19th IEEE International Conference on Dielectric Liquids (ICDL), Manchester, United Kingdom, 25 – 29 June, 2017
Insulation Resistance of Power Transformers –
Method for Optimized Analysis
André P. Marques, Cláudio H. B. Azevedo,
José A. L. dos Santos, Felipe R. de C. Sousa
CELG Distribuição S.A, and Federal Institute of Education,
Science and Technology of Goiás, Goiânia, Brazil
andre.pm@celg.com.br
Abstract –This paper proposes a method for evaluating and
classifying transformers, using equations to obtain scores and
ratings for insulation resistance tests as a function of the age of
these devices, resulting in recommendations to support decision-
making. Analyses were made of test results of power
transformers of 1 MVA to 50 MVA, wrapped in insulation Kraft
paper and immersed in insulating mineral oil, with nominal
voltages of 13.8 kV to 230 kV, ranging from 1 to 51 years old,
whose ratings varied from “A” (excellent) to “E” (very poor).
The methodology was developed using computational methods of
equation optimization involving the hill-climbing algorithm
associated with the 1/5th success rule. It was concluded that the
combined use of maintenance engineering criteria and
computational optimization techniques resulted in an efficient
diagnostic method for these devices.
Keywords: electrical test, insulation resistance; optimization;
power transformers
I. INTRODUCTION
Insulation resistance testing of power transformers stands
out, among other applications [1], in non-invasive predictive
techniques of the field of preventive maintenance, which
enables real-time detection of defects and failures through
quantitative and qualitative analysis as well as efficient
diagnoses.
In this context, this article proposes a method for the
analysis and classification of power transformers in terms of
insulation resistance testing, including dielectric discharge
magnitudes and polarization and absorption indexes according
to the age of the transformer, which differentiates this work.
The criteria for maintenance engineering defined here are
based on the following sources: the experience of specialists,
bibliographical research, and statistical surveys of a database
of electrical tests carried out over a 34-year period (from 1982
to 2016) at CELG Distribuição, an electric utility company
that supplies electricity to 2.8 million consumer units.
Insulation resistance, which is directly associated with the
dielectric robustness of the insulation, can be analyzed
quantitatively by means of its test. The polarization and
absorption indices, as well as the dielectric discharge factor,
are directly related to the state of conservation of the
insulation system. These indices allow one to evaluate the
insulator qualitatively and to infer if it is in a good state of
conservation or degraded. Considered together, one thus has a
quantitative and qualitative evaluation of the insulation
system.
The existing literature, for example [2-7], provides some
information about insulation resistance testing, but there are
1
978-1-5090-4877-9/17/$31.00 ©2017 IEEE
Nicolas K. Moura, Cacilda J. Ribeiro, Yuri A. Dias,
Adonay Rodrigues, Adson S. Rocha, Leonardo da C. Brito
Federal University of Goiás, and Goiano Federal Institute,
Goiás, Goiânia, Brazil
cacildaribeiro@gmail.com
gaps in the explanation of the results of analysis (ranges of
values), which makes it difficult to classify power
transformers; moreover, their ages are not taken into account.
Therefore, this paper aims to address these gaps and
contribute to existing studies by proposing an optimized and
efficient method of analysis to aid in power transformer
diagnostics.
II. INSULATION RESISTANCE TEMPERATURE CORRECTION
FACTOR FOR THE REFERENCE TEMPERATURE (20º C)
The value of insulation resistance changes according to
temperature, the presence of moisture and dirt, and also as a
function of the degradation of the insulation material.
Thus, it is possible to deduce the state of power
transformer insulation systems and to detect tendencies and
insulation decay rates by monitoring and comparing the results
of insulation resistance tests. However, before comparing
such results, these tests must be referenced to the same
temperature.
Complementing earlier studies, this article presents two
options of efficient mathematical equations to determine the
correction factors of resistance values measured at the same
reference temperature of 20ºC, and also discusses two
approximate conversion methods, as follows:
• The temperature conversion method by the 2/decade
factor [8] establishes that, at each increase of 10ºC, the
value of insulation resistance decreases by
approximately one half [2]. This rule has a direct
practical application, but may compromise the
accuracy of results, especially in non-multiple
temperature intervals of 10;
• The table of conversion factors recommended by the
American National Standards Institute - ANSI / NETA
[3] is directly applicable but may compromise the
accuracy of the results, especially with measurements
outside multiple ranges of 5; and
• The mathematical equations presented below are easy
to implement and provide accurate results.
A. Equation developed in this work: correction factor
An insulation resistance (IR) value at 20ºC can be referred
to, according to eq. (1),
R20°C =RT ∙ KT (1)
where:
R20°C: IRat MΩ, corrected at 20°C;
 19th IEEE International Conference on Dielectric Liquids (ICDL), Manchester, United Kingdom, 25 – 29 June, 2017

