Upload
others
View
5
Download
0
Embed Size (px)
Citation preview
i
Department of Electrical and Computer Engineering
A New Fuzzy Logic Approach to Identify Transformer Criticality
using Dissolved Gas Analysis
Sdood Abd Al-Gbar Hmood AL- Auqaili
This thesis is presented for the Degree of
Master of Philosophy
of
Curtin University
December 2013
ii
TABLE OF CONTENTS
Table of Contents………………………………………………………………………………………ii
List of Figures ....................................................................................................................................... iv
List of Tables .......................................................................................................................................... v
ABSTRACT ......................................................................................................................................... vii
Acknowledgments ................................................................................................................................. ix
List of Publications ................................................................................................................................. x
Acronyms .............................................................................................................................................. xi
1. CHAPTER 1 INTRODUCTION ................................................................................................ 5
1.1 PROJECT OVERVIEW-BACKGROUND OF FAULT DIAGNOSIS OF POWER TRANSFORMER ........ 5
1.2 CONDITION MONITORING BY DISSOLVED GAS ANALYSIS (DGA) ........................................ 7
1.2.1 Dissolved Gases Analysis Methodology .......................................................................... 7
1.3 MOTIVATION OF RESEARCH .................................................................................................. 8
1.4 OBJECTIVES .......................................................................................................................... 9
1.5 THESIS OUTLINE ................................................................................................................. 10
2 CHAPTER 2 POWER TRANSFORMER CONDITION MONITORING AND
DIAGNOSIS ........................................................................................................................................ 11
2.1 INTRODUCTION ............................................................................................................. 11
2.2 CRITICAL COMPONENTS FAILURE ............................................................................ 11
2.2.1 Core ............................................................................................................................... 11
2.2.2 Windings ........................................................................................................................ 12
2.2.3 Insulating Oil ................................................................................................................ 12
2.2.4 Insulating Paper (Cellulose) ......................................................................................... 13
2.2.5 Bushings ........................................................................................................................ 13
2.2.6 Tapchanger ................................................................................................................... 14
2.2.7 Fans ............................................................................................................................... 14
2.2.8 Pumps ............................................................................................................................ 14
2.3 SMALL GAS SAMPLES (SGS) OIL SAMPLING (PHYSICAL AND CHEMICAL TEST) ................ 15
2.4 DIAGNOSTICS ................................................................................................................. 15
2.4.1 Gas Chromatography .................................................................................................... 15
2.4.2 Interfacial Tension (IFT) ............................................................................................... 16
2.4.3 Acid Number.................................................................................................................. 17
2.4.4 Moisture ........................................................................................................................ 17
2.4.5 Dielectric Strength (Break Down voltage) .................................................................... 18
2.4.6 Partial Discharge .......................................................................................................... 18
iii
2.4.7 Dielectric Dissipation Factor ....................................................................................... 18
2.4.8 Degree of Polymerization ............................................................................................. 19
2.4.9 Furan Analysis ............................................................................................................. 19
2.4.10 Water Content .......................................................................................................... 19
2.4.11 Recovery Voltage Measurement .............................................................................. 20
2.4.12 Polarization- Depolarization Current Measurements ............................................. 20
2.4.13 Frequency Response Analysis .................................................................................. 20
2.4.14 Temperature ............................................................................................................ 21
2.4.15 Vibration Analysis ................................................................................................... 21
2.4.16 Bushing Monitoring ................................................................................................. 21
2.4.17 Tapchanger Monitoring ........................................................................................... 22
3 CHAPTER 3 DISSOLVED GAS ANALYSIS (DGA) ........................................................... 23
3.1 INTRODUCTION .................................................................................................................. 23
3.2 DGA QUANTIFICATION METHODS ..................................................................................... 24
3.2.1 Gas Chromatography (GC) .......................................................................................... 24
3.2.2 Hydrogen On-Line Monitor .......................................................................................... 28
3.2.3 Photo-Acoustic Spectroscopy (PAS) ............................................................................. 29
3.3 FACTORS AFFECTING THE OIL AND CORRESPONDING FAULTS............................................ 32
3.3.1 Arcing ........................................................................................................................... 34
3.3.2 Thermal heating/ Pyrolysis........................................................................................... 34
3.3.3 Corona .......................................................................................................................... 34
3.3.4 Overheating Cellulose .................................................................................................. 34
3.4 INTERPRETATION TECHNIQUES FOR DGA ANALYSIS ......................................................... 35
3.5 IEC METHOD ...................................................................................................................... 36
3.5.1 Roger Ratio Method ..................................................................................................... 38
3.5.2 Key Gas Method ........................................................................................................... 40
3.5.3 Doernenburg Ratio Method .......................................................................................... 43
3.5.4 Duval Triangle: ............................................................................................................ 44
3.6 RESULTS DISCUSSIONS AND OBSERVATIONS ...................................................................... 47
4 CHAPTER 4 FUZZY LOGIC MODEL ................................................................................. 49
4.1 INTRODUCTION ............................................................................................................ 49
4.2 FUZZY LOGIC APPLICATIONS ............................................................................................. 49
4.3 THE ADVANTAGES OF FUZZY LOGIC ................................................................................. 50
4.3.1 Solution to nonlinear problems .................................................................................... 50
4.3.2 Ability to handle linguistic variables ............................................................................ 50
4.3.3 Rule reduction in fuzzy rule base .................................................................................. 51
4.4 THE DISADVANTAGES OF FUZZY LOGIC ............................................................................ 51
4.4.1 Highly dependent on domain expert’s knowledge ........................................................ 51
iv
4.4.2 Lack of information ....................................................................................................... 51
4.4.3 Insufficient design standard or methodology ................................................................ 51
4.5 FUZZY LOGIC CONTROL SYSTEM ........................................................................................ 52
4.5.1 Fuzzification .................................................................................................................. 53
4.5.2 Fuzzy Knowledge Base .................................................................................................. 53
4.5.3 Fuzzy inference system (FIS) ......................................................................................... 54
4.5.4 Defuzzification............................................................................................................... 54
4.6 INTRODUCTION TO DGA FUZZY DIAGNOSTIC SYSTEM ....................................................... 56
4.7 THE DESIGN METHODOLOGY OF FUZZY DIAGNOSTIC SYSTEM ........................................... 56
4.8 ASSIGNMENT OF MEMBERSHIP FUNCTIONS ......................................................................... 57
4.8.1 Selection of fuzzy compositional operator (inference engine) ....................................... 57
4.9 FUZZY LOGIC MODELS FOR DGA INTERPRETATION TECHNIQUES ....................................... 58
4.9.1 Fuzzy Logic for Roger ratio method .............................................................................. 61
4.9.2 Fuzzy logic for IEC method ........................................................................................... 63
4.9.3 Fuzzy logic for Doernenburg ........................................................................................ 64
4.9.4 Fuzzy logic for Duval method ....................................................................................... 66
4.9.5 Fuzzy logic for Key Gas method .................................................................................... 67
5 CHAPTER 5 CONSISTENCY, ACCURACY ANALYSES AND PROPOSED FUZZY
LOGIC MODEL ................................................................................................................................. 70
