture in cellulose can be and a capacitive part. The resistive part represents the energy
derived from this model. losses in the material, which are dominated by two different
Originally proposed by Piper and Fessler [9], the isothermal mechanisms, one is due to dc conduction (movement of free
approach has also attracted the attention of some researchers charges) and the other is because of relaxation losses (re-
[10]. This approach uses temperature and vapor pressure to orientation of bonded charges). The capacitive part is associated
derive the local moisture content in cellulose from the with the capacitance of the material.
measurement of moisture-in-oil sensor. However, the sensor’s In many situations, it is more convenient to use the complex
location has a large influence on the moisture estimation. Given permittivity instead of the complex susceptibility. It can be
the complexity of a transformer’s construction and different defined as follows
types of cellulose used at different layers and locations inside the  
transformer, a sufficient number of moisture-in-oil sensors need = j (ω ) iωεε 0 [ (ω ) − iε (ω )]E (ω )
' "
(5)
to be installed to provide an accurate estimation of moisture in where ε ' (ω=
) ε ∞ + χ ' (ω ) and=
ε " (ω ) σε
dc / 0ω + c (ω ) .
"

the cellulose insulation of a transformer.

The above equations show that the dc conductivity σ dc , the
An alternative technique for estimating moisture content in
cellulose is the dielectric response method. It measures several high frequency component of the relative permittivity ε ∞ , and

dielectric response parameters, which are influenced by the complex dielectric susceptibility χ (ω ) , characterise the
temperature, moisture content and the ageing condition of the dielectric material in the frequency domain. It is possible to
cellulose insulation in a transformer. By using a database built determine these parameters by measuring the magnitude and
 3

phase angle of the resultant currents when the material is In Fig. 1, each layer of pressboard can be represented by a
subjected to an alternating voltage under different frequencies. parallel circuit consisting of a capacitor and a resistor. For the
i-th layer of the pressboard, its admittance can be written as
B. Dielectric Response Measurement for Moisture Estimation (7)
For estimating moisture content in a transformer, the According to frequency domain dielectric spectroscopy
geometric information of the winding insulation of the theory [11], (6) can be rewritten as (8)
transformer under investigation is needed. For a core type
transformer the main insulation usually consists of a number of
(8)
cylindrical shells of pressboard barriers, separated by axial
spacers. A so-called X-Y model is widely used to represent where denotes the total area of the cylindrical pressboard,
such a structure [6]. By making use of a database and X-Y
is the thickness of each sliced layer of the pressboard,
model, an algorithm can be implemented to find the best fit
between the response of the model and the measured response denotes the real part of complex permittivity of the i-th layer
of the transformer. This result can then be used for moisture of pressboard and represents the summation of conductive
estimation of a transformer’s cellulose insulation. and polarization losses of the i-th layer of pressboard (the loss
The above moisture estimation assumes the transformer of the whole piece of pressboard is shown in (5)).
under test is kept at a constant temperature and attains Assuming that the above pressboard can be sliced infinitely
moisture equilibrium. However, equilibrium is difficult to thin, (8) can be rewritten as (9)
attain in the transformer during measurement. It is therefore
(9)
necessary to investigate the correlation between temperature
dependent moisture dynamics and the dielectric response of
In dielectric response measurement, the phases and
cellulose insulation. This will pave a way for properly
magnitudes of the applied AC voltage and resultant current can
evaluating moisture in cellulose insulation of the transformer
be measured. Subsequently, the impedance of the
onsite, which is under a temperature transient.
In the next section a distributed parameter model is pressboard can be obtained. Then the complex capacitance of
proposed for studying the dielectric behavior of oil and the pressboard can be calculated by using (10)
cellulose insulation under non-equilibrium and non-uniform (10)
moisture distribution in a transformer. Combining (7) and (10), the complex capacitance of the
pressboard is expressed in (11)
IV. DISTRIBUTED PARAMETER DIELECTRIC RESPONSE MODEL (11)
The proposed distributed model is shown in Fig. 1. In this On the other hand, from (9) and (10) the complex
model, the pressboard with thickness is sliced into N layers capacitance of the pressboard can also be directly written as
( ). Assuming moisture in the transformer is in a
(12)
non-equilibrium state, the moisture is not uniformly distributed
along the thickness of the cellulose insulation. Consequently, By comparing (11) and (12), it can be expressed as
each layer of pressboard has a different dielectric loss and (13)
permittivity .
Eq. (13) reveals that in (8) equals the real part of
The impedance of the whole piece of the pressboard
permittivity while in (8) can be calculated from the
in the frequency domain can be written as
imaginary part of permittivity . Both the real and imaginary
(6) parts of permittivity of pressboard can be obtained from the
results of dielectric frequency response measurements. The
where denotes the total number of sliced layers of the above distributed model will be used to investigate the
pressboard, and are the impedance and moisture dynamics effect on dielectric response measurement
admittance of the i-th layer of the pressboard respectively. in the following sections.

