DGA Method-Based ANFIS Expert System for Diagnosing Faults and Assessing Quality of Power Transformer Insulation Oil

W2216242478 · 수집 2026-10-04 18:12:22

T10511 High voltage insulation and dielectric phenomena

CoT 라벨링

라벨링 전

PDF 파일

자동 다운로드한 PDF PDF 보기

기본 정보

DOI 제목 발행년도 저널 URL 페이지 열기 PDF URL PDF 열기 발행기관 피인용수 라이선스

추가 정보 초록

&lt;p class="zhengwen"&gt;&lt;span lang="EN-GB"&gt;Accurate fault diagnostics and assessment of electrical power transformer insulation oil for lifelong endurance are the key issues addressed in this research. The durability of a transformer is significantly determined by the quality of its insulation oil, which deteriorates over time due to temperature fluctuations and moisture content. Protecting transformers from potential failure through early and precise diagnosis of faults and through efficient assessment of oil quality during the actual conduct of the operation can avoid sizeable economic losses. The ANFIS Expert System that uses intelligent software plays an important role in this regard. The dissolved gas analysis (DGA) in oil is a reliable method for diagnosing faults and assessing insulation oil quality in transformers. The safeguarding teams of transformer power stations often suffer from the occurrence of sudden faults, which result in severe damages and heavy monetary loss. The oil in transformers must be appropriately treated to circumvent such failures. In this research, an ANFIS Expert System was used to diagnose faults and to assess the status and quality of insulation oil in power transformers. A suitable treatment was identified using the Rogers ratio method depending on the DGA in oil. The graphical user interface from the MATLAB environment was used and proven effective for fault diagnosis and oil quality evaluation. The training algorithm is capable of assessing oil quality according to the specifications of the IEEE standard C57-104 and the IEC standard 60599.&lt;/span&gt;&lt;/p&gt;

본문

Modern Applied Science; Vol. 10, No. 1; 2016

ISSN 1913-1844 E-ISSN 1913-1852
Published by Canadian Center of Science and Education

# DGA Method-Based ANFIS Expert System for Diagnosing Faults and Assessing Quality of Power Transformer Insulation Oil

Mohd Muhridza Bin Yaacobˡ, Ahmed Raisan Hussein 1 & Mohd Fauzi Bin Othman 2

1 Institute of High Voltage and High Current, Faculty of Electrical Engineering, Universiti Teknologi Malaysia,
Malaysia

2 Centre for Artificial Intelligence and Robotic, Faculty of Electrical Engineering, Universiti Teknologi Malaysia,
Malaysia

Correspondence: Ahmed Raisan Hussein, Institute of High Voltage and High Current, Faculty of Electrical
Engineering, Universiti Teknologi Malaysia, 81310, UTM Skudai, Johor, Malaysia. Tel: 60-107-553-5695. Fax:
6107-557-8150. E-mail: ahmad_rissan@yahoo.com

Received: April 28, 2015     Accepted: August 28, 2015    Online Published: December 21, 2015

doi:10.5539/mas.v10n1p13    URL: http://dx.doi.org/10.5539/mas.v10n1p13

Abstract

Accurate fault diagnostics and assessment of electrical power transformer insulation oil for lifelong endurance
are the key issues addressed in this research. The durability of a transformer is significantly determined by the
quality of its insulation oil, which deteriorates over time due to temperature fluctuations and moisture content.
Protecting transformers from potential failure through early and precise diagnosis of faults and through efficient
assessment of oil quality during the actual conduct of the operation can avoid sizeable economic losses. The
ANFIS Expert System that uses intelligent software plays an important role in this regard. The dissolved gas
analysis (DGA) in oil is a reliable method for diagnosing faults and assessing insulation oil quality in
transformers. The safeguarding teams of transformer power stations often suffer from the occurrence of sudden
faults, which result in severe damages and heavy monetary loss. The oil in transformers must be appropriately
treated to circumvent such failures. In this research, an ANFIS Expert System was used to diagnose faults and to
assess the status and quality of insulation oil in power transformers. A suitable treatment was identified using the
Rogers ratio method depending on the DGA in oil. The graphical user interface from the MATLAB environment
was used and proven effective for fault diagnosis and oil quality evaluation. The training algorithm is capable of
assessing oil quality according to the specifications of the IEEE standard C57-104 and the IEC standard 60599.