5.1 INTRODUCTION ............................................................................................................. 70
5.2 CONSISTENCY ANALYSIS .................................................................................................... 71
5.3 ACCURACY ANALYSIS ........................................................................................................ 73
5.4 PROPOSED APPROACH TO STANDARDIZE DGA INTERPRETATION ................................ 74
6 CHAPTER 6 CONCLUSION AND FUTURE WORK .......................................................... 80
6.1 RESEARCH CONCLUSION ............................................................................................ 80
6.2 FUTURE RESEARCH RECOMMENDATIONS ............................................................. 81
List of Figures
FIGURE 1. EXTRACTION OF GAS FROM INSULATING OIL USING VACUUM EXTRACTION METHOD [43] ..... 26
FIGURE 2. EXTRACTION OF GAS FROM INSULATING OIL BY USING HEADSPACE METHOD[43] ................. 26
FIGURE 3. AN EXAMPLE OF BASIC GAS CHROMATOGRAPHY [49] .......................................................... 27
FIGURE 4. HYDROGEN ON-LINE MONITOR PRINCIPLE SCHEMATIC DIAGRAM[40] .................................. 28
FIGURE 5. BASIC PRINCIPLE OF PAS PROCESS [62] ................................................................................ 30
FIGURE 6. AN EXAMPLE OF PAS-BASED DGA SYSTEM [61.................................................................... 30
FIGURE 7. CHARACTERISTIC ABSORPTION OF DIAGNOSTIC FAULT GASES [61] ....................................... 31
FIGURE 8. BREAKING CHAIN PROCESS OF FAULT ARCING, CORONA, THERMAL HEATING AND PYROLYSIS
OF CELLULOSE [17]. ..................................................................................................................... 33
v
FIGURE 9. PRINCIPAL GASES FOR EACH FAULT [65]............................................................................... 42
FIGURE 10. DUVAL TRIANGLE [38]. .................................................................................................... 44
FIGURE 11. DUVAL TRIANGLE DIAGNOSTIC EXAMPLE OF A RECLAMATION TRANSFORMER [36]. ......... 46
FIGURE 12. BASIC STRUCTURE OF FUZZY LOGIC CONTROL SYSTEM [17]. .............................................. 53
FIGURE 13. FIVE FUZZY INFERENCE SYSTEMS (2) ANALYSE GAS VALUES (1) REGARDING TO THE DEFECT
CONDITION (3A) OF THE POWER TRANSFORMER, ANALYTICAL RESULTS ARE DEFECT CONDITION
AND RELIABILITY FOR ALL LEAVES (3C) OF THE DEFECT CONDITION TREE (3B) [17]. .................. 55
FIGURE 14. FUZZY LOGIC MODEL FLOW CHART [19]. ............................................................................ 56
FIGURE 15. STEPS FOR CONSTRUCTING A FUZZY LOGIC SYSTEM [17] .................................................... 57
FIGURE 16. FUZZY LOGIC MODELS OUTPUT MEMBERSHIP FUNCTIONS (WHEN TWO THERMAL FAULTS F1
AND F2 ARE CONSIDERED). .......................................................................................................... 59
FIGURE 17.FUZZY LOGIC MODELS OUTPUT MEMBERSHIP FUNCTIONS FOR ONE THERMAL FAULT. ......... 59
FIGURE 18.TYPE OF FAULTS AND GENERATED GASES [19]. ................................................................... 60
List of Tables
TABLE 1: CHEMICAL STRUCTURE OF INSULATING OIL AND FAULT GASES[17] 8
TABLE 2. COMPARISON BETWEEN GC, HYDROGEN ON-LINE MONITOR AND PAS 32
TABLE 3: FAULT GASES GROUP [17] 35
TABLE 4: RELATION BETWEEN FAULT TYPE AND FAULT GASES [65] 35
TABLE 5: IEC RATIO CODES [65]. 37
TABLE 6: CLASSIFICATION OF FAULT BASED ON IEC RATIO CODES [65]. 37
TABLE 7: 20 SAMPLES FOR IEC METHOD. 38
TABLE 8: ROGER RATIO CODES. 39
TABLE 9: CLASSIFICATION OF FAULT BASED ON ROGER RATIO CODES 39
TABLE 10: 20 SAMPLES FOR ROGER RATIO METHOD 40
TABLE 11: RELATION OF FAULT GASES AND TEMPERATURE [17] 41
TABLE 12: L1 LIMIT CONCENTRATION FOR DOERNENBURG RATIO METHOD 43
TABLE 13: DOERNENBURG RATIO METHOD 43
TABLE 14: 20 SAMPLES FOR DOERNENBURG METHOD. 44
TABLE 15: LIMIT AND GENERATION RATE PER MONTH LIMITS [15 45
TABLE 16: FAULT TYPES BY CONSIDERING TWO THERMAL FAULTS 58
TABLE 17: FAULT TYPES BY CONSIDERING ONE THERMAL FAULT. 58
TABLE 18: FAULT TYPES FOR ONE THERMAL FAULT 60
TABLE 19: 20 SAMPLES RESULTS USING FUZZY LOGIC MODEL 69
TABLE 20: FAULT TYPES IDENTIFIED BY VARIOUS DGA [77]. 71
TABLE 21: CONSISTENCY TABLE [65] 72
TABLE 22: ACCURACY OF DGA METHODS 73
TABLE 23: ASSET MANAGEMENT DECISION [77]. 77
TABLE 24: - RESULTS OF 50 SAMPLES FROM 2000 (BASED ON CONSIDERING TWO THERMAL FAULTS, F5
IS NORMAL) 78
vi
TABLE 25: - RESULTS OF 25 SAMPLES FROM 2000 (BASED ON CONSIDERING ONE THERMAL
FAULT, F4 IS NORMAL) 79
vii
ABSTRACT
Power transformer is one of the main components in any power transmission or
distribution network. Transformers have complicated winding structures and are
subject to electrical, thermal and mechanical stresses. During the last few years, there
has been a trend of continuous increase of transformer failures. It is therefore vital to
correct diagnose their incipient faults for safety and reliability of an electrical
network. Thus, these transformers are needed to be routinely maintained. Due to the
large number of transformers of different makes and capacities, routine maintenance
and diagnosis of such transformers are difficult as different transformers exhibit
different characteristics and problems. By means of dissolved gas analysis (DGA), it
is possible to distinguish faults such as partial discharge (corona), overheating
(pyrolysis) and arcing in a great variety of oil-filled equipment. Dissolved gas
analysis is one of the most effective tools for power transformer condition
monitoring. There are many traditional interpretation techniques for DGA results
including Key Gas, Doernenburg, IEC Ratio, Roger’s Ratio and Duval Triangle.
However DGA interpretation is still a challenge issue as all available techniques rely
on personnel experience more than standard mathematical formulation. As a result,
various interpretation techniques do not necessarily lead to the same conclusion for
the same oil sample. Furthermore, significant number of DGA results fall outside the
proposed codes of the current based-ratio interpretation techniques and cannot be
diagnosed by these methods. Moreover, ratio methods fail to diagnose multiple fault
conditions due to the mixing up of produced gases.
To overcome these limitations, this thesis introduces a new fuzzy logic approach to
reduce dependency on expert personnel and to aid in standardizing DGA
interpretation techniques. The approach relies on incorporating all existing DGA
interpretation techniques into one expert model. DGA results of 2000 oil samples
that were collected from different transformers of different rating and different life
span are used to establish the model. Traditional DGA interpretation techniques are
used to analyze the collected DGA results to evaluate the consistency and accuracy
of each interpretation technique. Results of this analysis were then used to develop
the proposed fuzzy logic model.
Keywords: Power Transformer, condition monitoring, dissolved gas analysis, fuzzy
logic modelling.
ix
ACKNOWLEDGMENTS
I would like to take this opportunity to express my deep gratitude to my supervisor
Dr. Ahmed Abu- Siada for introducing me to the main working hypothesis for this
thesis plus his tal support, advice, stimulating suggestions and encouragement
throughout the research work. His guidance, valuable ideas, optimism as well as the
sense of humour helped me throughout this research and writing of this thesis.
I cordially thank Professor Mohammad A.S. Masoum for his consistent support and
encourage.
I would like to express my deep appreciation for Professor Syed M. Islam for his
support on many occasions.
I wish to thank my sisters (Wirood, Muazaz and Etabb) for their encouragement
and support.
Finally, I thank my husband Mundher and my children (Nooralhuda,
Zainulabdeen, Mujtaba, Ryaheen and Mariam) for their understanding and
support over the years, enabling me to complete this work.
x
LIST OF PUBLICATIONS
-A. Abu-Siada, S. Hmood and S. Islam, “A New Fuzzy Logic Approach for
Consistent Interpretation of Dissolved Gas-in-Oil Analysis”, IEEE Transactions on
Dielectrics and Electrical Insulation, vol 20, No.6, pp. 2343-2349, December 2013.
-A. Abu-Siada and S. Hmood, “Fuzzy Logic Approach for Power Transformer Asset
Management Based on Dissolved Gas-in-Oil Analysis”, Chemical Engineering
Transactions, Vol. 33, No. 2, pp. 997- 1002, September 2013.
-S. Hmood, A. Abu-Siada, Mohammad A. S. Masoum and Syed M. Islam,
“Standardization of DGA Interpretation Techniques using Fuzzy Logic Approach”,
presented at the Condition Monitoring and Diagnosis conference, Bali, Indonesia,
September 2012
-A. Abu-Siada and S. Hmood, “Fuzzy Logic Approach for Power Transformer Asset
Management Based on Dissolved Gas-in-Oil Analysis”, presented at the Prognostics
and System Health Management Conference, Milan 8-11 September, 2013.
xi
Acronyms
°C Degree Centigrade
AM Acoustic Measurements
AS Australian Standard
ASTM American Society for Testing and Materials
Av Average
B-Av Below Average
BDV Break Down Voltage
H2 Hydrogen
C2H2 Acetylene
C2H4 Ethylene
C2H6 Ethane
CBM Condition Based Maintenance
CH4 Methane
cm Centimetre
CM Condition Monitoring
CO Carbon Monoxide
CO2 Carbon Dioxide
Corr. Correction
Crit. Critical
CT Current Transformer
Cw Water Content
db Decibel
DBV Dielectric Breakdown Voltage
dc Direct Current
PD Partial Discharge
DDF Dielectric Dissipation Factor
Dev. Deviation
DGA Dissolved Gas Analysis
DP Degree of Polymerization
FLM Fuzzy Logic Model
xii
FIS Fuzzy Inference System
Elect. Electrical
EM Electrical Measurements
Eq. Equation
F Fast
FRA Frequency Response Analysis
ft Foot
G Good
GC Gas Chromatography
GE General Electric
gm Gram
HST Hot spot Temperature
HV High Voltage
Hz Hertz
IEEE Institute of Electrical and Electronics Engineers
IR Insulation Resistance
KHz Kilo Hertz
KV Kilovolt
KWh Kilowatt Hour
L Litre
LV Low Voltage
m Meter
M/DW Moisture by Dry Weight
Mech. Mechanical
Med. Medium
mg Milligram
mg/L Milligram per Litre
MHz Mega Hertz
mls Millilitres
MV Megavolt
MV/cm Megavolt per Centimetre
xiii
MVA Mega Volt Ampere
N2 Nitrogen
nm Nanometre
Nor. Normal
O2 Oxygen
OLTC On Load Tapchanger
PDC Polarization and Depolarization Current
IFT Interfacial Tension
ppb Part Per Billion
ppm Part Per Million
RVM Recovery Voltage Measurement
Sat. Satisfactory
Sig. Significant
TDCG Total Dissolved Combustible Gas
Th. Thermal
TTFF Time to First Failure
US United States
CFIDS Compared Fault Intelligent Diagnostic Solution
V Volts
VG Very Good
VH Very High
V-Poor Very Poor
L Micro Litre
5
1. CHAPTER 1
INTRODUCTION
1.1 PROJECT OVERVIEW-BACKGROUND OF FAULT DIAGNOSIS OF
POWER TRANSFORMER
Utilities rely heavily on their infrastructure to provide a reliable power supply to
customers. Power transformers are among the most important and expensive
equipment of the power system and their reliability directly affects the safety of
power operations. Reliable and efficient fault-free operation of large power
transformers is essential for a reliable electricity supply. Transformer failures are
often catastrophic and almost always cause irreversible internal damage. Therefore,
accurate evaluation of power transformer conditions is essential. Condition-based
monitoring uses advanced fault diagnosis techniques to identify on-line and off-line
incipient faults and to provide real-time transformer conditions. To optimize
maintenance schedules, various condition monitoring techniques of power
transformer have been developed in attempt to reduce operating costs, enhance the
reliability of operation, and improve power supply and service to customers.
