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

net/publication/317146177

Assessing degradation of power transformer solid insulation considering

thermal stress and moisture variation

Article in Electric Power Systems Research · October 2017

DOI: 10.1016/j.epsr.2017.04.006

CITATIONS READS

33 774

4 authors, including:

Ricardo David Medina A.A. Romero

23 PUBLICATIONS 160 CITATIONS

National Scientific and Technical Research Council
55 PUBLICATIONS 420 CITATIONS
SEE PROFILE
SEE PROFILE

Enrique Esteban Mombello

National University of San Juan
72 PUBLICATIONS 1,072 CITATIONS

SEE PROFILE

All content following this page was uploaded by Ricardo David Medina on 15 February 2018.

The user has requested enhancement of the downloaded file.

 Electric Power Systems Research 151 (2017) 1–11

Contents lists available at ScienceDirect

Electric Power Systems Research

journal homepage: www.elsevier.com/locate/epsr

Assessing degradation of power transformer solid insulation

considering thermal stress and moisture variation
Ricardo D. Medina a,b,∗ , Andrés A. Romero b , Enrique E. Mombello b , Giuseppe Rattá b
a
Grupo de Investigación en Energía GIE, Universidad Politécnica Salesiana Sede Cuenca, Calle Vieja 12-30 y Elia Liut, Cuenca EC 010102, Ecuador
b
Instituto de Energía Eléctrica, Universidad Nacional de San Juan, IEE – UNSJ – CONICET, Av. Libertador San Martín, 1109 (o), San Juan, J5400AR, Argentina

a r t i c l e i n f o a b s t r a c t

Article history: This paper presents a novel method for estimating the degradation of solid insulation in power transform-
Received 20 July 2016 ers, considering thermal ageing and paper moisture dynamics. Current ageing models are based on both,
Received in revised form 6 April 2017 experimental evidence and theoretical developments; considering that all models are approximation to
Accepted 7 April 2017
reality, loss of life estimation could be found in a large range depending on the evidence considered; this
amplitude could lead inaccurate results to make adequate decisions in an asset-management context.
Keywords:
These differences in results can be explained because the models consider only nominal operative temper-
Brownian random process
atures ranges overlooking low and high temperatures degradation process and the influence of variations
Insulation ageing
Life estimation
in paper moisture content. Considering the above, this document proposes a holistic methodology for
Oil-paper insulation solid insulation ageing assessing based on all thermal degradation process (oxidation, hydrolysis and
Paper moisture pyrolysis) and the influence of dynamics on paper moisture. Paper moisture is estimated using as input
Power transformer external variables such as: load, ambient temperature, transformer technical data and measurements
regarding oil moisture, in order to consider uncertain in oil moisture growing Arithmetic-Brownian-
Motion algorithms are presented. The proposed methodology was tested for four power transformers,
for which load and ambient temperature hourly profiles are available over a period of almost nine years.
In order to compare different degradation rates, three alternatives to model the chemical environment
in which cellulose is aged, are analysed. Results are presented, and conclusions are finally detailed.
© 2017 Elsevier B.V. All rights reserved.

1. Introduction DP value (that is related with the loss-of-life) can be used as an

alternative. For instance, according to IEEE C57.91 [4], the paper
Solid insulation failure is the leading cause of the end-of-life of loss-of-life is assessed as a function of the hot spot temperature
power transformers (PTs) [1]. Therefore, a measure of the state of ( HS ) during a time period, t, of PT operation. Such approach con-
the solid insulation (paper) is accepted as an indicator of the PT siders that heat is the main ageing agent, and assumes that both
condition [2]. Paper is composed by long chains of glucose rings humidity and acidity content in the oil-paper insulating system
that build the cellulose polymer molecule. The average length of remain constant.
these chains is termed degree of polymerization (DP). During the The following Arrhenius relation is widely accepted to model
PT life cycle, DP decreases due to ageing processes (i.e., oxidation, the ageing of paper [5].
hydrolysis and pyrolysis) [3], and consequently insulating paper
loses its dielectric and mechanical properties. 1 1 Ea
− = A · e R·T (t) · t (1)
Degree of polymerization can be directly measured from a sam- DP (t) DP (t0 )
ple of paper, but this practice for a PT in operation implies a
where DP(t0 ) and DP(t) are DP values at the start, t0 , and at the end,
non-desired disconnection and an invasive manipulation of the
t, of the analysed time interval t; Ea is the activation energy of
unit. Therefore, computational methods for indirect estimation of
the ageing reaction, expressed in [J/mol]; A is the pre-exponential
factor and it depends on the chemical environmental; R is the gas
constant (8.314 [J/mol/K]); and T is the temperature of the paper in
∗ Corresponding author at: Grupo de Investigación en Energía GIE, Universidad
[K], that is just  HS in the top of the windings, because at this place
Politécnica Salesiana Sede Cuenca, Calle Vieja 12-30 y Elia Liut, Cuenca EC 010102,
occurs the highest paper degradation in a PT [6].
Ecuador. A post-mortem evaluation of a PT was performed in order to
E-mail address: [email protected] (R.D. Medina). validate the Arrhenius models in Ref. [6]. The main conclusion of

http://dx.doi.org/10.1016/j.epsr.2017.04.006
0378-7796/© 2017 Elsevier B.V. All rights reserved.
2 R.D. Medina et al. / Electric Power Systems Research 151 (2017) 1–11

