Review and Implementation of Transformer Health

Index Methods in line with the Development of a

Condition Assessment Tool

Stefan Kittan*, Stefan Kornhuber Peter Kästel, Gunnar Nitsche, Gerd Valtin,
Faculty of Electrical Engineering and Computer Science Michael Weise
University of Applied Sciences Zittau/Görlitz Faculty of Electrical Engineering and Information
Germany Technology
Leipzig University of Applied Sciences
Germany

Abstract— Transformers are important equipment of the of an important transmission grid transformer can endanger the
electrical power supply system. A transformer defect can lead to security of energy supplies. For this reasons a transformer
heavy damage on the equipment itself as well as to its diagnostic tool is under development, to help experts taking
surroundings. Therefore, a defect is coherent with high costs. care of whole transformer fleets.
Numerous transformers are in operation since several decades.
Furthermore, in Germany an additional stress for the equipment The field of transformer diagnostic is widely explored.
yields from a rising energy input of regenerative sources. The Existing diagnostic and condition assessment solutions focus
situation of the distribution grid is additionally intensified by a some special methods, e.g. the dissolved gas analyses. Thus,
delayed system expansion. Hence, a transformer condition not all of the available information is used by those solutions.
assessment tool is in development, which will support Therefore, the development of the new diagnostic tool aims at
transformer expert at their daily work. This tool provides merging all relevant techniques for transformer diagnostic. To
automated measurement evaluation, generation of diagnostic achieve this goal, the diagnostic tool is designed modular. By
statements as well as the determination of a transformer overall this design the tool is flexible and can be easily enhanced e.g.
condition. by new transformer types or new diagnostic methods. Two
main components of the tool are the diagnostics module as well
Health index (HI) methods are approved techniques for as the condition assessment module. When a transformer
determining the general condition of a transformer. Thus, they
shows an abnormality, the diagnostic module is applied. This
are important for a condition assessment tool also. An overview
module consists of a measurement interpretation and a graph
about actual HI methods was generated by a literature study.
Afterward the most promising methods where selected and based diagnostic algorithm. All available measurements are
implemented for the condition assessment tool. This paper analyzed by the measurement interpretation. Each
recapitulates the literature study as well as the selection criteria measurement is rated as defective, inconspicuous or good and
and the implementation of the most interesting HI methods. transferred to the diagnostic algorithm subsequently. For the
diagnostic part a comprehensive database of transformer
Keywords—transformer; condition assessment; health index components with possible defects and failures as well as
suitable measurement methods was developed. Since
I. INTRODUCTION1 transformer failures and their causes can be very complex, this
database is under constant development. Therefore, the
Due to the high significance and the value of transmission database was implemented as graph, achieving a good
and power transformers, a reliable state assessment is of interpretability and expandability. The diagnostic algorithm
importance. In addition the transmission grid in Germany is evaluates this database by means of the given measurement
under high pressure, due to slow expansion, increasing amount interpretations. As result, a ranked list of diagnostic statements
of regenerative energy sources and with it a partially is calculated and the diagnosis process is narrowed down. In
displacement of supply and consumption. Therefore, a failure which each statement shows a component and the defect.
__________________________________________________
1
When a transformer is in normal operation and all
The project "Condition Assessment of Power and
Distribution Transformers in the Electrical Power
diagnostic measurements are inconspicuous, an assessment of
Engineering (Smart Grid - Smart Monitoring)" is the overall condition is of importance though. This enables
supported by funds of Federal Ministry of planning maintenance measures and determining the
Education and Research in line with the program operational risk. An approved method for transformer overall
research at universities of applied science condition assessment is the health index (HI). Thus, the
(SC03FH026PB5). condition assessment module implements recent HI methods.
Fig. 1 depicts the use cases for the different modules of the