Notably, monitoring should be distinguished from diagnostics. Monitoring refers
mainly to data acquisition, sensor development, data collection, and development of
methods for condition measurement of power transformers. Diagnostics is the step
following monitoring; in this sense, diagnostics can interpret data measured off-line
and online.
The insulation system within a power transformer consists of oil and paper. Due to
the high temperatures inside the power transformer; oil and paper decomposition
occurs and evolves gases inside the oil which will decrease the dielectric strength of
oil and paper. These gases are hydrogen (H2), methane (CH4), ethane (C2H6),
ethylene (C2H4), acetylene (C2H2), carbon monoxide (CO), and carbon dioxide (CO2).
Apart from CO, CO2 all other gases are due to oil decomposition while CO and CO2
are due to paper decomposition so it is very essential to measure the amount of these
gases in transformer oil to identify incipient faults within the power transformer.
Dissolved gas in-oil analysis (DGA) is one of the most effective tools for power
transformer condition monitoring. DGA is a sensitive and reliable technique
for the detection of incipient fault condition within oil-immersed
6
transformers. Through DGA technique, the presence of certain key gases is
monitored and quantified. There are a number of methods developed for
analysing these gases and interpreting their significance such as Key Gas,
Rogers Ratio, Doernenburg, IEC Ratio and Duval Triangle. This project
investigates the accuracy and consistency of these methods in interpreting
various transformer conditions. The evaluation is carried out on DGA data
obtained from the Western Power, OMICRON and Doble companies [13].
The Key gases considered are hydrogen (H2), methane (CH4), ethane (C2H6),
ethylene (C2H4) and acetylene (C2H2). Traditional DGA interpretation
techniques are used to analyze the collected DGA results to evaluate the
consistency and accuracy of each interpretation technique. Results of
consistency and accuracy analyses are then used to develop a fuzzy logic
model that incorporates the key features of various DGA interpretation
techniques.
Several techniques have been developed in the literature using artificial
intelligence for more accurate diagnosis and they are mostly suitable for
transformers with a single fault or a dominant fault [1-13]. In principle,
these methods are still based on the specific set of codes defined from
certain gas ratios. In addition, there are also some other limitations in the
previous fuzzy diagnosis methods. As an extension of previous work [14], a
new method has been developed to employ fuzzy boundaries between
different DGA methods and combine them with same fuzzy approach.
Due to the need for continuous demand of electricity, transformers will not
stop operating except when faults occur in them or during maintenance.
Because of this factor, the companies usually spend a lot of money for the
maintenance of the transformers to ensure that they are in good operating
conditions. However, the transformer is usually subjected to thermal and
electrical stresses when operated over a long period of time. These stresses
could break down the insulating material and release gaseous decomposition
products, which if excessive could cause explosion and, therefore, should be
avoided. Presently, with the development of new technologies and new
findings from researchers around the world, there are a variety of diagnostic
methods for detecting and predicting the condition of the power
7
transformers. An IEEE standard (C57.91-1995) [15] introduced the DGA as
one of the most accepted methods for detecting incipient fault conditions in
power transformers. From the DGA test results, appropriate actions can then
be taken to either carry out preventive maintenance or repairs the
transformer.
1.2 CONDITION MONITORING BY DISSOLVED GAS ANALYSIS (DGA)
Dissolved gas in oil analysis is an outstanding method to detect the incipient
insulation or concealed faults of the oil immersed power transformer. Some small
quantities of gases are liberated when insulating oil’s face to abnormal electrical or
thermal stresses. The degradation of these gases is related to fault types. By mean of
DGA, it is feasible to differentiate variety types of fault such as partial discharge
(PD), thermal faults or arcing in huge various oil filled equipment. To distinguish
trends and determine the severity of incipient faults some oil samples must be taken
regularly from power transformers. For accessing the health condition of a
transformer the information which is taken from the analysis of gases dissolved in
insulating oil is very essential and valuable. This considerable information Data from
DGA can provide
Progressive warning of developed faults
Monitoring the rate of fault development
Approve the existence of faults
Means for the schedules of repairs
Condition Monitoring within overload.
1.2.1 Dissolved Gases Analysis Methodology
Utilities and transformer owners are increasing their focus on transformer
maintenance in order to achieve extension in its life and have a better return on their
investment. Therefore, transformers are of high concern from the point of asset
management irrespective of their failure rate. Dissolved gas analysis consists of
sending transformer oil samples to a commercial laboratory for testing, quantification
of the key gases dissolved in the sample using chromatographic techniques and
interpreting the results. The oil in power transformer acts as a dielectric media which
is an insulator and as a heat transfer agent. Normally, the insulated oil fluids are
composed of saturated hydrocarbons called paraffin, whose general molecular
8
formula is CnH2n+2 with n in the range of 20 to 40 while the cellulose insulation
material is a polymeric substance whose general molecular format is
[C12H14O4(OH)6]n with n in the range of 300 to 750 [16]. These molecules are
connected and linked together to form a chain-liked manner by hydrogen and carbon.
The structured formula of the insulating oil is shown in table 1[17].
Table 1: Chemical structure of insulating oil and fault gases[17]
1.3 MOTIVATION OF RESEARCH
DGA is widely accepted as the most reliable technique for the earliest detection of
9
incipient faults in transformers. Nowadays, with the emergence of new technologies
and new finding from many researchers around the world, there are a variety of
diagnostic methods for detecting and predicting the condition of the power
transformers. Some of these studies are briefly summarised below:
Islam et al [18] has investigated how fuzzy logic technique can be used to interpret
the DGA method in fault diagnosis. In this paper, the researchers adopted a novel
fuzzy logic approach to develop computer software based intelligent interpretation
model. Abu-Siada et al [19] presented a novel fuzzy logic approach to identify the
transformer criticality based on the DGA of oil- filled power transformers. Fuzzy
logic model was developed using traditional DGA interpretation techniques.
Though considerable efforts in developing such software have been carried out
universally over the past few years, much of them are still in their under
development. Most of the software is being sponsored by utility companies for their
own in-house usage and none of them is available commercially. Even if such
software is available commercially, factors such as different manufacturer’s
specifications, trends of operations, and local climatic conditions, etc. may make
such software unsuitable or may not directly applicable in some of Countries. Fault
diagnosis software which is able to interpret diagnostic results more accurately so as
to cut cost and ensure better quality service. To achieve the best performance, this
intelligent fault diagnosis system must be developed to suit the natural characteristics
of local transformers. However, countries with similar environment, transformer
usage and other criteria may find this system useful and applicable with minor
modification. Hence, this project is motivated by two factors, the first is to predict
earlier fault that enable precautionary measures to be undertaken so as to minimize
the risk of transformer explosion. The second factor is to standardise the outcome of
various DGA interpretation techniques as they are not fully consistent and they may
lead to different interpretation for the same oil sample.
1.4 OBJECTIVES
This project main objective is as follows:
1. To develop Fuzzy Logic Models for various DGA interpretation
techniques to reduce the dependency on personnel experience.
2. To investigate the consistency and accuracy of various DGA
10
interpretation techniques.
3. To incorporate all techniques in one Fuzzy model based on the
consistency and accuracy analysis; Based on the proposed Fuzzy
Logic Model, transformer criticality depending on DGA will be
calculated.
1.5 THESIS OUTLINE
A brief description of the contents of each chapter of this thesis is as follows:
Chapter 1 (Introduction and research objectives)
Chapter 2 (Power Transformer Condition Monitoring and diagnosis):
It explains the critical components failure and various transformer diagnostics.
Chapter 3 (Dissolved Gas Analysis):
It explains the Factors Affecting the oil and corresponding Faults,
then explains the most commonly used methods for DGA
Interpretation.
Chapter 4 (Fuzzy Logic Model):
It will explain the proposed approach, Fault Diagnosis System Steps.
Chapter 5 (Consistency, Accuracy Analysis and Proposed Fuzzy
Logic Model):
It will explain the Consistency and Accuracy.
Chapter 6 (CONCLUSION and Future Work):
It will explain the overall fuzzy logic model that incorporates 5
traditional interpretation techniques.
11
2 CHAPTER 2
POWER TRANSFORMER CONDITION MONITORING
AND DIAGNOSIS
2.1 INTRODUCTION
Diagnostics play an important role in the life expectancy of a transformer. Besides
other stresses, transformer’s insulation deterioration is significantly affected by
operational stresses, adverse ambient conditions, geomagnetic storms and
contaminations thus shortening its design life [20]. Insulation aging in a transformer
is a complex and irreversible phenomenon. The majority of transformers (around
65%) are serving with an age of 30 years and above. Therefore, diagnostics and
surveillance become important in power transformers to determine their respective
life expectancy. Advanced monitoring and diagnostics provide an early warning of
abnormalities in a transformer. Non-intrusive diagnostics and accurate interpretation
together with expert system determine the transformer’s criticalities.
Condition monitoring (CM) in power transformers provides better information
regarding operating performance. The overall integrity of a transformer can be
assessed and enhanced asset management strategies be developed. Transformer life
can be extended by implementing the required operational criteria and cost effective
maintenance strategies. Condition monitoring facilitates detection of incipient faults,
thus the catastrophic failures can be averted.