this work are that the available A and E values are not sufficiently formed. In order to overcome this limitation, the main contribution
accurate to estimate DP, because they do not include the combined of this paper can be summarised as follows:
effects of moisture, acidity and oxygen.
In Ref. [7], the thermal life expectancy of a fleet of 185 PTs was (1) Proposal of a practical methodology for computational estima-
estimated. For this purpose, Eq. (1) was employed but assuming tion of DP and the loss of life.
different values for A, in order to account for oxidation and hydrol- (2) The proposal takes into account both hot spot temperature and
ysis processes. However, it was assumed that moisture content in chemical environment variations, which are modelled consid-
paper is a discrete function of DP value. In fact, it was assumed ering uncertainty and lack of information.
that moisture content is 1.0, 1.5 and 2.0% within the DP range of (3) Proposed method add information on life-span estimations.
1000–500, 500–250 and 250–200. The average life expectancy of
185 transformers was quantified as 83 years. The novelty of this proposal is just a methodology which inte-
In Ref. [8], general lines for power transformer asset man- grates several pieces of evidence (e.g., paper moisture content
agement are proposed. Transformer administration considers interaction with paper ageing, paper and oil moisture relationship,
condition monitoring, maintenance plans and ageing assessment. paper moisture and temperature relationship, influence on ageing
Condition takes into account: vibration analysis, partial discharge, of oxidation, hydrolysis and pyrolysis, etc.) reported in literature,
frequency response analysis (FRA) and thermal analysis. Thermal that have not been connected until now.
analysis is based on following tests: DP, furans, paper retailed ten- This article has been organised as follows. Section two describes
sion and thermal ageing. the following topics: solid insulation degradation process,  HS
In Ref. [9], a computational platform for data integration and an estimation, evaluation of water content in paper and loss of life
intelligent system for fault detection and diagnosis as well reliabil- computation. Section three describes the proposed methodology.
ity for PT is presented. This tool uses as input: top-oil-temperature, Section four presents the obtained results after applying the pro-
FRA and dissolved gas analysis. posed method by using acquired data from four units currently in
In Ref. [10], the authors presents an approach for calculating operation. Finally, conclusions are given in section five.
the thermal lifetime of transformer insulation using Montecarlo
techniques in order to consider uncertain in load and ambient 2. Theoretical framework
temperature artificial history. Thermal ageing is the most relevant
agent in insulation degradation. Paper is used in oil-cooled PTs as insulating material due to
In Ref. [11], an updated method to calculate remaining life of its excellent dielectric and mechanical properties. Meanwhile, oil
power transformer solid insulation was presented. The research degradation can be managed by treatment methods as dehumidifi-
was conducted during a period of three years, over three single- cation, purification and filtration, or even by oil replacement; there
phase 4 kV A–2 kV/230 V transformers that ran at high load and are not paper refurbishment methods.
reached the end of insulation life. In order to compare esti- Therefore, paper is one of the most critical insulating materials
mated and real degradation, samples of oil and paper were taken in a PT [14]. A well-known indicator to measure the condition of
along the experiment, and DP and water content were mea- paper is the aforementioned DP, which value decreases when paper
sured, along with the gas content of the oil. Main finding of is exposed to water, oxygen, heat and acids. It is widely accepted
such research are: (1) the activation energy value, Ea = 111 kJ/mol, that DP value is in the range of 1200–1000 at the beginning of the
proposed in Ref. [12] is probably almost right; (2) proposed transformer life. In a DP range between 1000–500, the mechanical
equations for modelling the relationship between oxygen and strength of paper remains almost constant. Mechanical strength of
paper moisture with the factor A, which was assumed to be paper decreases proportionally to DP value in the range 500–200.
independent of the temperature, are in a good agreement with Finally, when DP value is lower than 200–150, paper is not able to
the experimental evidence; and (3) it is improper to use a withstand mechanical stresses; in fact, the mechanical strength of
numerical average of temperature profiles since the relation- paper can be reduced to 20% of its initial strength, which is usu-
ship between the temperature and the reaction rate is not ally assumed to be the end-of-life criterion for power transformer
linear. It must be noted that the accelerated ageing experiment insulation [15].
performed considers only hydrolysis, thus passing over low tem-
perature degradation phenomena. In this work, paper moisture 2.1. Ageing of cellulose
influence is emphasised, but no practical evaluation method is pro-
posed. Cellulose ageing processes act simultaneously and synergisti-
In Ref. [13], models for estimating residual operative time of cally; consequently, to achieve a good model is a very complex
transformer oil are presented, these models are based on artificial task. Nevertheless, for practical purposes most researchers and
neural networks and non-linear models and works with statistical organizations assume three independent degradation processes:
information. Application of these models predicts oil characteristics oxidation, hydrolysis and pyrolysis, each one acting in a specific
and can help effective PT management. temperature range.
From the above brief review, it is concluded that an advanced In Ref. [16], the general Arrhenius relation expressed in Eq. (1)
model for estimation of the insulation paper degradation of a PT in was disaggregated in order to consider hydrolysis, oxidation and
operation, must suitably consider dynamics in both the paper tem- pyrolysis, as:
perature,  HS , and the chemical environment, A. As was stated, in
1 1 t
Eq. (1), DP depends on  HS , which can be directly measured or math- − = k (t) · t (2)
DP (t) DP (t0 ) t0
ematically approached. For instance,  HS in a PT can be measured
by using fibre-optical thermocouples installed at the windings or with:
estimated using e.g., a set of differential equations. Similarly, the  −Ea,oxi −Ea,hyd
k (t) = Aoxi (t) · e R·(273+HS (t)) + Ahyd (t) · e R·(273+HS (t))
A factor depends on the presence of water, oxygen and acids, that
can be also directly measured or computationally estimated.
−Ea,pyr 
+Apyr (t) · e R·(273+HS (t)) (3)
In practice, aged PT does not have pre-installed thermocouples,
and direct measures of water content in paper are not usually per- where k(t) is the degradation rate and oxi, hyd and pyr subscripts
correspond to oxidation, hydrolysis and pyrolysis respectively.
 R.D. Medina et al. / Electric Power Systems Research 151 (2017) 1–11 3