V. EXPERIMENTAL SETUP

A. Prototype Transformer Configuration

A prototype transformer is used for experimental study in
this paper. It is a single phase transformer rated at 5 kVA with
240 V secondary and 2.2 kV primary. Since the losses due to
current flowing through the transformer’s winding may not
generate the required temperature, a heater was installed at the
Fig. 1 Distributed model for modelling dielectric response of pressboard
 4

bottom of the transformer. This also helped to control the cycle sinusoidal temperature was 24 hours with the highest
prototype transformer’s temperature to simulate different temperature of 80 °C and the lowest temperature of 30 °C.
thermal conditions for the accelerated ageing and moisture (3) The transformer was kept at a constant temperature
diffusion experiments. Fig. 2 shows the geometry of the (55 °C) for another one week to let it attain moisture equilibrium.
insulation of the prototype transformer. Dielectric response measurement was then performed.
Fig. 3 presents one complete cycle of moisture
measurements under sinusoidal temperature. In the figure, the
temperature (in green), moisture in oil (in red) and water
activity (in blue, is defined as the ratio of the partial pressure
of water in the material and the saturated vapor pressure of
pure water at the same temperature [10]) were directly
obtained from the above moisture-in-oil sensor.
360 55 50 0.85
0.80
350 Temperature 50 0.75
340 0.70
45 40
Fig. 2 Insulation geometry of the prototype transformer 330 0.65
0.60
40

Moisture in paper(%)
Moisture in oil(ppm)
320 0.55
In the prototype transformer, the cellulose insulation

Temperature(K)

Water activity
310 Moisture in oil 30 0.50
35
0.45
includes hybrid Diamond Dotted Paper (DDPP), Mouldable 300
30 0.40

pressboard (MPB) and Spacers All Round (SAR) and they are 290
25
20 0.35
0.30
280
immersed in transformer oil (Nynas Libra). The temperature 270 Moisture in paper 20
0.25
Water activity 0.20
class of the transformer’s paper insulation is A. The ratio in X- 260 15
10
0.15

Y model is calculated= =
0.10
as X 38%, Y 16% , where X is the 250
10 0.05
240 0 0.00
lumped sum of the thickness of all barriers in the duct divided 0 50 100 150 200 250 300 350 400
Time (h)
by the duct width; Y is the total width of all spacers divided by
Fig. 3 Sinusoidal variation of temperature (green), water activity in oil (blue),
the total length of the periphery of the duct [6]. moisture in oil (red, in ppm) and water content at the cellulose surface in
contact with oil (black, in percentage) of the prototype transformer
B. Moisture Dynamics Experiments
After the commission of the prototype transformer, it was The moisture concentration at the cellulose surface in
subjected to both electrical and thermal loading to attain a contact with oil (in black in Fig. 3) was obtained using Fessler
certain degree of ageing of its cellulose insulation (equivalent to [9] equation as
34 years of life consumption based on the degree of W = 2.173 × 10−7 × Pv0.6685 × e 4275.6/T (14)
polymerization (DP) measurement of this transformer). where W is the concentration of absorbed water as the ratio of
Electrical loading was imposed by using a load bank with the mass of water to the mass of dry cellulose, Pv is the vapour
maximum power capacity of 6 kW. The prototype transformer pressure of water in atm and T is the temperature in Kelvin.
was kept at 110 ºC (using the abovementioned heater) with 30A
load current for an effective time period equal to 35 days (the VI. MODELLING OF MOISTURE DYNAMICS IN CELLULOSE
transformer was kept at 50 ºC during the night and weekends).
This section investigates the above sinusoidal temperature
After 35 days of accelerated electrical and thermal ageing
driven moisture dynamics in cellulose. The time of moisture
of the prototype transformer, experiments were arranged to
diffusion inside cellulose is much larger than that of moisture
study the moisture dynamics and its effect on dielectric
migration at the cellulose’s surface contacting oil. Therefore,
response of the cellulose insulation of the transformer. In the
this section mainly focuses on moisture diffusion in cellulose.
experiments, a sinusoidal temperature profile was imposed on
This section firstly models moisture diffusion in un-aged
the prototype transformer by using the above mentioned heater.
pressboards. Then it investigates moisture diffusion in the
The prototype transformer is sealed from the atmosphere. A
prototype transformer based on the experiments in Section V.
moisture-in-oil sensor (Vaisala MMT 330 [12]) was installed
in the prototype transformer. The sensor’s tip was close to the A. Modelling Moisture Diffusion in Pressboard
cellulose insulation of the winding. By using this setup, the It is assumed an un-aged dry pressboard (1 mm in thickness
moisture at the interface between oil and cellulose insulation with less than 0.5% water content) has one side of its surface
could be continuously monitored. Temperature was also contacting with a large volume of oil, which contains a
measured by this sensor. sufficient amount of water. When the pressboard is immersed
The moisture dynamics experiments were conducted in into the oil, the partial pressure of water in oil is significantly
three steps: larger than that in pressboard. Thus, the pressboard absorbs
(1) The transformer was kept at a constant temperature water from the oil when it is just immersed into the oil. The
(55 °C) for one week to let it attain moisture equilibrium. other side of the pressboard surface is sealed by an electrode
(2) The transformer was subjected to a sinusoidal for dielectric response measurement. The whole test cell is
temperature profile (Fig. 3) for one week. The period of one considered as a closed system. Under sinusoidal temperature
 5