2.2 CRITICAL COMPONENTS FAILURE
The main components that play an important role in a transformer failure,
performance and life expectancy are mentioned below [21]:
2.2.1 Core
Core is the main active, inaccessible and most expensive element of a transformer.
Its inspection requires removal of oil and exposure of the insulation to air having
moisture and other contaminants. It is usually not recommended as it increases the
failure risk. Core clamping parts in some designs are bonded electrically. This allows
the current to flow within the structure and any poor connection results in hotspots.
12
2.2.2 Windings
Power transformer windings are usually made of copper strands covered with paper
or continuously transposed enamelled copper strands with an overall paper covering.
In older transformers cotton insulation, varnished paper and tape were used. In some
designs, enamel covered conductors without paper insulation have also been used.
The recent designs use soft or hard pressboard for interlayer, inter-turn, end
insulation and winding cylinders.
Winding insulation (cellulose) condition can be assessed using a diagnostic test by
measuring the degree of polymerization in the cellulose chain. It is difficult to obtain
paper samples from the relevant winding location without its dismantling. Cellulose
samples taken from some accessible locations can provide useful information with
certain limitations. Moreover, furan in oil provides sufficient information on the
cellulose integrity.
2.2.3 Insulating Oil
Mineral oil is derived predominantly from naphthenic crude oils. The process
includes acid treatment, solvent extraction, de-waxing and hydrogen treatment in a
combination [22]. Insulating oil is an integral, inseparable component of the
transformer insulation system and serves as a dielectric and cooling medium. The oil
undergoes chemical changes due to oxidization, forms sludge and affects the
insulation properties. Oil dielectric properties become more critical in the area of
high electrical stresses, such as bushing, corona shields and end connections. The oil
poor dielectric properties may result in a failure.
Transformer oil analysis is a key source to detect incipient faults, fast developing
faults and insulation trending. Due to the abnormal electrical and thermal stresses,
insulation (oil and cellulose) decomposition occurs. This decomposition evolves
some key gases as mentioned in chapter 1. Transformer premature failures can be
limited with known gas generation rate and accurate faults interpretation.
The insulating oil being the immersion medium for core and coils is sampled for
analysis of various soluble and insoluble compounds produced by materials
immersed in the oil. The most common in practice assessments are dissolved gas
analysis (DGA), furans from paper degradation, dissolved metals from wear or
reactions with metallic components and sludge. The analysis provides a critical
13
assessment of the transformer [12-24]. The oil quality is of high concern and is
always required to be maintained by keeping the parameters within the limits.
2.2.4 Insulating Paper (Cellulose)
The electrical windings in a power transformer consist of paper insulation immersed
in insulating oil. Transformer life mainly depends on the integrity of its solid
insulation (cellulose). It is important to monitor the insulation condition continuously
for overall assessment of a transformer [5]. The insulating paper is developed from
cellulose mainly composed of long chains of glucose. In cellulose diagnostics, the
degree of polymerization (DP) is a measurement of an average number of glucose
units per molecular chain. Cellulose samples are difficult to test as it involves
removal of paper from the winding. There is a good correlation between DP and
tensile strength of an insulating paper [1, 4, 8].
2.2.5 Bushings
Bushings are the most significant ancillary components of a transformer.
Transformer bushings are designed to withstand the load current (including
overload), and fault currents. Bushings operate under extreme environmental
conditions and their performance is affected by moisture and pollution. The bushings
in construction are either oil impregnated paper or synthetic resin bonded paper
(SRBP). Both types are subject to moisture ingress if their seals fail and later
delamination occurs.
The dielectric failure of a bushing may cause short circuit on a transformer. A
flashover of external insulation can also result in damage to the bushing, it is
therefore, important to monitor its integrity. Failure of a bushing may damage
adjacent transformers and the plant. Failure may occur due the following reasons:
Mechanical defect in material.
Mechanical forces due to mishandling.
Overheating due to improper connection.
Poor impregnation.
Overstressing.
Thermal runaway due to high dielectric losses as a result of
contamination.
Aging.
14
2.2.6 Tapchanger
It is the only component in a transformer that has moving parts. Drive mechanism
failure may prevent its operation, leaving the transformer in-operation on a fixed tap.
A dielectric failure within a tapchanger can also result in a short circuit between
tapping sections of the transformer. This, in turn, can result in severe damage to
winding. Moving parts of the tapchanger are usually subjected to wear and tear.
Contamination, such as carbon and moisture may appear in the switch assembly due
to mechanical operation and can result in a dielectric failure.
Tapchanger in power transformers are of separate diverter and selector-switch
designs. The diverter switch is in a separate oil filled compartment either within the
transformer tank or external to the tank. The selector switches are inside the tank or
in a separate compartment sharing oil with the main tank.
The duty of the tapchanger is determined by the number of operations per day/year
and the magnitude of load current. Traditional condition monitoring is carried out by
inspection of the diverter switches, measuring contact wear and contact resistance.
Some on-line monitoring systems for tapchangers are also in practice.
2.2.7 Fans
Fans are used to provide more efficient oil to air heat exchange in a power
transformer. The fan failure can have a major effect on the transformer rating. Fan
failure occurs mainly due to the phase failure, over-voltage or bearings problem. A
fan failure results in de-rating the transformer as the coolers become ineffective
without them. Visual inspection and sound observation can provide information on
the performance and availability of fans. Fan motor winding integrity can be verified
by applying insulation resistance (IR) test during the routine maintenance program.
2.2.8 Pumps
Pumps are usually considered as reliable components. A pump failure results in loss
of oil/air heat exchangers and causes de-rating of the transformer. Failure of a pump
bearing is likely to result in metallic particles being pumped into the transformer
tank. These particles could result in a dielectric failure within the transformer.
15
2.3 SMALL GAS SAMPLES (SGS) OIL SAMPLING (PHYSICAL AND
CHEMICAL TEST)
SGS provides online transformer testing (OTT) to determine the basic electrical
properties of transformer oil. From this one can identify if the oil is suitable for
further use or whether necessary actions like filtration or regeneration are needed.
Transformer oil is crucial in the cooling process of the transformer but it is subject to
electrical and thermal stress or possible chemical contamination during its lifecycle.
Periodic testing provides an indicator of these issues as they arise. A fast reaction
ensures enhanced component life and reduced oil costs, while maximizing safety and
preventing untimely failures. The SGS OTT programs are innovating, such as
Laboratory Information Management Systems (LIMS) software and a live web
interface, continually being introduced. This program, combined with the
recommended testing schedule, provides a complete solution to transformer testing
needs. All analyses are carried out according to IEC, ASTM; ISO standards in
NATA certified laboratories.
2.4 DIAGNOSTICS
When sending oil samples to a laboratory for DGA, one should also specify other
tests that reveal oil quality. Various oil analysis techniques are summarised below.
2.4.1 Gas Chromatography
The DGA technique involves extracting or stripping the gases from the oil and
injecting them into a gas chromatograph (GC). Detection of gas concentrations
usually involves the use of a flame ionization detector (FID) and a thermal
conductivity detector (TCD). Most systems also employ a methanizer, which
converts any carbon monoxide and carbon dioxide present into methane so that it can
be burned and detected on the FID, a very sensitive sensor.
Removing the gas from the oil is one of the most difficult and critical portions of the
procedure. The original method, now ASTM D3612A, requires that the oil be
subjected to a high vacuum in an elaborate glass-sealed system to remove most of the
gas from the oil. The gas is then collected and measured in a graduated tube by
breaking the vacuum with a mercury piston. The gas is removed from the graduated
column through a septum with a gas-tight syringe and immediately injected into GC.
In the present modern day laboratory, however, mercury is not a favourite material of
16
chemists. For this reason, two additional extraction techniques have been developed
to eliminate mercury [40-42].
ASTM D3612B is called the direct injection technique. In this method, the stripping
of gases from the oil and gas analysis takes place inside the GC. Originally
developed in the mid-1980s for this application, the process involves injecting the oil
into a sample loop in the GC. When the GC run is initiated, the sample loop transfers
the oil through a series of valves into a stripper column. The stripper column is
composed of metal spheres in one end in which the oil overlays the surface of the
spheres to increase the surface area. Carrier gas is passed over the spheres and
extracts dissolved gases from the oil, which then pass through a series of columns
and on through the detectors. The oil is back-flushed and purged from the system
before the next sample is introduced [26].
2.4.2 Interfacial Tension (IFT)
This test, ASTM D-971-91, standard test method or interfacial tension of oil against
water by the ring method [27], is used by DGA laboratories to determine the
interfacial tension between the oil sample and distilled water. The oil sample is
placed in a beaker of distilled water at temperature of 25° C. The oil floats because
its specific gravity is less than that of water. There should be a distance line between
the two liquids. The IFT number is the amount of force (dynes) required to pull a
small wire ring upwards a distance of 1 centimetre through the water/oil inference. A
dyne is a very small unit of force equal to 0.000002247 pound. Good clean oil will
make a very distinct line on top of the water and give an IFT number of 40-50 dynes
per centimetre of travel of the wire ring.
As oil ages, it is contaminated by tiny particles (oxidation products) of the oil and
paper insulation. Particles on top of the water extend across the water/oil interface
line which weakens the surface tension between the two liquids. Particles in oil
weaken interfacial tension and lower the IFT number. IFT and acid number together
are an excellent indication of when oil needs to be reclaimed. It is recommended the
oil be reclaimed when the IFT number falls to 25 dynes per centimetre. At this level,
the oil is very contaminated and must be reclaimed to prevent slugging, which begins
around 22 dynes per centimetre [28]. If the oil is not reclaimed, sludge will settle on
windings, insulation, cooling surfaces, etc., and cause loading and cooling problems.