978-1-5386-4424-9/18/$31.00 ©2018 IEEE

diagnostic tool. In this paper, the literature study for the HI an effective age of a transformer by using a published curve.
methods is summarized. Furthermore, the selection and The effective age describes the transformer service time
implementation of the most interesting methods are described. regarding the mean load and the corresponding degradation. By
defining an end-of-life failure probability, the effective
transformer age at end-of-life can be determined. Hence,
yes
Diagnost ics
Det ect ion
of defect (s)
GABRIEL ET AL. can estimate the remaining lifetime of a
transformer [6]. The authors are testing this approach with four
Input Dat a
- laborat ory t est s Not iceable
no
yes
transformers. An important result of those tests is that the HI
problem s
- on- / off-line
m easurem ent s Ex ceeding of loses significance if individual measurements are near or
- visual inspect ion t hreshold values Short t erm
m aint enance
exceeding allowed limits. Due to the weighting of HI, those
- expert evaluat ion
- ... cases are not apparent very well. Therefore, further
Condit ion Long t erm investigation of the concerned components is advisable [6].
no Assessm ent m aint enance
ZHOU ET AL. present an approach for a calculation of the
Fig. 1. Differentiation between the use cases of the diagnostics and condition weighting factors. Where the diagnostic factors have to be time
assessment module series. The HI weights are determine by the signal entropy [7].
Furthermore, ASHKEZARI ET AL. are introducing a
technique for investigating the influence of diagnostic factors
II. LITERATURE STUDY ON HEALTH INDEX METHODS on HI as well as the correlation between those factors by
multivariate analysis [8]. This method is explained and tested
A. Overview on a set of oil test data for 170 transformers. As diagnostic
For the development of the condition assessment module, a factors the gas in oil concentrations, furfural concentration,
literature study about actual HI methods was conducted. Based breakdown voltage, acidity, moisture and resistivity where
on 21 studied papers, 13 different HI methods were selected. For each transformer of the dataset a HI rating was
investigated. The contents are subdividing into classical HI given under consideration of industry standards, utility
methods, artificial intelligence (AI) HI methods and further practices and expert judgements. With the help of a common
investigations. Classical HI methods means the HI calculation factor analysis, five influence factors on the HI could be
by a weighted sum of different diagnostic factors. Examples for identified. Additionally, the correlation between those factors
those factors are gas in oil concentrations, chemical and and the relation to the diagnostic factors was determined. The
physical oil analysis, results of visual inspections, furan common factors are grouping highly correlated diagnostic
concentrations or electrical measurements. The differences factors. For instance, the analysis shows a high correlation for
between classical HI methods are: moisture, furfural concentration and breakdown voltage. Where
the breakdown voltage is correlated negatively. Which can be
• Selection of diagnostic factors interpreted as with rising moisture the degradation of the
• Evaluation of each diagnostic factor insulation system advances. This leads to a rising amount of
furfural and a lowered breakdown voltage [8]. As conclusion,
• Summing up of diagnostic factors into sub-indices the multivariate analysis is an appropriate tool to facilitate the
transformer condition assessment by revealing correlations
• Weighting of diagnostic factors or sub-indices for final between large amounts of input data and their influence on the
HI calculation resulting HI. For the applicability of this method, a significant
AI based HI methods means the usage of data based test database is necessary, however. If this requirement is
classification or regression methods like artificial neuronal fulfilled, the multivariate analysis can be used to:
networks [1], [2], adaptive neuro-fuzzy inference systems [3],
• Determine appropriate weighting factors for a specific
support vector machines [4] or similar [5]. For a given set of
transformer fleet and a selected set of diagnostic factors
diagnostic factors, a database of about hundred up to several
hundred transformers is prepared. A HI specifies the condition • Examine weighting factors of a classic HI method and
of each transformer of such a database. The HI rating can be the suitability for a specific transformer fleet
carried out by utility experts, service provider or classical HI
methods. The database is divided into training and test data. • Differ diagnostic factors with low and high influence on
The training data is used for the learning process of the AI HI
method. In this process, the AI methods are recognizing the SCATIGGIO ET AL. are combining the HI with the
relation between the chosen diagnostic factors and the HI of the probability of a dangerous event as well as with the average
transformers. After the successful learning step, the AI model damage, which is caused by a transformer failure [9]. The HI is
is validated by means of a comparison between the calculated a measure for the probability of a transformer failure. By
HI of the AI method with the given HI value of the test data. multiplication of the three parameters, a yearly risk of a
Further investigations of HI methods are studies about transformer is calculated. This risk is a foundation of an aging
enhancement of HI, novel approaches or the application of orientated maintenance strategy. Further information about the
known methods. For instance, GABRIEL ET AL. are corresponding HI calculation is also presented by SCATIGGIO
combining a HI approach with failure probability estimations ET AL. [10].
[6]. With the known failure probability, the authors conclude to
 III. DETAILS ON INVESTIGATED HI METHODS Selection of
Transformer fleet
diagnostic factors