variation, the moisture content at this side of the surface hours sinusoidal temperature cycle. It can be seen from Fig. 4(d),
follows the curve in black shown in Fig. 3. the moisture distribution at the surface of the pressboard at the
The initial values of moisture content and boundary time instances t = 24, 48 and 72 hours are very close (the
condition of this pressboard is difference is less than 0.2%). The difference of moisture in the
W ( x ,=
t = W0，W0 < 0.5
0) depth of the pressboard between t = 24, 48 and 72 hours is about
(15)
W (=
x 0, t=
) Ws，∂W ( =
x l , t ) / ∂=
x 0 0.5%. Such a difference can be explained as follows.
where W0 is the moisture content in the whole bulk of un-aged After several cycles of diffusion, the pressboard becomes
humid and water migration is mainly dominated by temperature.
pressboards and is assumed less than 0.5% initially. Ws is the
At low temperature (30 ℃ at time instances t = 24, 48, 72 hours)
concentration of moisture at the interface between pressboard the slow water molecules’ mobility and strong bond between
and oil and it varies as sinusoidal shape (black curve in Fig. 3). cellulose molecules and water molecules imply that the moisture
By solving Fick’s second law (1) and (15), the moisture movement is not able to change its directions immediately when
distribution along the thickness of the pressboard under temperature changes. Therefore, a certain amount of moisture
sinusoidal temperature profile from the time instances t = 1 to 72 moves “back and forth” at a particular position of pressboard and a
hours are obtained and depicted in Fig. 4. “standing wave” of moisture distribution can be formed.
7.5
6.5 24 hours
1 hour 5.0 12 hours 7.0
80
6.0
16 hours 48 hours
4 hours 6.5
5.5 4.5 20 hours 72 hours
8 hours 6.0

Moisture concentration (%)

Moisture concentration (%)
5.0
Moisture concentration (%)

70 4.0 5.5
4.5
Temperature (C)

5.0
4.0 3.5
60 4.5
3.5
3.0 4.0
3.0
50 3.5
2.5
2.5
2.0 3.0

40 1.5 2.0 2.5

1.0 2.0
1.5
0.5 1.5
30
0.0 1.0 1.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
0 4 8 12 16 20 24 28 32 36 40 44 48
Time (h) Depth from surface (mm) Depth from surface (mm) Depth from surface (mm)

(a) (b) (c) (d)

Fig. 4 Simulation results of moisture distribution in an un-aged pressboard (one dimension moisture diffusion) (a) temperature variation at different time
instances (b)-(d) moisture distribution at different time instances (x=0 refers to the pressboard surface in contact with oil).
It is assumed that at t = 0 when the dry and un-aged
B. Modelling Moisture Diffusion in Prototype Transformer
pressboard sample was put into the oil tank, there exists a
significant moisture gradient between the oil and the pressboard Moisture distribution is not only a function of temperature
and at their interface vapor pressure of the oil is larger than that and moisture gradients but also a function of mass and
of the pressboard. Therefore, at t = 0 hour, the pressboard dimensions of the transformer’s insulation system. Therefore,
absorbs water from the oil and eventually water moves towards multi-physics modeling [13] is adopted to investigate moisture
the depth of the pressboard. At the time instance t = 1 hour, there distribution in three dimensions (3D) of the prototype
exists a large moisture gradient in the pressboard: moisture transformer as shown in Fig. 5. It considers the effects of
concentration at the pressboard surface is 5.7% and only 0.5% at electromagnetic, thermal, fluid flow and moisture migration
other locations away from the surface (Fig. 4(b)). This moisture physics on moisture dynamics in transformers. It especially
gradient dominates the water migration and water will takes into account the coupling and interactions of these
continuously move into the pressboard along its thickness for physics as they collectively influence the moisture dynamics
some time though temperature starts to increase. and moisture distribution inside the transformer. The details of
At the time instances t = 4 and 8 hours, the moisture moves multi-physics modeling will be provided in a future paper.
further into the depth of the pressboard. The moisture at the 7
2.1