17
This will greatly shorten transformer life [29, 30]. There is a definite relationship
between acid number, the IFT, and years-in-service.
2.4.3 Acid Number
Acid number is the amount of potassium hydroxide (KOH) in milligrams (mg) that it
is taken to neutralize the acid in 1 gram of transformer oil. The higher the acid
number, the more acid is in the oil. New transformer oils contain practically no acid.
Oxidation of insulation and oils forms acids as the transformer ages, oxidation
products form sludge particles in suspension in the oil which rains (precipitates out)
inside the transformer. The acids attack metals inside the tank and form soaps (more
sludge). Acid also attacks cellulose and accelerates insulation degradation. Slugging
has been found to begin when the acid number reaches 0.40; it is obvious that the oil
should be reclaimed long before it reaches 0.40. It is recommended that the oil be
reclaimed when the acid number reaches 0.20 mg KOH/gm. [28]. As with all others,
this decision must not be based on one DGA test; one should look for a rising trend
in the acid number each year [28].
2.4.4 Moisture
It is critical for life extension to keep transformers as dry and as free of oxygen as
possible. Moisture and oxygen cause paper insulation to decay faster than normal to
form acids, metal soaps, sludge, and more moisture. Sludge settles on windings and
inside the structure, causing transformer cooling to be less efficient, and slowly, over
time, temperature rises. Acids cause an increase in the rate of decay, which forms
more acid, sludge, and moisture at a fast rate [31]. This is a vicious cycle of
increasing speed with deterioration forming more acid and causing more decay. In
addition, oxygen inhibitor should be watched carefully in DGA testing. Moisture,
especially in the presence of oxygen, is extremely hazardous to transformer
insulation. Recent EPRI studies show that oxygen above 2,000 ppm dissolved in
transformer oil is extremely destructive. Each DGA and double test result should be
examined carefully to see if water content is increasing and to determine the moisture
by dry weight (M/DW) or precent saturation in the paper insulation. When 2%
M/DW is reached, plans should be made for a dry out. It is recommended not to
allow the M/DW to go above 2.5% in the paper or 30% oil saturation before drying
out the transformer [28]. Each time the moisture is doubled in a transformer, the life
18
of the insulation is cut by one-half as in general, life of the transformer is the life of
the paper, and the life of the paper is extended by keeping out moisture and oxygen
[28].
2.4.5 Dielectric Strength (Break Down voltage)
The dielectric strength test is not extremely valuable; moisture in combination with
oxygen and heat will destroy cellulose insulation long before the dielectric strength
of the oil has indicated anything is going wrong. This test measures the voltage at
which the oil electrically breaks down. The test gives an indication of the amount of
contaminants (water and oxidation particles) in the oil. DGA laboratories typically
use IEC 60156-95 or ASTM D1816 test, the minimum oil breakdown voltage is 20
kV for transformers rated less than 288 kV and 25 kV for transformers 287.5 kV and
above [28, 33]. If a dielectric strength test falls below these numbers, the oil should
be reclaimed. It is worth mentioning that one should not base any decisions on one
test result, information from several DGA and other tests should be investigated
before making any decision[28- 33].
2.4.6 Partial Discharge
Insulation system in a power transformer is quite susceptible to partial discharge
(PD) activity. The PD test is vital as there is always a chance that the transformer
might have suffered damage during transportation and commissioning. In addition,
power transformers are under intense electromechanical stresses while serving under
abnormal operating conditions. The forces due to fault current may cause windings
deformation and induced partial discharge activity. Generator transformer can suffer
electromechanical stress during improper synchronization [32]. Partial discharge is a
major source of insulation failure in power transformers. On-line PD measurements
on transformers can be performed using acoustic measurements and electric
detection methods [26, 32].
2.4.7 Dielectric Dissipation Factor
The dielectric dissipation factor (DDF) measurement is normally carried out at power
frequency using a voltage level that is high enough to overcome any local
interference at the test site, typically 10 to 30 KV [31]. The DDF represents the ratio
between the real and imaginary parts of the permittivity [32]. Testing is often
19
performed at different temperatures, as the presence of contaminates including
moisture, influences the DDF at different temperatures. This technique is used to
identify the concentration of these contaminants. DDF can also be used at very high
voltages, often above normal operating stresses and can highlight serious problems
that are developing but have not become apparent in a normal operation. This can
include contamination or physical deterioration.
2.4.8 Degree of Polymerization
The Degree of polymerization (DP) measurement is currently used to assess the
cellulose condition. A strong correlation also exists between DP and tensile strength
of cellulose [33]. These properties are used to assess end of reliable life of paper
insulation. This method helps to determine the overall degradation of the cellulose
within the transformer being sampled. Paper analysis using DP test plays a decisive
role in transformer rebuilding or scraping. Degree of polymerization value is the
main parameter in the relation between the insulation deterioration and formation of
aging products.
2.4.9 Furan Analysis
Furans are the major degradation products of paper found in oil [34]. The 2-
furaldehyde in oil is the most prominent component of paper decomposition and is
directly related to PD value [31- 34]. Paper in transformer does not age uniformly
and variations are expected with temperature, moisture distribution, oxygen levels
and operating conditions.
It is well known that the solid insulation in transformers degrades with time at rates
which depend on the temperature and the amount of moisture, oxygen and acids in
the insulation system [33]. Most of solid materials are cellulose-based products such
as paper, pressboard and tapes. When degradation of solid insulation occurs, the
cellulose molecular chains get shorter and chemical products such as furanic
derivatives; CO and CO2 are produced and dissolve into the oil [35, 39].
2.4.10 Water Content
Imperfect sealing of a transformer tank causes water to enter oil from the
atmosphere. The water also arises as a product of oxidation aging during poly-
condensation reactions. Water, which is emulsificated substantially decreases the
20
value of dielectric breakdown voltage and negatively affects its resistivity [22]. If
water contents exceed the limit values in transformer insulation oil, then it is
essential to carry out the drying of oil.
The life of the transformer mainly depends on that of the paper. The purpose of oil
dry out is to keep out moisture and oxygen. Result should be examined carefully to
know if water is increasing and to determine the moisture by dry weight or percent
saturation in paper insulation.
2.4.11 Recovery Voltage Measurement
The recovery voltage measurement (RVM) technique is applied by charging the
transformer up to 2 KV, and then discharging it for half the charging time [20]. After
the discharge cycle, the transformer is free to recover to a voltage which depends on
its internal condition, temperature and the applied voltage. The charging time
typically range from 20ms to as high as 105^s [20]. This test is performed by a
number of utilities in Europe, and now in Australia [18]. It has been demonstrated to
be more sensitive to moisture in paper than DDF measurements and provides an
indication of insulation deterioration [18], a voltage will build up between the
electrodes on dielectric.
2.4.12 Polarization- Depolarization Current Measurements
The polarization and depolarization current (PDC) uses the dielectric system
response in time domain [34]. A single measurement allows to distinguish the
influences of material properties and geometrical structures, e.g. in an oil- board
barrier system. Measurements on transformer insulation models have shown that oil
conductivities, board conductivities and water contents in the barriers can be
measured [33,34]. When a direct voltage is applied to a dielectric for a long period of
time, and then short circuited for a short period, thereafter, opening the short circuit
will cause the charge bounded by the polarisation turn into free charge.
2.4.13 Frequency Response Analysis
Frequency response analysis (FRA) is a measure of input impedance or transfer
function over a wide frequency range [30-33]. The comparison of the response prior
to and after an incident such as a through fault, provides a more sensitive
measurement of winding movement [21]. Frequency response measurements can be
21
used to detect winding movements in transformers arising from short circuit forces,
as well as many other faults [21].
2.4.14 Temperature
Transformers serving with overload rating will consequently shorten their expected
service life. A transformer’s loading capacity is related to the exposure of its
insulation to heat, the highest temperature of which is referred to as the hotspot
temperature (HST). The HST’s effect on the paper insulation is used to quantify the
limit of its temperature range over a calculated period of time [23]. Gas production
rates increase exponentially with temperature and directly with volume of oil and
paper insulation at high enough temperature to produce gases [23]. As distance
increases from the fault (hotspot), temperature goes down and the rate of gas
generation also falls. Because of the volume effect, a large heated volume of oil and
paper will produce the same amount of gas as a smaller volume at a higher
temperature [23]. This is one of the reasons that interpreting DGA is not an exact
science.
2.4.15 Vibration Analysis
The objective of this analysis is to detect the mechanical vibration of a transformer,
to detect its internal changes such as, ones caused by loss of winding clamping
pressure or other mechanical problems [22, 23]. The mechanical integrity of
transformer windings is maintained in part by the winding clamping system. When
there is a reduction or loss of clamping pressure, an excessive winding movement
results. This movement can result in mechanical failure of the windings or cause
damage to the winding insulation. This subsequently causes an electrical failure of
the transformer.
2.4.16 Bushing Monitoring
Power transformer bushing condition monitoring can be carried out off line as well
as online. The off-line measurements are [21-30]:
Oil level.
Insulation resistance.
DDF and capacitance.
DGA of oil (where this can be sampled).
22
The on-line measurements are [25]:
Thermographic scan.
Measurement of tapping current (crudest form of measurement).
Dielectric dissipation factor.
2.4.17 Tapchanger Monitoring
This system uses a current transformer (CT) in a diverter switch compartment. It is
used for protection too, when the time of current flowing in this section exceeds a
preset time [25-36]. The following are important monitoring procedures for an
efficient operation of the tapchanger [21, 37].