Keywords: expert system, IEC method, insulation oil, fault diagnosis, power transformer, Rogers method

1. Introduction

Electricity has become an important part of human life. At present, it is difficult to imagine life without
electricity. Electricity has become a basic need for people, agriculture, and the industry; hence, it has a strong
impact on the economy of any country. A suitable method for accurate fault diagnostics and assessment of the oil
quality of electrical power transformer is needed to ensure long-term protection. The longevity of a transformer
is determined by its insulating quality. Generally, insulation deteriorates over time due to fluctuations in
temperature, moisture, oxygen, and other environmental factors. Diagnosis of faults and assessment of oil quality
are important steps in protecting transformers from operational failures. Power transformers are exceptionally
expensive and damage to the insulation system often causes high economic loss(Sun et al., 2010) . In the past,
several methods were adopted for the diagnosis of faults and for assessing oil quality in transformers; different
standards with approved specifications were developed, including the IEEE standard C57-104 and the IEC
standard 60599 (Committee et al., 2009; Rogers, 1978). Most of the assessment and diagnosis methods were
based on dissolved gas analysis (DGA). Despite many efforts, efficient and precise methods for determining the
nature of faults and the subsequent rectification mechanisms that can attain superior performance.

Electrical transformer oils are known to possess a dual function: insulation and cooling. The superior insulation
and cooling capacity of electrical transformer oils that enables the transformer to withstand severely elevated
changes in temperature over an extended period of operation protects transformers from faults. However, high
electrical pressure, temperature, and harsh environmental conditions produce hydrocarbons (gases) in
transformer oils, which affect its function negatively. Therefore, preventive maintenance schedules for fault

www.ccsenet.org/mas Modern Applied Science Vol. 10, No. 1; 2016

diagnosis and quality assessment of transformer oils are obligatory(Shirkhodaie et al., 1998).

Several methods have been developed for fault diagnosis. Methods for assessing oil quality, on the other hand, is
another topic. The various types of incipient faults that occur in transformers are arcing, corona discharge, low
energy sparking, severe overloading, and overheating. These are some of the mechanisms that can result in
chemical decomposition of the insulating material, as well as in the formation of various combustible and
non-combustible gases, such as methane (CH4), ethane (C2H6), ethylene (C2H4), acetylene (C2H2), and hydrogen
(H2). Decomposition generates gaseous products that can be dissolved in mineral oil(Silva et al., 2000). DGA, a
popular conventional method for assessing the condition of an oil-filled transformer, involves analysis of the
gases dissolved in the insulating medium. This method is widely used by electric companies and has received
worldwide recognition because of its reliability in the detection of incipient faults in power transformers.
However, DGA method provides results based on empirical evidence; hence, the approach does not offer a
mathematical description. In response, several soft computing techniques based on DGA have been proposed to
enhance the diagnosis capabilities and effectiveness of the existing DGA method(Satyanarayana et al., 2008).

2. Dissolved Gas Analysisd

The DGA in oil involves taking a sample of transformer oil to determine the concentration and ratio of gases in
the sample. These gases are formed by the following factors: high temperatures, partial discharge, arcing, and
corona. These factors lead to the decomposition of the oil hydrocarbons (compounds of carbon and hydrogen).
These gases are classified into two types: combustible gas and non-combustible gas. Combustible gases include
carbon monoxide (CO), hydrogen (H2), methane (CH4), acetylene (C2H2), ethylene (C2H4), and ethane (C2H6).
Noncombustible gases include, oxygen (O2), nitrogen (N2), and carbon dioxide (CO2). When these gases are
present in the oil sample, these exist at different rates and are generated with different rates. Some of the gases
vary in ratio from time to time, depending on the changes in temperature and the age of the transformers(Anon
n.d, 2000) .

_2.1 Rogers Ratio Method_

In this method, four ratios are accounted for: CH4/H2, C2H2/CH4, C2H4/C2H6, and C2H2/C2H4. Diagnosis of faults
is accomplished via a simple coding scheme based on ranges presented in Table 1. Ratios are coded to identify
12 types of faults in transformers to identify the types of faults as shown Table 2(Muhamad et al., 2007).