2.1.1. Oxidation of cellulose sources of water contamination are: (i) air intake from atmo-
Oxidation is the combination of paper and oil with oxygen. This sphere, (ii) oil-paper decomposition and, (iii) residual humidity
reaction increases the acidity of the medium. It is predominant not removed during factory drying procedure. Water content in
at temperatures below 60–75 [◦ C]. This means that the insulat- paper can be measured from a paper sample by using Karl Fis-
ing paper of an unloaded PT is mainly degraded by acidification. cher titration, or inferred by using frequency-domain spectroscopy
In Refs. [17] and [18], it was found that the activation energy, Ea,oxi , [23]. However, in practice, for most PTs currently in operation, nei-
is lower for oxidation than for hydrolysis. Oxidation sub-products ther direct measurements nor dielectric response analysis (which
are CO, CO2 and H2 O. According to Ref. [19], from the oxidation is in some way a recent test technique) are usually performed. Nev-
point of view, thermally upgraded papers (TUP) do not offer great ertheless, physical–chemical oil testing analysis is a common and
advantages regarding non-thermal upgraded papers (No-TUP). frequent practice [24]. Therefore, historical data about water con-
tent in oil is available and it can be used to deal with the above
2.1.2. Hydrolysis of cellulose drawback.
This process requires heat, water and an acid environment to In this regard, mathematical relations that have been estab-
trigger the paper degradation. Hydrolysis is the main degradation lished in order to approach water content in paper, can be used.
process in the range 70–130 [◦ C]; it produces H+ ions, CO, CO2 , and For instance, ABB paper humidity formula, presented in Eq. (6), is
furans in No-TUP. TUP ages 1.5–3 times slower than No-TUP [19]. an empirical function that relates water content in paper to water
content in oil and  HS , assuming equilibrium conditions [25].
2.1.3. Pyrolysis of cellulose
Pyrolytic degradation is produced exclusively by heat. Its acti-
0.09733
vation energy, Ea,pyr , is 1.4–2 times bigger than hydrolysis Ea,hyd . Hp = 2.06915 · e(−0.0297·HS ) · (Ho )0.4089·HS (6)
Pyrolysis is the main degradation process above 130 [◦ C] and, start-
ing at 140 [◦ C] it becomes a self-accelerated reaction through water
and oxygen generation [20]. where Hp is the paper moisture, expressed in [%], and Ho is the oil
humidity expressed in [ppm].
2.2. Degradation accelerators The time to reach thermal equilibrium is around several hours
to a few days [26]. This condition is not reached in normal opera-
2.2.1. Temperature tion PT, because both load and ambient temperature vary during
Temperature inside a PT is not uniform. Hot spot temperature the day. In order to overcome this situation, it is possible to con-
( HS ), expressed in [◦ C], is located at the top of the tank, over the sider that PTs work under daily and weekly cyclic load profile,
windings.  HS is a function of load, ambient temperature, and con- this produces a cyclic  HS profile, then, paper moisture adsorp-
structive features of the unit. In Ref. [21], a dynamic model for  HS tion/desorption oscillates around an equilibrium point or area. This
estimation was proposed. The model is based on the solution of the idea has been verified in Ref. [27] for the range of operative temper-
following two differential equations: atures of the unit. Therefore, a practical way to determinate water in
     transformer paper insulation is to use ABB formulation with water
1 + R · KL 2 d oil content in oil and a smoothed  HS profile. Some smoothing meth-
n n
· pu · TO,R = pu · TO,R ·
(1 + R) dt ods are proposed in the state of the art: in Ref. [28] a daily average
is recommended, in Ref. [29] a long term average is proposed, in
 n + 1 Ref. [30] a three day average is computed, and in Ref. [31] the
TO − amb
+ n (4) daily moisture concentration of cellulose is calculated by averag-
TO,R ing the maximum and minimum moisture during one day interval.
Long term averages for temperature may provide an estimate of
 