Temperature monitoring of the tapchanger compartment.
Monitoring motor mechanism current or torque.
DGA monitoring of oil in diverter switch compartment.
Acoustic monitoring of diverter.
23
3 CHAPTER 3
DISSOLVED GAS ANALYSIS (DGA)
3.1 INTRODUCTION
The insulation system within a power transformer consists of oil and paper. Due to
the high temperature within the power transformer; oil and paper decomposition
occurs and evolves gases inside the oil which will decrease the dielectric strength of
the insulation system. By means of dissolved gas analysis, it is feasible to detect
faults such as overheating, partial discharge (corona) and arcing in huge variety of oil
filled equipment [25]. Similar to blood test or scanner examination of human body,
DGA can give early diagnosis and increase the chances of finding the appropriate
cure. Conventionally, DGA measurement is done in the laboratory environment due
to the complexity of the equipment required. Normally, an oil sample is taken from
operating transformers, and then transported to the laboratory for gas extraction and
measurement processes. There are three common techniques currently used in
laboratory to extract gases from oil sample; vacuum extraction, stripper extraction
and headspace sampling as stated in ASTM D3612 [35]. Another technique that can
be used is known as Shake Test [36]. After extraction process done, all the gases are
analyzed by using gas chromatography (GC). However, due to the time and costs
involved with GC analysis, practically DGA analysis is only performed once a year
for operating transformers. Frequent DGA measurements only take place when
significant fault gases were detected during routine analysis [37].
Dissolved gas analysis is a primary most effective condition monitoring tool to detect
incipient faults in a transformer and indicate overheating, partial discharge and
arcing [1-39]. It facilitates determination of the transformer failure ranking and
impact on the aging mechanism. Geographic sampling and timing of sampling
provide further valuable information about the fault.
DGA is based on the fact that during faults various gases are evolved from the
decomposition of insulating materials. Transformer with online DGA provides better
trending and facilitates to maintain high reliability. DGA is misleading if the oil
sample is taken after the transformer is de-energized and cooled down. Online DGA
gives real-time gassing information while the transformer is serving, several
24
abnormalities can be picked up immediately. Gassing trend is achievable repeatedly
with online DGA and failures could be restricted by taking the appropriate decision
such as restricted loading criteria. Transformer gassing behaviour can be analysed
against the changing load, thermal fluctuations, and temperature respectively. In
conventional DGA under routine maintenance, a de-energized transformer exhibits
different characteristics than being on load [1-12].
Carbon monoxide (CO) and carbon dioxide (CO2) in DGA represent a good source
of cellulose monitoring while a transformer is serving. Presence of acetylene (C2H2)
in the oil by few parts per million (ppm) is an indication of high energy arcing
(600°C and above). Continuous increase in acetylene indicates active internal arcing
and the unit should be taken out of service immediately [65-67].
3.2 DGA QUANTIFICATION METHODS
3.2.1 Gas Chromatography (GC)
Gas chromatography has been used to analyze gases dissolved in insulating oil
during the last 60 years [40]. This technique is introduced by James and Martin in
1952 [41]. Rogers reported in [42] that the first GC diagnose was attempted in 1956
by Howe et al, and later on, regularly used by C.E.G.B for monitoring and routine
assessment since 1968. However, this technique became more popular after IEEE,
IEC and ASTM published guidelines on how to measure and analyze gases dissolved
in transformer insulating oil. Currently, GC analysis is well accepted as the best
among DGA techniques to quantify all gases dissolved in transformer oil including
total dissolved gases (TDG), individual dissolved gases (IDG) and individual gases
present (IGP) in the gas blanket. However, due to the complexity of the equipment
required, GC analysis can only be conducted in laboratory environment, hence
several standards should be followed to handle the gas sample since extracting the oil
sample from an operating transformer and transporting it to the laboratory site till
extracting dissolved gases using GC [43-47].
According to ANSI/IEEEC57.104-1978 Standard [44], oil sample can be stored and
transported to the laboratory by using either calibrated stainless steel cylinders,
flexible metal cans, syringes or glass bottles. However, all containers used must meet
the leak criterion stated in this standard. ASTM D923 [47] stated that amber or clear
glass bottles may use glass-stopper or can be fitted with screw caps having a pulp-
25
board liner faced with tin or aluminium foil, or at least with a suitable oil-resistant
plastic such as polyethylene, polyte tra flouro ethylene (PTFE) or fluoro-elastomers.
ASTM D3612 [43] quotes that gases in the oil can be separated by using vacuum
extraction, stripper column extraction or headspace sampling methods. Vacuum
extraction (shown in Fig. 1) is suitable method to extract a portion of gases, while
stripper column extraction method can extract all gases in oil sample. On the other
hand, headspace sampling (shown in Fig. 2) is used to get a portion of the headspace
gases. Another method to extract dissolved gas in oil sample is developed by
Morgan–Shaffer in 1993 known as Shake Test [36]. Through shake test, dissolved
gases in the oil can be extracted quickly, even at the site.
After the extraction of dissolved gases from the oil sample, it is analyzed by using
GC. A basic GC as shown in Fig. 3 consists of a carrier gas source, a pressure
regulator, a sample injection port and chromatography columns, flow meter, detector,
and recorder or recording integrator [43]. Basic operating principle of GC involves
volatilization of the sample in injection port of a gas chromatograph, followed by
separation of the components of the mixture in chromatography columns [41].
Argon, helium, nitrogen and hydrogen are normally used as carrier gases to transfer
the sample from the injector, via the column, and into a detector or mass
spectrometer [41,43,48]. However, the nature of the carrier gases used may affect the
separation characteristics of the GC system and can alter the sensitivity of the
detection.
26
Figure 1. Extraction of gas from insulating oil using vacuum extraction method [43]
Figure 2. Extraction of gas from insulating oil by using headspace method[43]
27
Figure 3. An example of basic Gas Chromatography [49]
As reported in [48], GC columns can be divided into two categories; packed and
capillary columns. However, majority of GC users use capillary columns/tubes with
a stationary phase coated on the inner wall [41] which has the advantage of
substantially higher separation capacity when compared with the packed columns
[48]. Separation capacity is determined by the portioning of each component
between carrier gas and the stationary phase [41]. Component with delay in the
stationary phase is eluted, while remaining components in the carrier gas flow into a
detector or a mass spectrometer. It is necessary to retain a constant temperature of
gas chromatographic column in order to achieve effective and reliable separation.
The solute molecules existed in the column interacts with the detectors, and is
converted into an electrical signal [48]. This signal is sent to the recording or data-
storage device. The detectors used are specifically designed for gas chromatograph
such as thermal conductivity detector (TCD), flame ionization detector (FID),
nitrogen–phosphorus detector (NPD), flame photometric detector (FPD), electron
capture detector (ECD), atomic emission detector (AED) and electrolytic
conductivity detector (ELCD) [41,48].
As reported in [43], FID is normally used to detect hydrocarbons and carbon oxides
gases due to its greater sensitivity, while TCD is used to detect permanent gases such
as H2, O2 and N2. Alternatively, using mass spectrometer (MS), prior separation of
mixture component is an optional process.
By using GC, some individual gases can be identified included hydrogen, oxygen,
nitrogen, carbon monoxide, carbon dioxide, methane, ethane, ethylene, acetylene,
propane and propylene [43]. Other studies show that GC/MS is also capable of
detecting and analyzing methyl acetate, 2-methylfuran, phenol, methyl formate,
furan, methanol, ethanol, acetone, isopropyl alcohol and methyl ethyl ketone [50-53].
28
3.2.2 Hydrogen On-Line Monitor
Due to the time-consuming and expensive laboratory equipment required for GC
technique, a rugged low-cost with continuously monitoring device known as
Hydrogen On-line Monitor has been developed [40]. Hydrogen On-line Monitor
system has been invented by Syprotec [40, 54], and followed by extensively research
by the Institut de Recherche d’Hydro Quebec (IREQ) since 1974 in order to produce
a rugged low-cost device for site implementation. As it is widely accepted that
majority of faults in oil-filled electrical equipment generate hydrogen gas [55],
Hydrogen On-line Monitor systems is developed to mainly focus on monitoring key
gases such as hydrogen along with carbon monoxide instead of considering all
dissolved gases [56]. By monitoring H2 dissolved in transformer oil, an early
detection of faults growth especially for hot spots, partial discharges and arcing is
warned. Other gases that can be detected by Hydrogen On-line Monitor are ethylene
and acetylene, but in smaller amounts.
The basic principle of the Hydrogen On-line Monitor operation is illustrated in the
schematic diagram of Fig. 4, which shows that the hydrogen emitted from the oil
permeates through a membrane and reacts with the atmospheric oxygen [40],
resulting in a generation of a small current. This current is amplified by electronic
circuits, and translated into gas level in parts per million (ppm). Alarm is actuated
when the gas value reached certain level. Recent improvement in membrane
technology allowed Hydrogen On-line Monitor systems to detect the combination of
hydrogen, carbon monoxide, ethylene and acetylene. With 100% efficiency of
hydrogen, the sensitivity of other gases is yet low; approximately 15% for carbon
monoxide, 8% for acetylene and 1% for ethylene [40, 57].
Figure 4. Hydrogen On-line Monitor principle schematic diagram[40]
29
The basic Hydrogen On-line Monitor consists of a sensor, contacted with the oil and
an electronics unit. The sensor is placed in a rugged brass housing containing the fuel
cell, temperature sensing and the membrane. This sensor can be installed either into a
flange or valve on the transformer pipe work [40], between the cooling bank and the
main tank, or on the upper part of the transformer [57].