2
pressboard surface is decreased from 4% to 2.1% while the 6
Average moisture concentration(%)

1.9
moisture in most positions of the pressboard increased to above 5
1.8
4
1% from 0.5% (Fig. 4(b)). At the time instance t = 12 hours, the 1.7
3
moisture at the surface of the pressboard almost attains uniformity 1.6
2
of 1.75%. As can be observed from Fig. 4(a), at t = 12 hours the 1.5
1
temperature starts to decrease from 80 °C. The water in the oil 0
1.4
0 10 20 30 40 50 60 70 80 90

(close to the oil-pressboard interface) starts to move towards the (a) (b)
Time (h)

pressboard. Therefore, at the time instances t = 16 and 20 hours, Fig. 5 Moisture distribution of the prototype transformer (a) 3D moisture
the moisture concentration at the pressboard surface (in contact distribution (at time instance t = 72 hours); (b) average moisture
concentration at different time instances. Transformer was under thermal
with oil) increased. The moisture at the pressboard’s surface is
transients as shown in Fig. 3.
2.5%, 5.0% for t= 16 and 20 hours respectively (Fig. 4(c)). The
Fig. 5(a) presents the moisture distribution in the prototype
moisture at the pressboard surface reaches its peak of 6.4% at the
transformer at the time instance t = 72 hours. Based on the 3D
time instance t = 24 hours. Also the moisture inside the pressboard
moisture distribution results, the average moisture concentration
is larger compared to previous time instances in this first
of the whole bulk of cellulose insulation can be calculated
temperature cycle (Figs. 4(b) and 4(c)).
The above moisture dynamics are repeated for the next 24 ∫∫∫ W ( x, y, z ) dV / V , where V is the total bulk
as Waverage = �
W
 6

volume of the cellulose insulation and W ( x, y , z ) is the moisture estimation is verified by the experiments under sinusoidal
concentration at a particular location of the cellulose insulation. temperature profile. It needs to be mentioned that the analysis
Fig. 5(b) presents the calculated average moisture concentration in step 1 does not include the influence of the temperature on
of the cellulose insulation at different time instances. It can be the dielectric response. Therefore, the temperature corrections
observed from Fig. 5(b) that moisture variations in sinusoidal will be conducted (detailed in Section VII C).
shape are within the range from 1.45% - 2.0%. The overall A. Dielectric Loss of Cellulose Insulation vs. Moisture
moisture of the cellulose insulation of the prototype transformer
After the prototype transformer attains moisture equilibrium
is 1.7 % by taking the average of data in Fig. 5 (b). It has good
and the moisture is uniformly distributed (Section V B, Step 3),
agreement with the results from the Karl Fischer Titration
dielectric frequency response measurement is performed from 1
method, which indicates 1.5% moisture content in a cellulose
kHz to 1 mHz. Fig. 7 presents both real and imaginary parts of
specimen collected from the prototype transformer.
permittivity of the cellulose insulation at 50°C with four
Fig. 6 shows the moisture distribution in the prototype
different moisture contents, i.e. 1.08%, 1.37%, 1.61% and
transformer’s whole bulk cellulose insulation along its radial
3.23% in the prototype transformer. It can be seen from Fig. 7
centreline under different time instances (temperature varies
according to Fig. 3). In Fig. 6 the legend of the vertical axis is in that the imaginary part of permittivity is influenced by moisture
ppm for moisture concentration in oil and it is in percentage for concentration. At a particular frequency, the imaginary part of
moisture concentration in cellulose insulation. From Fig. 6 it can permittivity increases when more moisture resides in the
be seen that the moisture is not uniformly distributed inside the cellulose insulation.
cellulose insulation of the prototype transformer. A large 1000
e'-1.08%
100
e'-1.37%
proportion of the cellulose has a relatively low moisture level e'-1.61%
e'-3.23%

Imaginary part of permittvity (e'')

10
(less than 3%) while the moisture of the cellulose surface in

Real part of permittvity (e')

e"-1.08%
e"-1.37%
contact with oil (x = 0 and 0.4 m) may reach up to 5% - 7%. The 100 e"-1.61%
e"-3.23%
1
moisture distribution along the transformer height direction also e''
exhibits gradients (Fig. 5(a)). This is due to the non-uniform 0.1
thermal distribution along the height direction and the existence 10

of a fluid flow field inside the prototype transformer. e' 0.01

Moisture concentration in cellulose insulation (%)