A. Classical health index methods

During the research the most important facts of each Measurements
Data preprocessing
Condition assesment
by experts
investigated HI methods where noticed for the purpose of
comparison. The first of those parameters is the considered Selection of
transformer components. According to CIGRE relevant Database
regression method

components are active part and its isolation, bushings, tap

changer, cooling system, oil containment and preservation (oil Test data Training data Parameterisation
system) [11]. Already the failure of one components means a
risk for the whole transformer operation. The more extensive a
HI method is the merrier is the transformer condition rating. Validation Trained model Training
An important feature would be the possibility to distinguish the
component state, e.g. by sub-indexes. Since a defective
component can be understated due to the weighting of the HI
calculation. The number of diagnostic factors used is a good Quality no
sufficient
measure for the extent of an HI, also. Where the availability of
those parameters is also of relevance. Because the HI
yes
calculation can only be carried out, if all necessary parameters
were determined. Therefore, from the view of the diagnostic Databased
tool it is constructive to consider HI methods with less HI model

parameters also. In order that a condition assessment is even

possible with less data available. As last, the confirmability of Fig. 2. General procedure for developing an AI based HI model
the HI method descriptions is of relevance. Because only those
HI calculations can be implemented to the software. use cases is corresponding to an extrapolation of the trained
model. However, the AI methods are meant for interpolation.
Tab. 1 shows a comprehensive comparison of all To make things worse the necessary data sets for implementing
investigated classical HI methods. The HI rating quality of published AI methods are needed, but rarely available. Due to
HERNANDA ET AL. deviated due to challenging this limitations AI based HI methods are not considered for the
environment condition while taking the oil samples [12]. EN- diagnostic tool.
WEN AND BIN are also considering the mean transformer
load for the HI calculation [13]. The HI methods of JAHROMI
ET AL. [14], [15] and GORGAN ET AL. [16] respectively are IV. SELECTION AND IMPLEMENTATION
based on each other. Since these methods are treated as one HI, An important requirement for the HI calculation is that all
which was continuously developed further. All methods are necessary diagnostic factors are determined. Therefore, HI
using besides the classical measurements further parameters regarding a lot of different factors implying great effort. Since
(load history, age, visual inspections). By this, the authors are such an effort cannot be carried out for each transformer, the
able to consider the whole transformer in the HI calculation. diagnostic tool needs HI methods with less input values as
well. For this category, the methods of HERNANDA ET AL.
B. Artificial intelligence based health index methods and EN-WEN AND BIN were chosen. Both methods are using
All AI based methods have in common, that they need a good available diagnostic factors. In addition, the whole
database to learn the wanted correlations. The general procedure of the HI calculation differs between both methods.
procedure to achieve a well-trained model is depicted in Figure Furthermore, EN-WEN AND BIN are also considering the
2. Where this procedure is nearly the same for each AI transformer load and age in their calculation. This enables the
technique. condition assessment by different points of view.

Since the AI based HI methods are data based models, the The HI methods of JAHROMI ET AL. and GORGAN ET
applicability is limited to the corresponding transformer fleet, AL. respectively are using a huge set of diagnostic factors. As
environmental and load conditions. The propagation to other all three methods are based on each other, they are treated as
different development steps. Those methods were implemented