Table 1. Rogers ratio code

< 0.1 5

CH4/H2 0.1- 1 0

1--3 1

> 3 2
C2H6/CH4 < 1 0

CH4/H2

C2H6/CH4 < 1 0

> 1 1

< 1 0

C2H4/C2H6 1-3 1

< 1 0

C2H4/C2H6 1-3 1

> 3 2

< 0.5 0

C2H2/C2H4 0.5-3 1

< 0.5 0

C2H2/C2H4 0.5-3 1

> 3 2

Table 2. Fault types according to rogers ratio method

No. CH 4 /H 2 C 2 H 6 /CH 4 C 2 H 4 /C 2 H 6 C 2 H 2 /C 2 H 4 Fault type
1 0 0 0 0 No fault
2 1-2 0 0 0 <150 ºC thermal fault
3 1-2 1 0 0 150–200 ºC thermal fault
4 0 1 0 0 200–300 ºC thermal fault
5 0 0 1 0 General conductor overheating
6 1 0 1 0 Winding circulating currents
7 1 0 2 0 Core and tank circulating currents overheated joints
8 5 0 0 0 Partial discharge
9 5 0 0 1-2 Partial discharge with tracking

www.ccsenet.org/mas Modern Applied Science Vol. 10, No. 1; 2016

10 0 0 0 1 Flashover without power follow-through
11 0 0 1-2 1-2 Arc with power follow- through
12 0 0 2 2 Continuous sparking to floating potential

_2.2 IEC Ratio Method_

In the IEC ratio method, the objective is to create three ratios (C2H2/C2H4, CH4/H2, and C2H4/C2H6), to allow
comparison between the result of this method and that of the Rogers method. In this method, the ratio, C2H6/CH4,
was excluded because it indicates a limited temperature range of decomposition. The coding of the three gas
ratios are presented in Table 3. The nine fault types are shown in Table 4(Sarma & Kalyani, 2004).

Table 3. Iec code ratio

IEC

Ratio of characteristic gases

C ~~2~~ H ~~2~~
C2H4

CH ~~4~~

C ~~2~~ H ~~4~~

C2H6

< 0.1 0 1 0

0.1-1 1 0 0

1-3 1 2 1

> 3 2 2 2

Table 4. Fault types according to the iec ratio method

Case No. C ~~2~~ H ~~2~~

C2H4

CH ~~4~~

H2

C ~~2~~ H ~~4~~

C2H6

Characteristic fault

0 0 0 0 No fault
1 0 1 0 Partial discharges of low energy density
But not
significant

2 1 1 0 Partial discharges of low energy density

3 1-2 0 1-2 Discharges of low energy
4 1 0 2 Discharges of High Energy
5 0 0 1 Thermal fault of low Temperature <150 ºC

6 0 2 0 Thermal fault of low Temperature
<150 ºC–300 ºC

7 0 2 1 Thermal fault of medium temperature range
300 ºC–700 ºC

8 0 2 2 Thermal fault of high temperature >700 ºC

3. ANFIS Expert System

Expert systems fall under the artificial intelligence category of computer applications. Designing an expert
system needs a knowledge engineer - an individual who studies how human experts make decisions and
translates the findings into rules that a computer can understand. ANFIS, derived from the term adaptive
network-based fuzzy inference engine, was first proposed. This system was designed to allow if-then rules and
membership functions to be constructed based on the historical data of established metrics. The system was also
designed with an adaptive nature for automatic tuning purposes (Foong et al., 2009).

www.ccsenet.org/mas Modern Applied Science Vol. 10, No. 1; 2016

_3.1 Design ANFIS Expert System_

The expert system was designed according to the flowchart shown fig 1. The design provides an interface where
the user can input gas values derived from the analysis of samples of transformer oil through dissolved gas ratios.
Subsequently, the user can choose the method to be used in the analysis process. The system then uses ANFIS to
process the data.