Although Hydrogen On-line Monitor is incapable to provide the concentrations of all
fault gases like GC, its reliability to detect incipient faults of power transformer is
proven. Due to the benefit offered by this system, it is reported that approximately
18,000 Hydrogen On-line Monitor systems have been globally installed in 2003 [36].
Additionally, Hydrogen On-line Monitor is not only performed when the membrane
is in contact with moving oil, but also with static oil [40]. There are however few
complications of using Hydrogen On-line Monitor since it is sensitive to temperature
and the reading of the unit may vary with the variation of oil temperature [54].
Hydrogen On-line Monitor accuracy is proven to be ±10% at temperature range
between 20°C to 40°C and cannot provide precise concentrations for fault gases [57].
3.2.3 Photo-Acoustic Spectroscopy (PAS)
Photo-Acoustic Spectroscopy (PAS) is another DGA on-line monitoring technique
that utilizes spectral analysis to detect the volume of absorbed gases based on photo-
acoustic effect [58]. Instead of using composite gas detectors likes Hydrogen On-line
Monitor, PAS employs the common headspace gas extraction technique along with
infrared/acoustic based detector for gas measurement [59]. According to [60], the
first photo-acoustic application was used by Alexander Graham Bell in 1880, when
he found that a sound is emitted when a thin disk is exposed to mechanically
chopped sunlight. Similar effects are observed by using infrared or ultraviolet light.
Since that, PAS has been used in various applications such as ambient air
monitoring, air polluting emission from car exhausts, biological and medical
experiments. The use of PAS technology to monitor power transformer health
condition is still new and is not fully matured yet.
The basic principle of PAS is that fault gases absorb the infrared light energy and
convert it into kinetic energy [59] that lead to sequences of pressure waves (sound)
that can be detected by a microphone. This microphone converts the amount of
30
pressure in the measurement chamber (where the gas sample is exposed to the light)
into electrical signal [61]. The photo-acoustic spectrum of fault gases is recorded by
measuring the sound at different wavelengths, which is used to identify the
concentration of the faults gases involved. By using PAS, several fault gases can be
detected such as H2, CO, CO2, CH4, C2H4, C2H2 and C2H6 [59, 61]. Figs. 5 and 6
show the basic operation of PAS-based DGA system while Fig. 7 shows the
absorption characteristic of fault gases [61].
Figure 5. Basic principle of PAS process [62]
Figure 6. An example of PAS-based DGA system [61
31
A historical analysis reported in [59] concluded that PAS is a very stable diagnostic
tool and suitable for monitoring critical transformers. However, although each fault
gas absorbs the infrared light at specific wavelength, selecting the centre wavelength
is a critical process. Incorrect centre wavelength will cause mutual interference
between gases involved [61]. In fact, there is cross-interference between various
gases including water vapor as shown in Fig. 7. Meanwhile the sensitivity of each
gas is influenced by the wave number of the optical filter and its characteristics
absorption spectrum. Investigation done by Fu et al [58] found that the detection
accuracy of PAS is also influenced by the external gas pressure, vibration, light
temperature and environmental factors.
Figure 7. Characteristic absorption of diagnostic fault gases [61]
A comprehensive comparison between the above three DGA measurement
techniques is given in Table 2.
32
Table 2. Comparison between GC, Hydrogen On-line Monitor and PAS
3.3 FACTORS AFFECTING THE OIL AND CORRESPONDING FAULTS
During normal use, there is usually a slow degradation of the mineral oil which
yields certain gases that dissolve in the oil. However, when an electrical fault
happens inside the transformer, the oil starts to degrade and temperature will rise
abnormally which generates various fault gases at a rapid rate. Different patterns of 6
Method Advantage Disadvantage
GC
- Able to detect and analyze
every individual gas dissolved
in transformer oil.
- Provides highest accuracy
and repeatability results.
- Results can be used to
interpret the faults roots.
- Only can be done in
laboratory due to complex
equipment required.
- Long time required to
complete each test.
- Possibility of missed
diagnostic opportunity due
to limited sample collected
per annum.
- Need an expert to
conduct the test.
HYDROGEN
ON-LINE
MONITOR
- Rugged low-cost with
continuous online monitoring.
- Able to detect incipient
faults.
- Only capable to detect
H2, CO, C2H2 and C2H4.
- Low accuracy at
temperature range 20⁰C
TO 40⁰C.
- Sensitive to oil
temperature variation.
- Unable to interpret faults
roots.
PAS
- Able to provide continuous
online monitoring.
- Capable to detect more
dissolve fault gases than
Hydrogen On- line Monitor
- Results may be used to
interpret the fault roots.
- Results influenced by the
wave number of the
optical filter and its
absorption characteristics.
- Detection accuracy
influenced by external gas
pressure, vibration and
light temperature.
- Still not fully matured.
33
gases are generated due to the different intensities of energy dissipated according to
the type of faults. This phenomenon happens mainly due to the broken chain of the
chemical structure of the insulating oil. As a result, the broken-chain molecule will
form an individual chemical structure known as hydrocarbon or fault gases. The
cause of the dissipation of the fault gases can be divided into 3 categories which are
corona or partial discharge, pyrolysis or thermal heating and arcing. Among the three
common fault cases, the most severe intensity of energy dissipation occurs with
arcing, less with thermal heating and least with corona. Figure 8 illustrates the
breaking chain process within the insulating oil chemical structure of the fault arcing,
thermal heating, corona and pyrolysis of cellulose.
Figure 8. Breaking chain process of fault arcing, corona, thermal heating and pyrolysis of cellulose
[17].
Gases which are produced by the degradation of oil as a result of elevated
temperatures may be caused by several factors as listed below [63]
Severe overloading
Lightening
Switching transients
34
Mechanical flaws
Chemical decomposition of oil or insulation
Overheated areas of the windings
Bad connections which have a high contact resistance
The type of gases present in an oil sample makes it possible to
identify the corresponding type of fault that occurs in the transformer.
This is usually done by analysing the type and amount of the gases
that are present when abnormality occurs or during routine
maintenance. The characteristic of the transformer faults are described
as below [64]:
3.3.1 Arcing
Arcing is the most severe of all faults processes. Large amount of
hydrogen and acetylene are produced, with minor quantities of
methane and ethylene. Arcing occurs in high current and high
temperature conditions. Carbon dioxide and carbon monoxide may
also be formed if the fault involves cellulose. In some instances, the
oil may become carbonized [64].
3.3.2 Thermal heating/ Pyrolysis
Decomposition products include ethylene and methane, together with
smaller quantities of hydrogen and ethane. Traces of acetylene may be
formed if the fault is severe or involves electrical contacts [64].
3.3.3 Corona
Corona is a low-energy electrical fault. Low-energy electrical
discharges produce hydrogen and methane, with small quantities of
ethane and ethylene. Comparable amounts of carbon monoxide and
dioxide may result from discharge in cellulose [64].
3.3.4 Overheating Cellulose
Large quantities of carbon dioxide and carbon monoxide are evolved
from overheated cellulose. Hydrogen-based gases, such as methane
and ethylene, will be formed if the fault involved is an oil-
35
impregnated structure.
The dissolved gas analysis method involves sampling of the oil inside
the transformer at various locations. Then, chromatographic analysis
will be carried out on the oil sample to measure the concentration of
the dissolved gases .The extracted gases are then separated, identified
and quantitatively determined such that
reliable diagnosis can be obtained [65]. The extracted gases meant for
analysis purpose are Hydrogen (H2), Methane (CH4), Ethane (CH6),
Ethylene (C2H4), acetylene (C2H2), Carbon Monoxide (CO), Carbon
Dioxide (CO2), Nitrogen (N2) and Oxygen (O2). These fault gases can
be classified into 3 groups which are shown in Table 3 below.
Table 3: Fault gases group [17]
Group
GasesN2,O2CO,CO2CH4, H2,CH6,C2H4,C2H2
Non-fault gasesCarbon oxidesHydrocarbons and
Depending on the concentration of the dissolved gases, the condition of the
transformer can be determined. This is because each type of fault burns the oil in a
different way which correspondingly generate different pattern of gases. This makes
it possible for experts to identify the nature of the fault type based on the gas type
and its concentration. For example, arcing may cause high concentration of acetylene
dissolved in the oil as shown in Table 4 below.
Table 4: Relation between fault type and fault gases [65]
Fault Material Involved Fault Gases Present
Cellulose H2, CO, CO2
Thermal Heating/ Pyrolysis
Oil- Low Temp CH4, C2H6
Oil- High Temp C2H4, H2 (CH4, C2H6)
Cellulose - Low Temp CO2 (CO)
Cellulose - High Temp CO (CO2)
Arcing Oil/ Cellulose C2H2, H2 (CH4, C2H6, C2H4)
3.4 INTERPRETATION TECHNIQUES FOR DGA ANALYSIS
There are numbers of DGA interpretation techniques available that include the Key
Gas analysis, Rogers ratio method, IEC ratio method, Doernenburg Ratio method,
Duval method, [16]. All these methods are quite similar where different patterns and
36
concentration of gases are matched with the characteristic of fault types. Among
these methods, key Gas method and Rogers ratio method are the most popular [65].
key Gas method employs rules to diagnose abnormalities such as thermal, corona or
arcing problems while Roger's ratio method uses four different ratio codes to
determine the corresponding fault [65]. IEC method originated from Roger ratio
method but it uses only three gas ratios [65].