 2
  n
 n dHS the bulk cellulose water content. However, its use will not detect
K · PW,pu HS · pu · HS,R = pu · W,R · fluctuations in the surface of cellulose nor thin structures mois-
dt
ture, in Ref. [28] the authors recommends use of dynamic models
 m+1 for determining Hp based on daily measurements. Considering pre-
HS − TO
+ m (5) vious evidence, in this work a daily average was selected due the
HS,R suppression of hourly variation.
where  TO is the top oil temperature, R is the ratio between load and A sophisticated method have been developed to estimate paper
no-load losses, KL is the load factor, pu is the oil viscosity in pu, n is a water content for no-equilibrium conditions [28], nevertheless this
constant that depends on oil circulation, m is an empirical constant model requires additional information about physical–chemical
taken from IEEE Loading Guide-Annex G [4].  TO,R is the rated top- characteristics of the paper and oil. When this information is avail-
oil  TO rise above the ambient temperature,  HS,R is the rated hot- able, it is advisable to use that model. Otherwise, Eq. (6) can be
spot temperature above the top-oil temperature,  TO,R is the rated used. A method to estimate oil moisture is proposed in Section 3.2.
thermal time constant,  W,R is the rated winding time constant and
 amb is the ambient temperature. Further explanations about the
constants n and m related to oil viscosity variation can be found in
Refs. [21] and [22]. 2.2.3. Chemical environment
Solving simultaneously Eqs. (4) and (5) for each interval i,  HS,i In Eqs. (1) and (2), the pre-exponential factor A represents
is obtained and can be used to calculate the de-polymerization by the environmental influence on the degradation rate k(t). Many
using Eqs. (2) and (3). accelerated ageing experiments have been performed in order to
quantify the relationship between humidity, heat and paper con-
2.2.2. Humidity dition.
Water accelerates the insulating paper ageing [11]. PT are dried For instance, in Refs. [32] and [33], the following quadratic equa-
during the manufacturing process. The moisture content of the tions that relates Ahyd value, paper moisture and oxygen were
insulation system will continually increase afterward. The main presented: for No-TUP (7), and (8) for TUP. Oxygen ranges are:
4 R.D. Medina et al. / Electric Power Systems Research 151 (2017) 1–11

Table 1 between consumed life and normal insulation life. The value of nor-
Ea ,oxi and Aoxi .
mal life is 150,000 h at standard conditions, i.e., at rated hot spot
No-TUP TUP temperature and constant humidity, it corresponds to the time to
Author Feng Lelekakis Lelekakis reach a DP value equal to 200.
Ea,oxi [kJ/mol] 89 89 82 FAA
Aoxi 4.6 × 105 4.6 × 105 3.2 × 104 %LifeLoss = × 100 (11)
Normal insulation life

Table 2

n
15000 15000
FAA = e HSR +273 − HSi +273 · ti (12)
Epyr and Apyr .
i=1
Author Conesa Capart Kashiwagi
Ea,pyr [kJ/mol] 215.7 255 220 where, FAA is the ageing acceleration factor,  HSi is the hot spot
Apyr 4 × 1017 5.8 × 1020 7.2 × 1020 temperature for each interval i,  HSR is the rated hot spot tempera-
ture: 98 ◦ C for no thermal upgraded papers and 110 ◦ C for thermal
upgraded paper,  HSi is the hot spot temperature for ever ti time
low content (<7000 ppm), medium (7000–14,000 ppm) and high
interval.
content (>16,500 ppm).
⎧ 1.78 · 108 · H 2 + 1.1 · 108 · H + 5.28 · 107 → Low O2
⎨ p p
3. Problem description and proposed methodology
Ahyd = 8 2 8 8
2.07 · 10 · Hp + 5.61 · 10 · Hp + 2.31 · 10 → Med O2 (7)
⎩ 3.1. Problem description
2.29 · 108 · Hp 2 + 9.78 · 108 · Hp + 3.86 · 108 → High O2
⎧ 6.92 · 107 · H 2 + 2.61 · 108 · H + 1.03 · 107 → Low O2
⎨ p p As it was mentioned in the previous sections, in practice, direct
Ahyd = 8 2 8 9
2.64 · 10 · Hp + 7.32 · 10 · Hp + 2.37 · 10 → Med O2 (8) measurements of  HS and water content in paper are not available,
⎩ especially in the case of old transformers. As the historical evolution
4.29 · 108 · Hp 2 + 2.03 · 109 · Hp + 4.27 · 109 → High O2 of these two parameters is needed to suitably estimate the solid
where Hp is the paper moisture, expressed in [%]. insulation degradation of a PT by means of the Arrhenius relation;
Considering that, it is desirable to maintain low oxygen levels then, the aim of this paper is to present a robust methodology for the
for units in service [16]. If oxygen content is high, there are some indirect estimation of insulating paper loss-of-life. The proposed
possible actions as: add oxygen inhibitors into oil or to conduct a method is able to account for the moisture influence and the low
degasification process. For most common service conditions, it is and high temperature degradation processes in the thermal ageing
recommendable to use low oxygen equation. model.
The value of Ahyd computed through the above expressions is
considered to be used directly in Eq. (1). However, as previously 3.2. Oil moisture estimation method
mentioned, according with empirical evidence [18], at low and high
temperatures oxidation and pyrolysis are respectively the most sig- As it was noted before, to assess the DP value, the historical
nificant degradation processes. Therefore, if the aim is to model evolution of the water content in paper is needed. In this section,
together all three degradation processes in solid insulation loss of a method to generate this paper moisture profile is proposed. The
life, as in the case of the approach proposed in this paper, then, Eqs. method uses historical information of the unit, or even, in absence
(2) and (3) must be used. of any measure, data is extracted from power transformer surveys.
Values for Ea,ox and Aox were obtained by Feng et al. [7] and If the unit owner has historical data of water content in oil, it
Lelekakis et al. [32]. These values are reported in Table 1. For pyrol- is possible to analyse tendencies and forecast humidity evolution
ysis, Ea,pyr and Apyr values can be found in the works of Conesa et al. in paper. If such information is not available, and assuming that
[34], Capart et al. [35] and Kashiwagi and Nambu [36] and, these the unit has a good sealing, it is possible to estimate the amount
are presented in Table 2. of water contamination inside the tank using empirical values to
create probable profiles.
2.3. Estimation of paper loss of life Oil humidity variations inside power transformers satisfy the
following conditions: (i) stochasticity, (ii) continuity, (iii) temporal
1) DP based remaining life independence, (iv) self-similarity and, (v) is a memoryless pro-
cess. Therefore, oil humidity content can be approached by using
In Ref. [37], the following relationship between DP and loss of the generalised Wiener Process, also called Arithmetic Brownian
life for No-TUP units was proposed. Motion (ABM) [39]. The corresponding mathematical formulation
is:
%LifeLoss = [log10 (DP) − 2.903]⁄−0.006021 (9) 
(m) (m)
Zt = Zt−1 +  · t +  · t · Xn (13)
Furthermore, the following equation for TUP was introduced in
Ref. [38]. where Zt (m) is the tth term of the m path,  is the expected value,
 DP   is the standard deviation, t is time period T divided in N equal
%LifeLoss = 100 · −0.881 · ln ⁄622 (10)
intervals and Xn is Gaussian white noise N(0,1). For all paths, there
Both equations are similar, since considers DP = 200 as the end- is only one initial value, that is, there exists exactly one Z0 for all m.
of-life criterion. The main difference is the degradation beginning If there are two or more known points, it is possible to create
value, since it was assumed that the paper begins to get degraded a Brownian Bridge (BB) assuming that the uncertainty is zero in
at DP = 820 in Eq. (9), and such initial value was DP = 622 in Eq. (10). these points or nodes, for all paths, the maximum uncertainty is
in the middle of two know points. The increments in BB are not
2) Time based remaining life independent. The corresponding equation is Eq. (14), where BZt (m)
is the tth term of the m path between the two known points Wn
The percentage of loss-of-life proposed in IEEE Loading-Guide and Wn+1 with (tn < t < tn+1 ),  is the standard deviation, and Xn is
C.57.91 [4] is shown in Eq. (11), and it represents the relation the Gaussian white noise N(0,1). There are common points for all
 R.D. Medina et al. / Electric Power Systems Research 151 (2017) 1–11 5