RT: IRat MΩ, at the test temperature inoC; and : IR (MΩ) measured after 1 min of testing;
KT: IR correction factor for the temperature of 20ºC. : IR (MΩ) measured after 30 seconds of testing.
Using exponential interpolation tools and taking as TABLE I. CLASSIFICATION CRITERIA FOR RHL INSULATION
reference the values listed by manufacturers of measuring RESISTANCE AS A FUNCTION OF THE AGE OF THE EQUIPMENT
instruments, equation (2) was developed in this work for the age ≤ 10 years 10 <age ≤ 20 age> 20 years
correction factor KT, Rating
(MΩ) years (MΩ) (MΩ)
(2) A (Excellent) RHL≥41,743 RHL≥ 20,085 RHL≥13,233
= 0.23991 ∙ e . 0.0222
20,112≤RHL< 12,359≤RHL< 6,738≤RHL<
where: B (Good)
41,743 20,085 13,233
KT: IR correction factor for the temperature of 20ºC; and 15,135 ≤RHL< 8,674≤RHL< 4,972≤RHL<
C (Marginal)
T: Temperature of the insulation system measured during 20,112 12,359 6,738
the test, in oC. 5,369≤RHL< 3,318≤RHL< 2,051≤RHL<
D (Poor)
15,135 8,674 4,972
B. Equation of the ANSI/NETA MTS table: correction factor E (Very Poor) RHL< 5,369 RHL< 3,318 RHL< 2,051
From the data of the ANSI/NETA MTS table [2], and
TABLE II. CLASSIFICATION CRITERIA FOR RH AND RLINSULATION
using the mathematical tool, we obtained the correction factor RESISTANCES AS A FUNCTION OF THE AGE OF THE EQUIPMENT
given by (3) for application in equation (1),
age ≤ 10 years 10 <age ≤ 20 age> 20 years
Rating
(MΩ) years (MΩ) (MΩ)
.
= 0.2525 ∙ e (3) A (Excellent) R ≥ 24,100 R ≥ 11,596 R ≥ 7,640
where: B (Good)
11,612≤ R< 7,135≤ 3,890≤ R<
KT: IR correction factor for the temperature of 20ºC; and 24,100 R<11,596 7,640
T: Temperature of the insulation system measured during 8,738≤ R< 5,008≤ R< 2,870 ≤ R<
C (Marginal)
11,612 7,135 3,890
the test, in oC. 3,100≤ R< 1,916≤ R< 1,184 ≤ R<
The mathematical equations presented here both have the D (Poor)
8,738 5,008 2,870
above described advantages, and it is up to the analyst to E (Very Poor) R < 3,100 R< 1,916 R< 1,184
choose the one that best suits him.
TABLE III. CLASSIFICATION CRITERIA FOR THE