Figure 1. Flowchart for ANFIS Expert System Design

_3.2 Design of the ANFIS Expert System Interface_

The user interface of the ANFIS expert system was designed using GUI access provided in the MATLAB
environment. As depicted in Figure 2, the user input interface allows the following items to be entered: gas
values, analysis methods (Rogers ratio method or IEC ratio method), gas ratios, fault type diagnosis, and
assessment of the quality of insulating oil (Gao et al., 1998).

www.ccsenet.org/mas Modern Applied Science Vol. 10, No. 1; 2016

Figure 2. Interface of the ANFIS Expert System

_3.3 ANFIS Models_

Fuzzy inference is the process of formulating the mapping from a given input to an output using fuzzy logic. The
mapping is then used as a basis for making decisions. The process of fuzzy inference involves membership
functions, logical operations, and if-then rules. Fuzzy Inference System (FIS) 5 maps the input characteristics to
input membership functions in rules to output membership function to single-value output. The shape of
membership functions depends on the parameters. ANFIS chooses the parameters of membership functions
automatically during the learning phase to match the system's output with the target output. Hence, ANFIS can
be used for modeling a system where input/output data are available for modeling but no predetermined model
for the system exists. Parameters of membership functions could be defined to fit the membership functions of
the input/output data, as depicted in Figure 3.

Figure 3. ANFIS structure

ANFIS was tested using 180 training samples. In ANFIS, the diagnosis of each crisp value of gas ratio is
represented by a trapezoidal membership function (Figure 4). The set of fuzzy rules relates the input variables to
the output. The fuzzy inference engine in the ANFIS network uses if-then rule-based system, given by IF

www.ccsenet.org/mas Modern Applied Science Vol. 10, No. 1; 2016

antecedent and Then Consequent, a fuzzy rule set of 12 rules for Rogers Ratio Method based ANFIS and a fuzzy
rule set of 9 rules for IEC Ratio Method based ANFIS.

Figure 4. Input membership functions for ANFIS

For the assessment of the quality of transformer oil, the input variable TCG uses a different set of terms (Low,
Moderate, High, V. High, and Severe). The input membership functions and the different set of terms are shown
in Figure 5. Only one output variable is assigned to transformer oil quality. This is called oil degradation index
(ODI). ODI value is computed for each input.The input membership functions are designed to give a constant
output over a substantial range. The membership of each input variable varies over a definite range, uniformly on
both sides (trapezoidal). This is an extremely useful property for monitoring normal changes due to operation
and ageing. Another significant feature of the ODI value is its correlation with the specific reconditioning
procedures for transformer oil. The output membership functions are shown in Figure 6. MF 1= Good Oil, MF 2
= Single Filtering and Degassing, MF 3 = Double Filtering and Degassing, MF 4 = Reclamation and MF 5 =
Don’t Use the Oil.

Figure 5. Input membership function used in oil reconditioning

www.ccsenet.org/mas Modern Applied Science Vol. 10, No. 1; 2016

Figure 6. Output membership function

The ANFIS responses obtained from the tested set are shown in Figure 7 and 8. The FIS output clearly shows
that the proposed method can be effectively used for diagnosing incipient faults of power transformers.

Figure 7. Testing output for Rogers Ratio Method-Based ANFIS

Figure 8. Testing the output of IEC Ratio Method-Based ANFIS

4. Results and Discussion

Sampling of power transformer oils was performed at the sites where these transformers are located for more
than 15 years. The 30 samples used for the detection of the DGA in oil and the results of the analysis, are
presented in Table 5. Insertion of the data ratios of gas-derived samples from Table 5 to the expert system

www.ccsenet.org/mas Modern Applied Science Vol. 10, No. 1; 2016

interface was tested in each case. The results, as summarized in Table 6, are good and in agreement with the
specifications of approved standards for diagnosing faults and assessing quality of insulating oils. The results
obtained through the ANFIS Expert System suggest that identifying the status of transformer oil while in service
is an important and necessary maintenance step to ensure long-term functioning of the transformer oil. The
results of the test, as presented in Table 6, are satisfactory. Results observed prior to identifying faults are
accurate statements of the type of fault and the assessment of the quality of the oil, which are guides in providing
treatment. In some cases, faults that could not be diagnosed by the Rogers ratio method were diagnosed
successfully using the IEC ratio method, and vice versa.