Table 3
Characteristics of studied units.

T1 T2 T3 T4

Rated power [MVA] 30/30/20 30/30/20 15 30/30/20

Voltage [kV] 132/34,5/13.8 132/34,5/13.8 132/34,5 132/34,5/13.8
Connections Yy0/Yd11/Yd11 Yy0/Yd11/Yd11 YNy0 Yy0/Yd11/Yd11
Cooling class ONAN/ONAF 0–70%/70%–100% 0–70%/70%–100% 0–70%/70%–100% 0–70%/70%–100%
Year of manufacture 1969 1994 1963 1984

Table 4
Reported oil moisture content, in [ppm].

Date T1 T2 T3 T4

January 2008 16.7 4.8 16.2 8.5

July 2008 17 5 16 8.6
July 2009 18 7 17 9
May 2011 18.5 6 18.7 9.7
January 2013 19.2 8 20 11.8
February 2014 19.9 7.6 20.6 12.6
January 2015 20.5 7.4 21 13.2
May 2016 21 7 22 13

Step 5. Assessment of depolymerization using Eq. (2) where

DP0 = 1000 corresponds to a new unit; for old units is possible to
use a DP0 value derived from other sources e.g. furan analysis or
direct sample measure.
Fig. 1. Proposed loss-of-life methodology.
Then, a set of final degree of polymerization values are obtained,
the probable DP final value is obtained by calculating the average
value of final DP value of all random simulations.
paths, that is, there exists exactly one Wn . for all m.
  Step 6. Estimation of loss-of-life using Eq. (9) for No TUP and Eq.
(m) t − tn (10) for TUP. These values can be used to define a distribution of
ZBt = Wn + (Wn+1 − Wn )
tn+1 − tn loss-of-life percentage value.

(t − tn ) · (tn+1 − t)
+ · · Xn (14)
tn+1 − tn 4. Case study

ABM will be used when limited or no oil moisture information is 4.1. Case study description
available, in the other hand, BB considers uncertainties when some
data exist. The loss-of-life of four step down PT (termed T1–T4) is assessed,
technical characteristics are enlisted in Table 3; the operator began
3.3. New degradation estimation approach to store load and ambient temperature data on 20/01/2008, this
study comprises the period from that date until 01/09/2016. There-
Fig. 1 shows a methodology to assess the degradation of a power fore, twenty-four values of both load and ambient temperature for
transformer considering thermal stress and paper moisture; every each of the 3147 days listed in the database are available. Ambient
step is numbered and described in detail in the following: temperature, load profile and the computed hot spot temperature
Step 1. Generation of random oil humidity profiles for the anal- for unit T1 are shown in Fig. 2.
ysis period, this information is needed in order to model moisture Oil humidity measurements were performed simultaneously for
dynamics and water migration from oil to paper. Oil humidity all units; the water content is presented on Table 4. Fig. 3 shows
depends on sealing quality of the unit and environmental condi- 1000 profiles generated with the BB tool from January 2008 to May
tions. 2016. Considering the lack of information, the ABM is used from
There are two forecast alternatives: (a) if there are some water May to September 2016, because there is no water in oil content
in oil historical records, then use Eq. (14); otherwise, (b) use Eq. information but initial value, humidity evolution starts in the last
(13) with statistical information. known value; with  = 0.00709 and  = 0.0068 computed from the
Step 2. Calculation of  HS solving simultaneously Eqs. (4) and average previous growing rates, this values are consistent with Ref.
(5); the transformer characteristics, load and ambient temperature [40]; the red dotted line shows the mean value of all paths, this is a
profiles are required as inputs. diffusive process for which uncertainty grows over time. Note that
Step 3. Estimation of hot-spot paper moisture using Eq. (6) all paths in Fig. 3 passes through the known points.
with smoothed  HS profile, and oil humidity profiles as inputs. It is To compare results, three alternatives to consider the influence
important to consider that estimated paper moisture corresponds of the chemical environment are proposed:
to the most degraded point of the windings, where the temperature Alternative 1: proposed method (Aoxi + Ahyd + Apyr ), consists on
is the highest. assess paper degradation considering all degradation process and
Step 4. The degradation rate k(t) is computed using Eq. (3). paper moisture dynamics.
Aoxi and Eoxi values are taken from Table 1. Considering that Apyr Alternative 2: simplified proposed method (Ahyd ), consist on
and Epyr values were experimentally obtained, the average value to assess paper ageing considering both, thermal degradation and
describes well pyrolysis degradation, mean values are taken from moisture variation but ignoring oxidation and pyrolysis. Ahyd val-
Table 2, Ahyd value is estimated with Eq. (7) for No-TUP and Eq. (8) ues are taken from Eq. (7) or (8) depending on the paper class, final
for TUP. Ea,hyd is 111 [kJ/mol]. DP value is computed using Eq. (1) with Ahyd instead of A.
6 R.D. Medina et al. / Electric Power Systems Research 151 (2017) 1–11