III. CLASSIFICATION CRITERIA POLARIZATION INDEX (PI) AS A FUNCTION OF THE AGE OF THE
EQUIPMENT
Based on the experience of specialists, bibliographical
5 <age ≤ 10
research, and statistical surveys in a database of 529 electrical Rating age ≤ 5 years
years
age> 10 years
tests performed on 237 devices – including power and A (Excellent) PI ≥ 3,070 PI ≥ 2,045 PI ≥ 1,850
distribution transformers and mobile substations – with 2,333 ≤ PI< 1,667 ≤ PI< 1,500 ≤ PI<
B (Good)
nominal voltages of 13.8 kV to 230 kV, power rating of 1 3,070 2,045 1,850
MVA to 50 MVA, ranging from 1 to 51 years old, carried out 1,720 ≤ PI< 1,400 ≤ PI< 1,286 ≤ PI<
C (Marginal)
2,333 1,667 1,500
over a 34-year period (from 1982 to 2016), criteria were 1,294 ≤ PI< 1,201 ≤ PI< 1,151 ≤ PI <
developed in this work for the evaluation and classification of D (Poor)
1,720 1,400 1,286
RH, RHL and RL insulation resistances, according to the age of E (Very Poor) PI< 1,294 PI< 1,201 PI < 1,151
the device. The values of these resistances were corrected to
20ºC by means of equations (1) and (2), as indicated below: TABLE IV.CLASSIFICATION CRITERIA FOR THE ABSORPTION
INDEX (AI) AS A FUNCTION OF THE AGE OF THE EQUIPMENT
RH : IR from HV (high voltage) winding to ground; 5 <age ≤ 10
Rating age ≤ 5 years Age> 10 years
RHL : IR between HV and LV (low voltage) windings; years
A (Excellent) AI≥ 1,600 AI ≥ 1,520 AI ≥ 1,440
RL : IR from LV winding to ground.
1,350 ≤ AI< 1,520 ≤ AI< 1,215 ≤ AI<
It should be noted that statistical studies of the insulation B (Good)
1,600 1,283 1,440
resistance were also carried out as a function of the nominal 1,250 ≤ AI< 1,188 ≤ AI< 1,125 ≤ AI<
voltage of the equipment. However, the differences observed C (Marginal)
1,350 1,283 1,215
in the statistical distribution were not enough to justify a D (Poor)
1,100 ≤ AI< 1,045 ≤ AI< 1,000 ≤ AI<
refinement of the criteria. Tables I and II describe the criteria 1,250 1,188 1,125
defined according to age. E (Very Poor) AI< 1,100 AI< 1,045 AI< 1,000