TABLE I. COMPARISON OF CLASSICAL HI METHODS

Hernanda En-Wen
Jahromi et al. [14], [15]; Malik et Haema and Phadungthin
et al. [17], and Bin
Gorgan et al. [16] al. [18] [19], [20]
[12] [12], [13]
active part, bushings, tap active part, bushings, tap
Transformer Components active part active part changer, cooling system, oil active part changer, cooling system, oil
system system
Number of diagnostic factors 8 8 20 / 24 / 27 8 21
Availability of diagnostic factors Very good Very good Good fair Good
Confirmability Yes Yes Yes No Yes
in such a way, that the software determines the appropriate [9] F. Scatiggio, L. Calcara und M. Pompili, „Risk prevention for HV
method according to the available input data. Additionally the transformers: Beyond the health index,“ in IEEE Electrical Insulation
Conference (EIC), 2016.
user can select the set of boundary values for the evaluation of
[10] F. Scatiggio und M. Pompili, „Health index: The TERNA's practical
each diagnostic factor. The HI method of HAEMA AND approach for transformers fleet management,“ in IEEE Electrical
PHADUNGTHIN is comparable to the methods of JAHROMI Insulation Conference (EIC), 2013
ET AL. Since some details are realized different, this method is [11] CIGRE_WG_A2.44, Guide 630: Guide on Transformer Intelligent
implemented as an alternative. Condition Monitoring (TICM) Systems, CIGRE, 2015.
[12] F. Ortiz, I. Fernandez, A. Ortiz, C. J. Renedo, F. Delgado und C.
A benefit of using multiple HI methods is that a consistent Fernandez, „Health indexes for power transformers: a case study,“ IEEE
rating over all methods rises the reliability of the result. On the Electrical Insulation Magazine, Bd. 32, pp. 7-17, 9 2016.
other hand, divergent results can give additional information [13] L. En-Wen und S. Bin, „Transformer health status evaluation model
due to the different aspects of each HI method. Hence the based on multi-feature factors,“ in Power System Technology
disadvantage of HI by neglecting single information due (POWERCON), 2014.
weighting, is countered. [14] A. N. Jahromi, S. Cress, R. Piercy, F. Wang und J. Service, „An
approach to determine the health index of power transformers,“ in IEEE
International Symposium on Electrical Insulation (ISEI), 2008.
V. SUMMARY [15] A. N. Jahromi, R. Piercy, S. Cress, J. Service und W. Fan, „An approach
A new transformer diagnostic tool is in development, with to power transformer asset management using health index,“ IEEE
Electrical Insulation Magazine, Bd. 25, pp. 20-34, 3 2009.
the objective of merging all relevant diagnostic techniques into
one tool. To achieve this goal, the developed software has to be [16] B. Gorgan, P. Notingher, L. V. Badicu und G. Tanasescu, „Calculation
of power transformers health indexes,“ Annals of the University of
flexible. Hence, the tool is build up modular. Craiova, Electrical Engineering series, p. 6, 2010.
HI methods are well suited for a general transformer [17] I. G. N. S. Hernanda, A. C. Mulyana, D. A. Asfani, I. M. Y. Negara und
condition assessment. Since those methods are used for a D. Fahmi, „Application of health index method for transformer
condition assessment,“ in TENCON 2014.
corresponding module of the diagnostic tool. By a literature
[18] H. Malik, A. Azeem und R. K. Jarial, „Application research based on
study, actual HI methods were investigated. The HI can be modern-technology for transformer health index estimation,“ in
determined by classical weighting either or by an AI based data Systems, Signals and Devices (SSD), 2012.
recognition. Where the AI based HI methods are not suitable [19] J. Haema und R. Phadungthin, „Condition assessment of the health
for a generic diagnostic tool, due to limitations in the index for power transformer,“ in Power Engineering and Automation
applicability. Since only the classical HI methods were Conference (PEAM), 2012.
considered in detail. Furthermore, all methods were compared [20] J. Haema und R. Phadungthin, „Development of condition evaluation for
and the suitable methods were implemented for the use in the power transformer maintenance,“ in Power Engineering, Energy and
Electrical Drives (POWERENG), 2013.
diagnostic software.

VI. REFERENCES
[1] M. Ahmed, M. Elkhatib, M. Salama und K. B. Shaban, „Transformer
health hndex estimation using orthogonal wavelet network,“ in Electrical
Power and Energy Conference (EPEC), 2015.
[2] A. E. B. Abu-Elanien, M. M. A. Salama und M. Ibrahim,
„Determination of transformer health condition using artificial neural
networks,“ in Innovations in Intelligent Systems and Applications
(INISTA), 2011.
[3] H. Zeinoddini-Meymand und B. Vahidi, „Health index calculation for
power transformers using technical and economical parameters,“ IET
Science, Measurement Technology, Bd. 10, pp. 823-830, 2016.
[4] K. I. Mohamadeen, R. M. Sharkawy und M. M. Salama, „Binary cat
swarm optimization versus binary particle swarm optimization for
transformer health index determination,“ in Engineering and Technology
(ICET), 2014.
[5] A. D. Ashkezari, H. Ma, T. K. Saha und C. Ekanayake, „Application of
fuzzy support vector machine for determining the health index of the
insulation system of in-service power transformers,“ IEEE Transactions
on Dielectrics and Electrical Insulation, Bd. 20, pp. 965-973, 6 2013.
[6] G. Tanasescu, O. Dragomir, L. Voinescu, B. Gorgan, P. V. Notingher, T.
Suru und C. Melinte, „Assessment of power transformers conditions
based on health index,“ in CIGRÉ Regional South-East European
Conference, At Sibiu, 2012.
[7] Y. Zhou, L. Ma, J. Yang und C. Xia, „Entropy weight health index
method of power transformer condition assessment,“ in Reliability,
Maintainability and Safety (ICRMS), 2011.
[8] A. D. Ashkezari, H. Ma, C. Ekanayake und T. K. Saha, „Multivariate
analysis for correlations among different transformer oil parameters to
determine transformer health index,“ in IEEE Power and Energy Society
General Meeting, 2012.