1 1E-3
1E-3 0.01 0.1 1 10 100 1000
Frequency (Hz)

Fig. 7 Real (ε’) and imaginary (ε”) parts of permittivity of the prototype
transformer insulation under uniform moisture distribution at T = 50 °C
The dielectric loss of cellulose insulation under each
frequency with different moisture contents can be obtained from
Fig. 7 and (13), which is shown in Fig. 8 (data points). Suppose
the dielectric loss and the moisture can be modelled as
(16) δ = α1e β ⋅W 1

Fig. 6 Moisture distribution of the whole bulk cellulose insulation in the where W denotes the moisture concentration in the cellulose
prototype transformer (d = 0.2 m denotes the symmetric line of the cellulose insulation, α1 and β1 are parameters, which can be computed
insulation). Temperature varies according to Fig. 3.
using the least squares method. Fig. 8 presents the modelled
VII. ANALYSIS OF MOISTURE DYNAMICS’ EFFECT ON curves of dielectric loss with respect to moisture contents in the
DIELECTRIC RESPONSE OF CELLULOSE INSULATION cellulose insulation under uniform moisture distribution at 50°C.

Based on the experiments and modeling results provided in 1000Hz

10Hz
1Hz
0.1Hz

Sections V and VI, this section applies the distributed parameter 0.01Hz

model proposed in Section IV to study the moisture dynamics’ 1E-10

Loss component

effect on dielectric response of cellulose insulation of the

prototype transformer. The study is carried out in two steps:
(1) By using a database built upon the dielectric response 1E-11

measurements on the prototype transformer under constant

temperature, the distributed model is applied to explore the
correlations between the cellulose’s position along its thickness
1E-12
and the cellulose’s dielectric response parameters (loss and 1.2 1.5 1.8 2.1 2.4 2.7 3.0 3.3
Moisture concentration(%)
permittivity) under uniform moisture distribution conditions.
Fig. 8 Dependency between dielectric loss and moisture contents under
(2) Based on the analysis in step 1, the dielectric response uniform moisture distribution condition. T = 50 °C.
of the cellulose insulation in the prototype transformer under
Once the relationship between dielectric loss and moisture
moisture dynamics and temperature transient is estimated. The
content is obtained by using Fig. 6 and (16), the dielectric loss of
 7

5
the cellulose insulation at different diffusion positions (x axis in 1000Hz
10 Hz
Fig. 6) at any specific time instance can be estimated. The 4 1 Hz
0.1 Hz
estimated values at time instance t = 6 hours are shown in Fig. 9 0.01 Hz

Real part of permittivity

3
(denoted as data points). Due to the axis symmetry, only half of
the curve (x axis from 0 - 0.2 m) in Fig. 6 is presented.
Suppose that the dielectric loss and diffusion position of the 2

cellulose has an exponential relationship

δ = α 2 e β2 ⋅ x (17)
where x denotes the position of cellulose insulation, α 2 and 1

β 2 are parameters which can be computed using the least 0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16 0.18 0.20

squares method. The modelled dielectric loss with respect to Diffusion position (m)

the position of cellulose at the time instance t = 6 hours is Fig. 11 Dependency between real part of permittivity and the depth of
presented in Fig. 9 (denoted as curves, T = 50 °C). It should be cellulose insulation (at time instance t = 6 hours).
mentioned that the curves in Fig. 9 are applied to non-
C. Validation of Distributed Parameter Model
equilibrium and non-uniform moisture distribution in the
cellulose insulation. After obtaining the complex permittivity through the
1E-8 procedures described in Section VII A and VII B, dielectric
1000Hz
10Hz
1Hz
response measurement is performed on the prototype
0.1Hz
0.01Hz
transformer. The comparison between the measured and the
1E-9
estimated complex permittivities is then made. The
Loss component