The polarization index (PI) [2] and the absorption index

(AI) were calculated based on the results of the insulation The Dielectric Discharge (DD) index was also calculated
resistance tests, according to equations (4) and (5), based on the test results, according to equation (6).
respectively, and their classification criteria are presented in
Tables III and IV, as a function of the age of the equipment. = (6)
∙
= / (4)
= / (5) where:
where: : current measured after 1 minute (mA);
: IR (MΩ) measured after 10 min of testing; V: test voltage (V); and

2
978-1-5090-4877-9/17/$31.00 ©2017 IEEE
 19th IEEE International Conference on Dielectric Liquids (ICDL), Manchester, United Kingdom, 25 – 29 June, 2017

C: measured capacitance (F). TABLE VI.EXPECTED AND OPTIMIZED VALUES OF THE WEIGHTS
However, it was only included recently in the database of Parameter Expected Values Optimized Values
the tests carried out by the company, after the acquisition of , 0.2800 0.2790
new instruments that enabled this complementation of the , 0.0533 0.0417
diagnosis, and it was not possible to draw up statistics based , 0.2800 0.2790
on age. Table V lists the classification criteria for DD. , 0.0533 0.0641
, 0.2800 0.2721
TABLE V.CLASSIFICATION CRITERIA FOR THE DIELECTRIC , 0.0533 0.0641
DISCHARGE (DD)
Total: 1.000 1.000
Rating Dielectric Discharge (DD)
As can be seen, the optimization results are close to the
A (Excellent) DD ≤ 1 ms
expected values, meeting the previously specified maintenance
B (Good) 1 ms <DD ≤ 2 ms
engineering criteria.
C (Marginal) 2 ms <DD ≤ 4 ms
D (Poor) 4 ms <DD ≤ 7 ms The values were double weighted, first according to the
E (VeryPoor) DD> 7 ms importance of the variable (Weight of the Variable, ) in the
evaluation of the equipment, and then as a function of the
The ranges of values developed here offer advantages over individual score attributed to it (Weight of the Score, ( )),
those reported in the literature, e.g. [7], because they consider according to equation (7).
age as a parameter, associating variations in IR to the aging of
. ∙
equipment. Furthermore, the values listed in Tables I to V are ( ) = 7.0876 ∙ e (7)
more rigorous, and therefore, more conservative than those in
the literature. Figures 1 to 3 show graphs of the individual scores – with
respective ratings (“A” to “E”) – as a function of the variable,
IV. OPTIMIZED ANALYSIS OF INSULATION RESISTANCE for each of the parameters of interest. Note that, in all the
cases, the criterion varies according to the age (in years) of the
This work was developed using the hill-climbing equipment.
optimization algorithm [9,10], which is associated with the
mechanism of parameter adaptation by the “1/5th rule,” which
resulted in the optimized analysis of insulation resistance
values, in accordance with maintenance engineering criteria.
The objective function used here served to minimize the
sum of the error modules of the comparisons between the
classification provided by the method and that provided by the
specialists. The error was calculated from the integer distance
between the expected rating and the obtained rating,
presenting a maximum error equal to 1.
The data used are diverse and come from 150 significant
tests – 70% of the cases used for training the method, and the
remaining 30% corresponding to different transformers to
validate the results, for which the diagnostic criteria were
verified –, extracted from the abovementioned main database,
and were classified by a team of experts, using final ratings
that varied from “A” (excellent) to “E” (very poor). Accuracy Fig.1. Injective functions (score versus magnitude) for RH and RL
rates of 86.7% and 88.9% of the cases were obtained in
training and validation sets, respectively.
Within the scope of this work, which focused on insulation
resistance tests of two-winding power transformers, one has
the following optimizable parameters:
, : Weight of Resistance of insulation RH;
, :Weight of the PI of insulation RH;
, :Weight of Resistance of insulation RHL;
, :Weight of the PI of RHL;
, :Weight of Resistance of insulation RL;
, :Weight of the PI of RL.

Table VI presents the weights obtained through

computational optimization and compares them with the
expected values, based on the expertise of the team of
engineers and on maintenance engineering criteria. Fig.2. Injective functions (score versus magnitude) for RHL

3
978-1-5090-4877-9/17/$31.00 ©2017 IEEE
 19th IEEE International Conference on Dielectric Liquids (ICDL), Manchester, United Kingdom, 25 – 29 June, 2017