Fig. 2. Load, ambient temperature and  HS profiles for the period January 2008–September 2010.

Alternative 3: (Afix ) this alternative consist in to evaluate paper Three previous presented alternatives uses the same  HS and Hp
ageing by means of thermal degradation overlooking the influence profiles computed with method proposed in Section 3.2.
of moisture variation; to this effect, pre-exponential factor, A, is Once the final DP is computed, loss-of-life calculations were per-
assumed as a fixed value. The A value was proposed in Ref. [5] and formed by using Eq. (9) or (10), another degradation value was
corresponds to: obtained from the IEEE C.57.91 approach [4].
Table 5 presents an overview of the main features of these three
alternatives and IEEE C.57.91 proposal. It is important to remark
 that only T2 is insulated with thermally upgraded paper, the other
3.62 · 108 → No TUP
Afix = (15) units are insulated with regular paper.
1.07 · 107 → TUP
 R.D. Medina et al. / Electric Power Systems Research 151 (2017) 1–11 7

Fig. 3. Thousand random oil humidity profiles generated using BB and ABM.

Table 5 For all units, the DP profiles assessed with Alternative 3, Afix ,
Main features of the compared loss-of-life methods.
presents the higher DP value than the other alternatives; this
Alternative Paper moisture k(t) reflects that the omission of the influence of chemical environ-
Alternative 1 Dynamic moisture Aoxi Ahyd Apyr ment inside the tank of PT trigger lower degradation rates. To ignore
Proposed approach the influence of moisture lead the thermal degradation as the only
Alternative 2 Dynamic moisture Ahyd ageing process in paper.
Simplified DP profiles assessed with Alternative 1, the proposed method,
proposed approach
presents the lower value than the ones obtained with Alternative
Alternative 3 Not considered Afix
IEEE C57.91 Not considered Not considered 2, simplified proposed method, although both alternatives consid-
approach [4] ers moisture dynamic the simplified method ignores oxidative and
pyrolytic degradation. Units operates under variable load cycles
and in several periods, when units are uncharged oxidation is the
There are no direct samples of paper taken from any unit then, main degradation process.
initial DP value must be inferred from another source. Consider- Table 6 summarizes the average final DP value and the varia-
ing that, furan content analysis is a good alternative, the DP0 was tion in percentage of loss-of-life, at the beginning of the analysis
obtained from furan test applied over all units in January 2008. Pro- period units have a value of percentage of loss of life, then in order
posed initial DP are: 684, 828, 625 and 688 for T1–T4 respectively. to compare the variation in degradation the difference between
It is noted that, when part of the data of load and ambient initial and final loss of life is shown. The analysis was performed
temperature profiles for the complete operational period of the over 3147 days and it represents 75,528 h (around 50% of normal
transformer is missing, as it is just this case, available methods life—150,000 h). For all units alternative 3 presents the greater loss
to estimate the unrecorded historical load data can be used [41]. of life percentage.
However, such an estimation is out of the scope of this paper.

5. Conclusions
4.2. Results and discussion
This paper introduces a novel methodology to assess degra-
For each unit, the DP values were computed according the previ- dation of power transformer solid insulation considering thermal
ously outlined alternatives to consider the chemical environment. stress and moisture dynamics. A comparative study was conducted
To this effect the same hot-spot-temperature profile was used, and on four transformers for an 8.7 years period. The obtained results
in order to consider oil moisture uncertainty 1000 oil humidity pro- shows that the proposed method yields lower DP and loss-of-life
files were proposed, Fig. 4 presents the 1000 computed Hp profiles values than alternatives that ignores oxidation and pyrolysis pro-
for T1. Figs. 5–8 show assessed DP profiles for units T1–T4 respec- cess.
tively. There are a small dispersion in final DP values; this is because It is wide accepted that state of the art methods:
the only uncertain source is oil moisture.
It is remarkable that the  HS profile follows an annual cycle,
with peaks every summer (from December to March), this situation • do not consider paper moisture variations properly,
produces an acceleration in the depolymerisation process. • do not consider pyrolysis or oxidation,
8 R.D. Medina et al. / Electric Power Systems Research 151 (2017) 1–11

Fig. 4. Computed one thousand paper moisture profiles for T1.

Fig. 5. DP profiles for unit T1 considering three proposed alternatives.