comparison provides a validation for the proposed distributed

1E-10
parameter model.
The dielectric response measurement was repeated every
hour for 24 hours while the sinusoidal temperature profile (the
1E-11 same as shown in Fig. 3) was imposed on the prototype
transformer. During measurement, sinusoidal temperature
0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16 0.18 0.20
variation can significantly influence the dielectric response of
Diffusion position (m) cellulose, especially for low frequencies because it takes a
Fig. 9 Dependency between loss and the depth of cellulose insulation (at time relatively longer time to take measurement at low frequencies.
instance t = 6 hours diffusion). Data points denote the values obtained from The dependency between permittivity and temperature [11]
(16) and Fig. 6; Curves are obtained using (17).
can be expressed as (18)
B. Permittivity of Cellulosic Insulation vs. Moisture
e 'T =
e r∞ +
e rs + e r∞
，e '' (e rs − e r∞ ) ωτ h e H / kT (18)
=
The above approaches are also applied to investigate the
1 + ω 2τ h 2 e 2 H / kT 1 + ω 2τ h2 e 2 H / kT
T
correlation between the real part of permittivity of cellulose
insulation and moisture content. Fig. 10 depicts the modelled where ε rs is the dielectric constant at zero excitation frequency
curves of the real permittivity with respect to moisture
or dc value, ε r ∞ is the dielectric constant at very high frequency,
concentrations under different frequencies at T = 50 °C. Fig. 11
depicts the modelled real permittivity with respect to the depth τ h is a pre exponential factor, H is the activation energy, T is
of cellulose insulation at T = 50 °C. In Fig. 11, the modelled real the absolute temperature and k is the Boltzmann constant.
part of permittivity is for the time instance t = 6 hours. As the dependency between dielectric response and
10
9 1000Hz
10Hz
diffusion positions in Section VI is obtained under a constant
8 1Hz
7 0.1Hz temperature of 50℃, necessary temperature corrections are
0.01Hz
6
needed to compensate the influence of the temperature on the
Real part of permittivity

4
complex permittivity in dielectric response modelling when
sinusoidal temperature is applied. The temperature correction
3
is presented in Fig. 12.
2 In Fig. 12, for each round of dielectric response
measurements, the average temperature at a particular
frequency is corrected by Tave = (1 / Pi ) ∫ Tdt , where Pi is the
P i

1 0

1.2 1.5 1.8 2.1 2.4 2.7 3.0 3.3 total measurement duration at the i-th frequency and T is the
Moisture concentration(%)
temperature at different time instances as depicted in Fig. 4 (a).
Fig. 10 Dependency between real part of permittivity and moisture contents
By substituting Tave into T of (18), the estimated real and
under uniform distribution condition. T = 50 °C.
imaginary part of permittivity at each frequency under
sinusoidal temperature can be corrected. The estimated and
 8

measured real and imaginary parts of permittivity of cellulose

insulation after temperature correction under non-equilibrium User Input Steps of Software

and non-uniform moisture distribution at different times are Essential: 1. Database of dielectric response.
1. Nameplate information. 2. Moisture distribution model.
drawn in Fig. 13. 2. Dielectric response 3. Analysis
From Fig. 13 it can be observed there is a similar trend in measurement. a) Modelled dielectric response of the
transformer under thermal transients.
the estimated and measured dielectric responses at different Optional: b) Corrections on the dielectric response
time instances. Some degree of discrepancy between the 1. Geometry information. measurement.
2. Historic dielectric response c) Moisture distribution of the
estimated and measured permittivities can be observed: (1) at measurement dataset.
transformer under thermal transients.
low frequencies, which may be due to the modeling errors (can Dielectric response 4. Results output
measurement
be observed in Fig. 9 and Fig. 11); and (2) at time instance t =
6 hours, which may be due to the lack of data for high Results
moisture content in modeling (Fig. 8 and Fig. 10). 1. Moisture distribution.
2. Average moisture content.
From Fig. 13 it can also be seen that there is a decreasing
trend in both estimated and measured real permittivities at Fig. 14 Components of software tool for moisture estimation
different time instances (i.e. different moisture distribution in
the cellulose insulation). When the moisture concentration (1) Database of dielectric response of transformers: This
increases, the real permittivity also increases at each database consists of dielectric response measurement from
frequency. The imaginary permittivity increases throughout the various transformers of different manufacturers. Users can also
whole frequency range when more moisture moves into expand the database by including dielectric response
cellulose insulation. However, the increase of the imaginary measurement results of their transformers. If the transformer of
part of permittivity at low frequencies is more dominant. interest is not included in the database and the user doesn’t
355 have any historic measurement dataset of this transformer, the
R1
350 R3
R6
datasets of other transformers with similar rating, insulation
345 R9
R12
geometry and age will be utilized by the software in its
340 R15
R18 calculation.
335
Temperature (K)

R21

330
R24 (2) Moisture distribution model of transformers: The
325 software tool consists of moisture distribution models of many
320 transformers, which are developed from transformers’
315 geometrical information of different manufacturers. For the
310 transformer of which the geometry information is not
305
available, the software will approximate its geometry by using
300
1E-3 0.01 0.1 1 10 100 1000 the available information of the transformer (e.g. rating, core
Frequency (Hz) type, temperature rise, oil volume etc.). It is assumed that the
Fig. 12 Temperature correction (Rj denotes the j-th round of dielectric
response measurement, data points denote the maximum temperature at each transformer’s insulation construction follows common
frequency, solid lines denote the corrected temperature at each frequency) engineering practice and meets certain criteria (standards). It is
100
ε'-1 hour modεllεd
100 also assumed that the moisture dissolved in oil is measured (by
ε'-2 hours modεllεd
ε'-3 hours modεllεd
either water-in-oil probes or DGA/oil test of the transformer).
10
ε'-4 hours modεllεd
Imaginary part of pεrmittivity(ε'')