Table 5. Value of gases in ppm data input to ANFIS expert system

NO. CH4 H2 C2H6 C2H4 C2H2
1 47 191 43 15 0.0001
2 200 50 300 200 50
3 245 36 144 332 0.0001
4 187 86 136 363 0.0001
5 24 10 372 24 0.0001
6 7 17 13 3 0
7 42 206 16 82 221
8 999 769 234 1599 31
9 50 1000 40 30 10
10 107 127 11 154 224
11 9739 2004 2750 5114 0
12 144 60 67 449 9
13 3 4 2 1 0
14 53 65 41 95 14
15 79 33 30 215 5
16 370 1249 56 606 1371
17 1053 507 297 1440 17
18 77 57 19 21 0
19 10 160 3 1 1
20 770 199 217 1508 72
21 695 416 74 867 0
22 27 10 49 4 1
23 940 1450 211 322 61
24 1.7 2.35 0.6 1.2 0.1
25 96 240 20 8 5
26 4000 960 1290 1560 6
27 73 980 58 12 0
28 1500 280 1200 150 40
29 8517 2844 10196 4422 39
30 4 8 350 24 2

Table 6. Results of diagnosing faults and assessng the quality of insulating oil

Fault Type Oil
NO.
Rogers Ratio IEC Ratio Treatment
1 Normal deterioration No Fault Single filtering &
degassing
2 Overheating – 150–200 °C Unidentifiable Single filtering &
degassing
3 Unidentifiable Thermal fault of medium Double filtering &
Temperature 300–700 °C degassing

3 Unidentifiable Thermal fault of medium Double filtering &
Temperature 300–700 °C degassing

4 Winding Circulating Currents Thermal fault of medium Double filtering &
Temperature 300–700 °C degassing

5 Overheating – 150–200 °C Thermal fault of low temperature – Single filtering &

www.ccsenet.org/mas Modern Applied Science Vol. 10, No. 1; 2016

150–300 °C degassing
6 Overheating – 200–300 °C No Fault Good Oil
7 Arc with Power Follow-Through Discharges of High Energy Double filtering &
degassing
8 Core and Tank Circulating Currents Thermal fault of High Temperature – Reclamation
above 700 °C

9 Partial discharge Partial Discharges of High Energy Double filtering &
Density degassing

10 Arc with Power Follow-Through Discharges of High Energy Double filtering &
degassing
11 Unidentifiable Thermal fault of medium Do Not use
Temperature 300–700 °C

9 Partial discharge Partial Discharges of High Energy
Density

Do Not use

12 Core and Tank Circulating Currents Thermal fault of High Temperature –
above 700 °C

Double filtering &
degassing

13 Flashover without Power FollowThrough

Unidentifiable Good Oil

14 General Conductor Overheating Discharges of Low Energy Single filtering &
degassing
15 Core and Tank Circulating Currents Thermal fault of High Temperature – Single filtering &
above 700 °C degassing

15 Core and Tank Circulating Currents Thermal fault of High Temperature – Single filtering &
above 700 °C degassing

16 Arc with Power Follow-Through Discharges of High Energy Reclamation
17 Core and Tank Circulating Currents Thermal fault of High Temperature – Reclamation
above 700 °C

Reclamation

18 Winding Circulating Currents Thermal fault of medium
Temperature 300–700 °C

18 Winding Circulating Currents Thermal fault of medium Single filtering &
Temperature 300–700 °C degassing

19 Partial Discharge with Tracking Partial Discharges of High Energy Single filtering &
Density degassing

20 Unidentifiable Thermal fault of High Temperature – Reclamation
above 700 °C

Reclamation

21 Core and Tank Circulating Currents,
overheated joints

21 Core and Tank Circulating Currents, Thermal fault of High Temperature – Double filtering &
overheated joints above 700 °C degassing

22 Overheating – 150–200 °C Unidentifiable Good Oil
23 General Conductor Overheating Discharges of Low Energy Reclamation
24 General Conductor Overheating Thermal fault of low temperature – Good Oil
below 150 °C

Thermal fault of High Temperature –
above 700 °C

Good Oil

25 Flashover without Power Follow
-Through

25 Flashover without Power Follow Unidentifiable Single filtering &
-Through degassing

26 Unidentifiable Thermal fault of medium Do Not use
Temperature 300–700 °C

Do Not use

27 Partial discharge Partial Discharges of Low Energy Double filtering &
Density degassing

28 Slight Overheating - below 150 °C Unidentifiable Reclamation
29 Overheating – 150–200 °C Thermal fault of low temperature – Do Not use
150–300 °C