another for LV/TV, and in the end, considering the worst case.
In addition, it should be noted that this work was based on
statistical data from transformers ranging from 13.8 kV to
230 kV, wrapped in insulation Kraft paper and immersed in
insulating mineral oil. However, this method is also expected
to present satisfactory results for power transformers with
higher voltages.
This article highlights the importance of evaluations to
obtain scores and ratings, generating an expression
characterized as a double-weighted and normalized sum,
which, applied to the evaluation and classification of
insulation resistance tests, presents good results (accuracy rate
of 88.9% of cases) and only minor variations from the values
expected by maintenance engineering criteria.
Fig.3. Injective functions (score versus magnitude) for the PI Thus, it can be concluded that the criteria presented here
contribute significantly to the analysis. The combined use of
maintenance engineering criteria and computational
Thus, the best individual scores related to variables in good
optimization techniques resulted in a practical and efficient
condition are associated with considerably lower weights than
method that is widely applicable to the power sector for the
scores attributed to variables in poor condition. Therefore,
analysis of insulation resistance tests and the diagnosis of
poor scores imply higher weights, and hence, greater impacts
power transformers.
on the evaluation of that equipment.
With the individual scores , the weights of each ACKNOWLEDGMENTS
variable and the weights as a function of the individual The authors thank the Federal University of Goiás and the
score ( ), the Final Score (SF) is calculated using (8). Federal Institute of Education, Science and Technology of
Goiás for their practical support, as well as the Goiás State
∑ ∙ ( )∙ , Research Foundation (FAPEG), the Federal Agency for the
= (8) Support and Improvement of Higher Education (CAPES), and
∑ ( )∙ ,
the Brazilian Electricity Regulatory Agency (ANEEL) for
The variable participates in all the constituent parameters their financial support of this work
of the electrical tests under analysis. REFERENCES
According to the final score, actions are recommended to
underpin decision-making for power transformers, as shown in [1] A.P. Marques et al.,“Method for the Evaluation and Classification of
Power Transformer Insulating Oil based on Physicochemical Analyses,”
Table VII. IEEE Electrical Insulation Magazine, Vol. 33, No. 1, p.40-50, 2017.
[2] IEEE Guide for Field Testing of Electric Power Apparatus –Part 1: Oil
TABLE VII.CRITERIA FOR OBTAINING FINAL RATING AND Filled Power Transformers, Regulators, and Reactors, IEEE Std.62,
RECOMMENDED ACTIONS
Mar. 2005.
Final Final [3] American National Standards Institute, ANSI/NETA MTS-2011:
Recommended Actions Standard for Maintenance Testing Specifications for Electrical Power
Rating Score
Equipment and Systems, May 2011
A [4] Associação Brasileira de Normas Técnicas, NBR 5356-1:2010:
SF> 0.8536 Continue operating the equipment normally
(Excellent) Transformadores de Potência. Parte 1: Generalidades, Rio de Janeiro,
0.6618 <SF Continue operating the equipment while July 2010. [Brazilian Association of Technical Standards, NBR 5356-1:
B
monitoring the evolution of the results in the 2010: Power Transformers. Part 1: Generalities, Rio de Janeiro, July
(Good) ≤ 0.8536
subsequent records 2010] (in Portuguese)
C 0.5659 <SF Investigate and perform other tests in the [5] A. JAHROMI et al.,“An approach to power transformer asset
(Marginal) ≤ 0.6618 short term to confirm results and tendencies management using health index,” IEEE Electrical Insulation Magazine,
0.3070 <SF Schedule the removal of the equipment from vol. 25. pp. 20-34. 2009.
D
operation for internal inspection, location [6] IEEE Guide for Failure Investigation, Documentation, and Analysis for
(Poor) ≤ 0.5659
and correction of defects Power Transformers and Shunt Reactors, IEEEC57.125, May 2005.
E Remove the equipment from operation
[7] IEEE Guide for Diagnostic Field Testing of Fluid-Filled Power
(Very SF ≤ 0.3070 immediately for internal inspection, location
Transformers, Regulators, and Reactors, IEEE C57.152, Mar. 2013.
Poor) and correction of defects
[8] MEGGER Co., “A Stitch in Time: The complete guide to electrical
insulation test,” Available at:
V. CONCLUSIONS
<https://www.instrumart.com/assets/Megger-insulationtester.pdf>.
The diagnostic criteria presented here refer to two- [9] A. P. ENGELBRECHT, “Computational Intelligence – An
winding power transformers. However, for the diagnosis of Introduction,” 2nd ed., John Wiley & Sons, 2007.
three-winding transformers (HV, LV, TV), the equipment can [10] S. RUSSELL, and P. NORVIG, “Artificial intelligence - A modern
approach,” 2nd ed., Prentice Hall. Englewood Cliffs, 2003.
be analyzed in subsets of two-winding transformers, i.e., by
making a diagnosis for HV/LV, another for HV/TV and

4
978-1-5090-4877-9/17/$31.00 ©2017 IEEE