ε'-5 hours modεllεd

Based on the above approximated transformer geometry and
Rεal part of pεrmittivity(ε')

ε'-6 hours modεllεd

ε''-1 hour modεllεd 1 the measured moisture in oil, the software will produce a
ε'' ε''-2 hours modεllεd
moisture distribution model for the transformer of interest.
ε''-3 hours modεllεd
10
ε''-4 hours modεllεd 0.1
ε''-5 hours modεllεd (3) Analysis: Users conduct dielectric response measurement
ε''-6 hours modεllεd
0.01 on the transformer of interest and input the dielectric response
ε' data to the software. The software will automatically perform
1E-3 the following calculations:
1
1E-4
(a) Combine the above database and moisture distribution
1E-3 0.01 0.1 1
Frεquεncy (Hz)
10 100 1000
model to compute the dielectric response of the transformer
Fig. 13 Comparison between measured and estimated dielectric response considering the effect of temperature variations.
(temperature varies according to Fig. 3, from T = 30°C to T = 80°C. Lines (b) Apply the computed dielectric response in (a) to make
denote the modelled data and dots denote the measured data)
corrections on the measured dielectric response of the
To improve the applicability of the proposed method a transformer. Such correction can be done by adjusting the
software tool is implemented, which can be used by practicing parameters (α and β in Equations 16 and 17) in the model until
engineers as an alternative to current dielectric response the error between computed and measured dielectric response
measurement analysis software available with the commercial is minimized.
products. Fig. 14 illustrates the major components of this software
(c) Compute the moisture distribution inside the
tool, including:
transformer’s cellulose insulation under thermal transients.
 9