30 Unidentifiable No Fault Single filtering
&degassing

5. Conclusion

We introduced a simple but accurate ANFIS Expert System to identify the type of fault and to assess the quality
of insulating oil in power transformers. The DGA in oil, being the reliable method, was used for fault diagnosis
and insulation oil quality evaluation. The Rogers ratio and the IEC ratio methods using DGA in oil were used
where the software employed ANFIS through the MATLAB program. The expert system was found reliable for
diagnosing faults and assessing the quality of insulating oil. The results are satisfactory based on their adherence
to the specifications of the IEEE standard C57-104 and the IEC standard 60599. Thus, the software results can

www.ccsenet.org/mas Modern Applied Science Vol. 10, No. 1; 2016

be used by maintenance teams to rectify the occurrence of sudden faults that result in power disruption and
economic loss. The nature of faults was accurately and efficiently detected and the oil replacement was
determined as the output.

References

Anon. The J & P Transformer Book J & P Books.

Committee, T., Power, I., & Society, E. (2009). IEEE Std C57.104’-2008 (Revision of IEEE Std C57.104-1991).

IEEE Guide for the Interpretation of Gases Generated in Oil-Immersed Transformers.

Foong, K. C., Chee, C. T., & Wei, L. S. (2009). _Adaptive network fuzzy inference system (ANFIS) handoff_

_algorithm_ . _International Conference on Future Computer and Communication, ICFCC, 1_ (2), 195–198.

Gao, N., Zhang, G. J., Qian, Z., Yan, Z., & Zhu, D. H. (1998). _Diagnosis of DGA based on fuzzy and ANN_

_methods_ . Proceedings of 1998 International Symposium on Electrical Insulating Materials. 1998 Asian
International Conference on Dielectrics and Electrical Insulation. 30th Symposium on Electrical Insulating
Ma.

Muhamad, N. A. et al. (2007). Comparative Study and Analysis of DGA Methods for Transformer Mineral Oil.

_2007_ _IEEE_ _Lausanne_ _Power_ _Tech_, 45–50. Retrieved from
http://ieeexplore.ieee.org/lpdocs/epic03/wrapper.htm?arnumber=4538290

Nunes, D. S. I., Imamura, M. M., & Souza, A. N. D. (2000). _The application of neural networks to the analysis_

_of dissolved gases in insulating oil used in transformers_ . Smc 2000 conference proceedings. 2000 IEEE
International Conference on systems, man and cybernetics. Cybernetics evolving to systems, humans,
organizations, and their complex interactions, 4, 2643–2648.

Rogers, R. R. (1978). IEEE and IEC Codes to Interpret Incipient Faults in Transformers, Using Gas in Oil

Analysis. _IEEE Transactions on Electrical Insulation, 13_ (5), 349-354.

Sarma, D. V. S. S. S., & Kalyani, G. N. S. (2004). _Ann approach for condition monitoring of power transformers_

_using_ _DGA_ . IEEE Region 10 Conference TENCON 2004, 444–447. Retrieved from
http://ieeexplore.ieee.org/lpdocs/epic03/wrapper.htm?arnumber=1414803

Satyanarayana, K. V. et al. (2008). Application of Artificial Intelligence for the Assessment of the Status of

Power Transformers, 104–107.

Shirkhodaie, A., Muthuswamy, S. K., & Saravanan, S. S. (1998). A Hybrid Neuro-Fuzzy Based Fault Diagnostic.

_Technique for Bearings Condition-Based Maintenance, 724._

Sun, X., Sun, X., Liang, Y., Lv, D., Liu, D., & Bian, J. (2010). _Application of Kernel C-means clustering and_

_Dempster-Shafer theory of evidence in power transformers fault diagnosis_ . Proceedings - 2010 7th
International Conference on Fuzzy Systems and Knowledge Discovery, 3, 1060–1063.

Copyrights

This is an open-access article distributed under the terms and conditions of the Creative Commons Attribution
license (http://creativecommons.org/licenses/by/3.0/).

본문 출처 PDF: PDF 보기 · 다른 PDF는 위 [PDF 업로드]로 올릴 수 있습니다.

활용가능성

이력 0건

아직 편집 이력이 없습니다. 저장할 때마다 이전 내용이 여기에 백업됩니다.