Table 6
Final DP and loss-of-life results.

Average final DP % loss-of-life

T1 T2 T3 T4 T1 T2 T3 T4

Alternative 1 405 562 407 399 49.08 27.93 48.73 50.16

Alternative 2 516 752 541 538 31.62 0 28.20 28.60
Alternative 3 650 790 595 650 14.96 0 21.34 14.96
IEEE C57.91 – – – – 2.65 1.99 2.54 2.68
 R.D. Medina et al. / Electric Power Systems Research 151 (2017) 1–11 9

Fig. 6. DP profiles for unit T2 considering three proposed alternatives.

Fig. 7. DP profiles for unit T3 considering three proposed alternatives.

IEEE C57.91 considers thermal ageing ignoring paper moisture dences lead more accurate degradation assessing e.g. when the
influence. transformer operates unloaded, the degradation is mainly pro-
The proposed method introduces paper moisture dynamic and duced by oxidation instead of hydrolysis and on the other hand,
considers all paper-oil degradation processes; inclusion of new evi- paper moisture increases due to low temperatures.
10 R.D. Medina et al. / Electric Power Systems Research 151 (2017) 1–11

Fig. 8. DP profiles for unit T4 considering three proposed alternatives.

The main advantages of the proposed methodology are sum- [5] A.M. Emsley, G.C. Stevens, Review of chemical indicators of degradation of
marised as follows: cellulosic electrical paper insulation in oil-filled transformers, IEEE Proc. Sci.
Meas. Technol. 141 (1994) 324–334.
[6] N. Lelekakis, G. Wenyu, D. Martin, J. Wijaya, D. Susa, A field study of aging in
• Paper moisture can be roughly estimated considering operative paper-oil insulation systems, IEEE Electr. Insul. Mag. 28 (2012) 12–19.
[7] D.Y. Feng, Z.D. Wang, P. Jarman, Modeling thermal life expectancy of the UK
conditions of the unit and test information.
transmission power transformers, ICHVE IEEE (2012) 540–543.
• This work is supported by well-known methods and uses all evi- [8] A.E.B. Abu-Elanien, M.M.A. Salama, Asset management techniques for
dences. transformers, Electr. Power Syst. Res. 80 (2010) 456–464.
• This method can be widely used for NO-TUP and TUP. [9] J.L. Velasquez-Contreras, M.A. Sanz-Bobi, S. Galceran Arellano, General asset
management model in the context of an electric utility: application to power
transformers, Electr. Power Syst. Res. 81 (2011) 2015–2037.
Although the results presented have proven the effectiveness [10] A.E.B. Abu-Elanien, M.M.A. Salama, A Monte Carlo approach for calculating
the thermal lifetime of transformer insulation, Int. J. Electr. Power Energy
of this method, future work will be focused on: (i) analysis of the Syst. 43 (2012) 481–487.
influence of different initial moisture values, (ii) generation of the [11] D. Martin, Y. Cui, C. Ekanayake, H. Ma, T. Saha, An updated model to
complete historical profiles of load and ambient temperature from determine the life remaining of transformer insulation, IEEE Trans. Power
Deliv. 30 (2015) 395–402.
commissioning date to date using existent information to assess
[12] L.E. Lundgaard, W. Hansen, D. Linhjell, T.J. Painter, Aging of oil-impregnated
the present end-of-life value and, (iii) forecasting future load and paper in power transformers, IEEE Trans. Power Deliv. 19 (2004) 230–239.
ambient profiles in order to forecast end-of-life probable date. [13] M.A.A. Wahab, M.M. Hamada, A. Mohamed, Artificial neural network and
non-linear models for prediction of transformer oil residual operating time,
Electr. Power Syst. Res. 81 (2011) 219–227.
Acknowledgements [14] A.M. Emsley, The kinetics and mechanisms of degradation of cellulosic
insulation in power transformers, Polym. Degrad. Stab. 44 (1994) 343–349.
[15] T.K. Saha, Review of modern diagnostic techniques for assessing insulation
This work was supported in part by DAAD, Grupo de Inves- condition in aged transformers, IEEE Trans. Dielectr. Electr. Insul. 10 (2003)
tigación en Energía – Universidad Politécnica Salesiana – Sede 903–917.
Cuenca and Agencia de Promoción Científica y Tecnológica (PICT [16] Brochure Cigré, 323: Ageing of cellulose in mineral oil insulated transformers,
Study Committee D 1 (2007).
2012 - 0388). [17] L.E. Lundgaard, W. Hansen, S. Ingebrigtsen, Ageing of mineral oil impregnated
cellulose by acid catalysis, IEEE Trans. Dielectr. Electr. Insul. 15 (2008)
540–546.
References [18] K.B. Liland, M. Kes, M.H.G. Ese, L.E. Lundgaard, B.E. Christensen, Study of
oxidation and hydrolysis of oil impregnated paper insulation for transformers
[1] W. Bartley, An analysis of international transformer failures, part 1, Hartford using a microcalorimeter, IEEE Trans. Dielectr. Electr. Insul. 18 (2011)
Steam Boiler Inspection and Insurance Company, 2004. 2059–2068.
[2] W. Flores, E. Mombello, G. Ratta, J.A. Jardini, Life of power transformers [19] Brochure Cigre, 393, Thermal performance of transformers. Working Group
immerserd in oil. -State-of-the-art-part I. Correlation between life and A224 (2009).
temperature, IEEE Lat. Am. Trans. 5 (2007) 50–54. [20] CIGRE, A227, Life Management Techniques for power transformers, CIGRE
[3] A. Abu-Siada, S. Islam, A new approach to identify power transformer Work Group A2.18, 2003.
criticality and asset management decision based on dissolved gas-in-oil [21] D. Susa, Dynamic Thermal Modelling of Power Transformers, Department of
analysis, IEEE Trans. Dielectr. Electr. Insul. 19 (2012) 1007–1012. Electrical and Communications Engineering, Helsinki University of
[4] IEEE, IEEE Guide for Loading Mineral-Oil-Immersed Transformers and Technology, 2005.
Step-Voltage Regulators, IEEE Std C57.91-2011 (Revision of IEEE Std
C57.91-1995), 2012, pp. 1–123.
 R.D. Medina et al. / Electric Power Systems Research 151 (2017) 1–11 11