(4) Output: The software will provide the following [9] W. A. Fessler, T. O. Rouse, W. J. McNutt and O. R. Compton, "A refined
mathematical model for prediction of bubble evolution in transformers,"
information. IEEE Trans. Power Del., vol.4, pp. 391-404, 1989.
(a) Moisture distribution of transformer’s cellulose [10] D. Martin, C. Perkasa and N. Lelekakis, "Measuring paper water content
insulation. of transformers: a new approach using cellulose isotherms in
nonequilibrium conditions," IEEE Trans. Power Del., vol.28, pp. 1433-
(b) Average moisture content in transformer’s cellulose 1439, 2013.
insulation. [11] A. K. Jonscher, Dielectric Relaxation in Solids, London: Chelsea
Dielectrics Press, 1983.
With the advancement of sensor technology, more sensors [12] "MMT330 Moisture and Temperature Transmitter Series for Oil"
can be installed at critical positions for continuously monitoring available at http://www.vaisala.com/en/industrialmeasurements/products
/moistureinoil/Pages/MMT330.aspx.
the temperature and moisture in transformer oil. By integrating [13] M. A. Tsili, E. I. Amoiralis, A. G. Kladas and A. T. Souflaris, "Power
the online sensor based monitoring, modeling approach, and transformer thermal analysis by using an advanced coupled 3D heat
dielectric response measurement, the reliability of estimation of transfer and fluid flow FEM model," Int. J. Therm. Sci., vol.53, pp. 188-
201, 2012.
moisture and its distribution in transformer insulation system can
be improved by the proposed method described in this paper.
IX. BIOGRAPHIES
VIII. CONCLUSIONS Yi Cui (S’12) received the B.Sc. (Engg.) and M.Sc
This paper developed a distributed parameter model to (Engg.) in Electrical Engineering from Southwest
Jiaotong University, China in 2009 and 2012
interpret the dielectric response of cellulose insulation of
respectively. He is currently pursuing the Ph.D.
transformers under thermal and moisture transients. By using degree in the School of Information Technology and
the methodologies proposed in this paper, the dielectric Electrical Engineering, the University of
response of a transformers’ cellulose insulation under thermal Queensland, Australia. He is a member of IEEE
since 2012. His research interests include
transients and moisture non-equilibrium condition can be condition assessment and fault diagnosis of power
estimated. It is expected that the proposed methodology can transformer.
provide proper interpretation of dielectric response
measurement and accurate moisture estimation for Hui Ma (M’95) received his B.Eng and M.Eng
transformers at substations. degrees from Xi’an Jiaotong University, China
in 1991 and 1994, M.Eng (by research) degree
from Nanyang Technological University,
ACKNOWLEDGMENT
Singapore in 1998, and PhD degree from the
The authors gratefully acknowledge Australian Research University of Adelaide, Adelaide, Australia in 2008.
Council, and industry partners Powerlink Queensland, Energex, Currently Dr. Ma is a research fellow in the School
of Information Technology and Electrical
Ergon Energy, and TransGrid for supporting this work. Engineering, the University of Queensland,
Australia. Prior to joining the University of
REFERENCES Queensland, Dr. Ma has many years research and development
experience. From 1994 to 1995, he was a researcher in Xi’an
[1] L. E. Lundgaard, W. Hansen, D. Linhjell and T. J. Painter, "Aging of oil-
JiaotongUniversity, China. From 1997 to 1999, he worked as a firmware
impregnated paper in power transformers," IEEE Trans. Power Del.,
development engineer in CET Technologies Pte. Ltd., Singapore. He was
vol.19, pp. 230-239, 2004.
with Singapore Institute of Manufacturing Technology as a research
[2] Pahlavanpour, M. Martins and Eklund, "Study of moisture equilibrium in
engineer from 1999 to 2003. His research interests include industrial
oil-paper system with temperature variation," in Proceedings of the 7th
informatics, condition monitoring and diagnosis, power systems, wireless
International Conference on Properties and Applications of Dielectric
sensor networks, and sensor signal processing.
Materials, 2003,1124-1129.
[3] T. V. Oommen, "Moisture equilibrium charts for transformer insulation
drying practice," IEEE Transactions on Power Apparatus and Systems, Tapan Kumar Saha (M’93, SM’97) was born
vol.103, pp. 3062-3067, 1984. in Bangladesh in 1959 and immigrated to Australia
[4] Y. Du, M. Zahn, B. C. Lesieutre, A. V. Mamishev and S. R. Lindgren, in 1989. He received his B.Sc Engineering
"Moisture equilibrium in transformer paper-oil systems," IEEE Electr. (electrical and electronic) in 1982 from the
Insul. Mag., vol.15, pp. 11-20, 1999. Bangladesh University of Engineering &
[5] B. Buerschaper, O. Kleboth-Lugova and T. Leibfried, "The electrical Technology, Dhaka, Bangladesh, M. Tech
strength of transformer oil in a transformerboard-oil system during (electrical engineering) in1985 from the Indian
moisture non-equilibrium," in Proceedings of Annual Conference on Institute of Technology, New Delhi, India and
Electrical Insulation and Dielectric Phenomena, 2003,269-272. PhD in 1994 from the University of
[6] C. Ekanayake, S. M. Gubanski, A. Graczkowski and K. Walczak, Queensland, Brisbane, Australia. Tapan is
"Frequency response of oil impregnated pressboard and paper samples currently Professor of Electrical Engineering in the School of
for estimating moisture in transformer insulation," IEEE Trans. Power Information Technology and Electrical Engineering, University of
Del., vol.21, pp. 1309-1317, 2006. Queensland, Australia. Previously he has had visiting appointments for a
[7] A. F. Howe, "Diffusion of moisture through power-transformer semester at both the Royal Institute of Technology (KTH), Stockholm,
insulation," Proceedings of the Institution of Electrical Engineers, Sweden and at the University of Newcastle (Australia). He is a Fellow of the
vol.125, pp. 978-986, 1978. Institution of Engineers, Australia. His research interests include condition
[8] B. Garcia, J. C. Burgos, A. M. Alonso and J. Sanz, "A moisture-in-oil monitoring of electrical plants, power systems and power quality.
model for power transformer monitoring - Part I: Theoretical
foundation," IEEE Trans. Power Del., vol.20, pp. 1417-1422, 2005.
 10

Chandima Ekanayake (M’00) received his

B.Sc.Eng.(Hons) in 1999 from University of
Peradeniya Sri Lanka. He obtained his Tech. Lic.
and PhD from Chalmers University of Technology
Sweden in 2003 and 2006 respectively. Currently he
is a lecturer in the School of Information
Technology and Electrical Engineering, the
University of Queensland (UQ), Brisbane, Australia.
Before joining UQ he was with University of
Peradeniya Sri Lanka as a Senior lecturer. During
his PhD studies he was working for a European Union Project called
REDIATOOL where he engaged in research related to Diagnostics of
Transformer Insulation from dielectric response measurements. From 2001,
he has been involving on condition monitoring of transformers installed at
Ceylon Electricity Board, Sri Lanka. He was the Chair of IEEE Sri Lanka
Section in year 2006 and 2007. His research interests are condition
monitoring of power apparatus, alternatives for insulating oil, transient
studies on power systems and energy related studies.