[22] R.D. Medina, A.A. Romero, E.E. Mombello, G. Ratta, Comparative study of two [32] N. Lelekakis, D. Martin, J. Wijaya, Ageing rate of paper insulation used in
thermal aging estimating methods for power transformers, IEEE Lat. Am. power transformers part 1: oil/paper system with low oxygen concentration,
Trans. 13 (2015) 3287–3293. IEEE Trans. Dielectr. Electr. Insul. 19 (2012) 1999–2008.
[23] T.K. Saha, P. Purkait, Understanding the impacts of moisture and thermal [33] N. Lelekakis, D. Martin, J. Wijaya, Ageing rate of paper insulation used in
ageing on transformer’s insulation by dielectric response and molecular power transformers part 2: oil/paper system with medium and high oxygen
weight measurements, IEEE Trans. Dielectr. Electr. Insul. 15 (2008) 568–582. concentration, IEEE Trans. Dielectr. Electr. Insul. 19 (2012) 2009–2018.
[24] J.P. Lata, R.D. Medina, W.P. Borja, D.X. Morales, Transformer common test [34] J.A. Conesa, J. Caballero, A. Marcilla, R. Font, Analysis of different kinetic
description for condition management, 2016 IEEE International Conference models in the dynamic pyrolysis of cellulose, Thermochim. Acta 254 (1995)
on Automatica, ICA-ACCA 2016 (2016). 175–192.
[25] Pahlavanpour, M. Martins, Eklund, Study of moisture equilibrium in oil-paper [35] R. Capart, L. Khezami, A.K. Burnham, Assessment of various kinetic models for
system with temperature variation, Proceedings of the 7th International the pyrolysis of a microgranular cellulose, Thermochim. Acta 417 (2004)
Conference on Properties and Applications of Dielectric Materials, 2003 1123 79–89.
(2003) 1124–1129. [36] T. Kashiwagi, H. Nambu, Global kinetic constants for thermal oxidative
[26] Y. Du, M. Zahn, B.C. Lesieutre, A.V. Mamishev, S.R. Lindgren, Moisture degradation of a cellulosic paper, Combust. Flame 88 (1992) 345–368.
equilibrium in transformer paper-oil systems, IEEE Electr. Insul. Mag. 15 [37] R.D. Stebbins, D.S. Myers, A.B. Shkolnik, Furanic compounds in dielectric liquid
(1999) 11–20. samples: review and update of diagnostic interpretation and estimation of
[27] W. Wei, W. Xin, M. Zhiqing, L. Chengrong, L. Shaoyu, W. Yanwei, H. Zhen, M. insulation ageing, Proceedings of the 7th International Conference on
Jixian, L. Kaihua, Disequilibrium of moisture in oil-pressboard insulation Properties and Applications of Dielectric Materials, 2003 923 (2003) 921–926.
under transient temperature condition, 2012 International Conference on [38] P.K. Sen, P. Sarunpong, Overloading and loss-of-life assessment guidelines of
Condition Monitoring and Diagnosis (CMD) (2012) 894–896. oil-cooled transformers, Rural Electric Power Conference, 2001 (2001), pp.
[28] D. Martin, C. Perkasa, N. Lelekakis, Measuring paper water content of B4/1–B4/8.
transformers: a new approach using cellulose isotherms in nonequilibrium [39] I. Karatzas, S. Shreve, Brownian Motion and Stochastic Calculus, Springer
conditions, IEEE Transactions on Power Delivery (2013) 1. Science & Business Media, 2012.
[29] M. Koch, S. Tenbohlen, T. Stirl, Diagnostic application of moisture equilibrium [40] P. Griffin, V. Sokolv, B. Vanin, Moisture equilibrium and moisture migration
for power transformers, IEEE Trans. Power Deliv. 25 (2010) 2574–2581. within transformer insulation systems, CIGRE Working Group A2 30 3 (2008).
[30] D. Martin, T. Saha, M. Gibson, G. Buckley, S. Chinnarajan, R. Dee, G. Russell, T. [41] A.A. Romero-Quete, E.E. Mombello, G. Rattá, Assessing the loss-of-insulation
Gray, K. Wyper, R. Li, K. Williams, Field studies on comparing techniques to life of power transformers by estimating their historical loads and ambient
measure the water content of power transformer insulation, 2015 IEEE PES temperature profiles using ANNs and Monte Carlo simulations, Dyna 83
Asia-Pacific Power and Energy Engineering Conference (APPEEC) (2015) 1–5. (2016) 104–113.
[31] Y. Cui, H. Ma, T. Saha, C. Ekanayake, D. Martin, Moisture-dependent thermal
modelling of power transformer, IEEE Trans. Power Deliv. 31 (2016)
2140–2150.

View publication stats