In this research, five most popular DGA diagnostic methods have been applied; these
techniques are Rogers ratio method, Key Gas method, IEC ratio method,
Doernenburg ratio method and Duval Triangle method to form a Fuzzy Logic Model
(FLM). The combination of these techniques ensures more accurate and more
reliable outcome. The most common diagnostics for DGA interpretation are given
below [13, 14, 25, 65-68]:
3.5 IEC METHOD
In dissolved gas analysis, the IEC codes have been used for several
decades and considerable experience accumulated throughout the
world to diagnose incipient faults in transformers as shown in Table 5
and 6 below. Early interpretations were concentrated on specific gas
components such as hydrogen and methane for the determination of
discharges in the oil. This simplistic approach was refined by a
number of researchers who investigated the ratios of certain gases to
establish more comprehensive diagnostic techniques.
37
Table 5: IEC ratio codes [65].
Ratio code Range Code
R1=C2H2/C2
H4
×<0.1 0
0.1≤×≤3 1
×>3 2
R2=CH4/H2
×<0.1 1
0.1≤×≤1 0
×>1 2
R3=C2H4/C2
H6
×<1 0
1≤×≤3 1
×>3 2
Table 6: Classification of fault based on IEC Ratio codes [65].
Ri1 Ri2 Ri3 Characteristic fault
0 0 0 Normal ageing
0 1 0 Partial discharge of low energy density
1 1 0 Partial discharge of high energy density
1--2 0 1--2 Discharge of low(continuous sparking )
1 0 2 Discharge of high energy(arcing)
0 0 1 Thermal fault <150°C
0 2 0 Thermal fault 150°c -300 °C
0 2 1 Thermal fault 300-700°C
0 2 2 Thermal fault>700°C
38
Case studies using IEC method
2000 samples were collected from various transformers of various life span and
operating conditions. Various interpretation techniques are used to analyse the 2000
oil DGA samples. 20 samples using IEC method are shown in Table 7 below.
Table 7: 20 samples for IEC method.
No. of H2 CH4 C2H2 C2H4 C2H6 CO CO2 RI1 RI2 RI3 CODES F IEC
samples
1 495 1775 2 2438 276 293 2999 0.00082 3.585859 0.278873 0 2 2 Thermal
2 80 619 0 2480 326 268 2952 0 7.7375 0.129241 0 2 2 Thermal
3 21 24 0 98 23 159 917 0 1.142857 0.875 0 0 2 Out of
4 231 3997 0 5584 1726 0 2194 0 17.30303 0.057793 0 2 2 Thermal
5 127 24 81 32 0 0 2024 2.53125 0.188976 5.291667 1 0 2 Arcing
6 2 7 0 0 0 0 132 #DIV/0! 3.5 0.285714 2 2 2 Out of
7 217 286 884 458 14 176 1544 1.930131 1.317972 0.758741 1 0 2 Arcing
8 54 0 0 4 0 106 1303 0 0 #DIV/0! 0 1 2 Out of
9 246 43 53 21 0 218 2069 2.52381 0.174797 5.72093 1 0 2 Arcing
10 9474 4066 12,997 6552 353 553 1156 1.983669 0.429175 2.330054 1 0 2 Arcing
11 507 1053 17 1440 297 22 2562 0.011806 2.076923 0.481481 0 0 2 Out of
12 416 695 0 867 74 200 14,316 0 1.670673 0.598561 0 0 2 Out of
13 47 12 0 8 0 115 1113 0 0.255319 3.916667 0 0 2 Out of
14 441 207 261 224 43 161 1123 1.165179 0.469388 2.130435 1 0 2 Arcing
15 18.9 46.9 0 61.54 6.9 371 4257 0 2.481481 0.402985 0 0 2 Out of
16 116.6 623 2.87 1683.5 416 317 3876 0.001705 5.343053 0.187159 0 2 2 Thermal
17 200 700 1 740 250 N N 0.001351 3.5 0.285714 0 2 1 Thermal
18 300 490 95 360 180 N N 0.263889 1.633333 0.612245 1 0 1 Arcing
19 56 61 31 32 75 N N 0.96875 1.089286 0.918033 1 0 0 Out of
20 33 26 0.2 5.3 6 N N 0.037736 0.787879 1.269231 0 0 0 Normal
3.5.1 Roger Ratio Method
Four ratios; CH4/H2, C2H6/CH4, C2H4/C2H6 and C2H2/C2H4 are used in the original
Roger’s ratio method diagnosis [29]. Two tables (8 and 9) are utilized in Roger’s
Ratio method; one shows the code of the ratio, the other presents the diagnoses. A
limited temperature range of the decomposition is only indicated with the ratio
C2H6/CH4.
39
Table 8: Roger ratio codes.
Ratio code Range code
RG1
CH4 /H2
x<0.1 5
0.1≤x≤1.0 0
1.0≤x≤3.0 1
x>3.0 2
RG2
C2H6 /CH4
x<1.0 0
x≥1.0 1
RG3
C2H4 /C2H6
x<1.0 0
1.0≤x≤3.0 1
x>3.0 2
RG4
C2H2 /C2H4
x<0.1 0
0.1≤x≤3.0 1
x>3.0 2
Table 9: Classification of fault based on Roger ratio codes
RG1 RG2 RG3 RG4 Diagnosis
0 0 0 0 Normal deterioration
5 0 0 0 Partial discharge
1--2 0 0 0 Slight overheating<150°C
1--2 1 0 0 Overheating 150-200°C
0 1 0 0 Overheating 200°c-300°C
0 0 1 0 General conductor overheating
1 0 1 0 Winding circulating currents
1 0 2 0 Core & tank circulating currents
0 0 0 1 Flashover without power follow
through
0 0 1--2 1--2 Arc with power follow through
0 0 2 2 continuous sparking to floating
5 0 0 1--2 Partial discharge with tracking
Case studies using Roger Ratio method
The collected 2000 samples were analysed using Roger’s ratio method; 20 samples
are shown in Table 10.
40
Table 10: 20 samples for Roger ratio method
No. of
samples H2 CH4 C2H2 C2H4 C2H6 CO CO2 RG1 RG2 RG3 RG4 CODES F Roger
1 495 1775 2 2438 276 293 2999 3.585859 0.155493 8.833333 0.00082 2020 out of
2 80 619 0 2480 326 268 2952 7.7375 0.526656 7.607362 0 2020 out of
3 21 24 0 98 23 159 917 1.142857 0.958333 4.26087 0 1020 Thermal
4 231 3997 0 5584 1726 0 2194 17.30303 0.431824 3.235226 0 2020 out of
5 127 24 81 32 0 0 2024 0.188976 0 #DIV/0! 2.53125 0 0 2 1 Arcing
6 2 7 0 0 0 0 132 3.5 0 #DIV/0! #DIV/0! 2022 out of
7 217 286 884 458 14 176 1544 1.317972 0.048951 32.71429 1.930131 1021 out of
8 54 0 0 4 0 106 1303 0 #DIV/0! #DIV/0! 0 5120 out of
9 246 43 53 21 0 218 2069 0.174797 0 #DIV/0! 2.52381 0 0 2 1 Arcing
10 9474 4066 12,997 6552 353 553 1156 0.429175 0.086818 18.56091 1.983669 0 0 2 1 Arcing
11 507 1053 17 1440 297 22 2562 2.076923 0.282051 4.848485 0.011806 1 0 2 0 Thermal
12 416 695 0 867 74 200 14,316 2 0 12 0 1 0 2 0 Thermal
13 47 12 0 8 0 115 1113 0.255319 0 #DIV/0! 0 0 0 2 0 out of
14 441 207 261 224 43 161 1123 0.469388 0.207729 5.209302 1.165179 0 0 2 1 Arcing
15 18.9 46.9 0 61.54 6.9 371 4257 2.481481 0.147122 8.918841 0 1 0 2 0 Thermal
16 116.6 623 2.87 1683.5 416 317 3876 5.343053 0.667737 4.046875 0.001705 2 0 2 0 out of
17 200 700 1 740 250 N N 3.5 0.357143 2.96 0.001351 2 0 1 0 out of
18 300 490 95 360 180 N N 1.633333 0.367347 2 0.263889 1 0 10 Thermal
19 56 61 31 32 75 N N 1.089286 1.229508 0.426667 0.96875 1 1 0 1 out of
20 33 26 0.2 5.3 6 N N 0.787879 0.230769 0.883333 0.037736 0 0 0 0 Normal
3.5.2 Key Gas Method
When a fault occurs, the temperature in the power transformer increases therefore
some fault gases are released from the insulating oil. Key Gas method is significantly
based on the quantity of the fault gases released from the insulating oil. Presence of
fault gases causes breakdown to the insulating oil chemical structure. In this method
the individual gases are used rather than the calculation of gas ratios for diagnosing
fault [65]. The Key Gas method employs a set of rules to diagnose abnormalities
such as Thermal, Corona or Arcing problems. It is considered as a conservative
diagnostic method and it can diagnose the condition of the transformer even though
when only a few gases are obtained from the oil sample. Different levels of
temperature will release different type of fault gases dissolved in the insulating oil.
For example, under slight overheating at about 130°C, some Methane and Hydrogen
gases are produced and as the temperature increases, Ethane is formed in higher
relative quantities with rising temperature between 350-400°C [65]. If the
temperature continues to rise up until 400°C or higher, Ethylene begins to form and
Acetylene will be released when the temperature reaches 700°C as shown in Table
11 below [17].
41
Table 11: Relation of fault gases and temperature [17]
Temperature range (C°) Fault gas
130- 150 Hydrogen (H2), Methane (CH4)
350- 400 Ethane (C2H6)
400- 600 Ethylene (C2H4)
> 700 Acetylene (C2H2)
The key gases are used to predict a specific problem are C2H2 (Acetylene) for arcing
H2 (Hydrogen) for corona in oil, C2H4 (Ethylene) for severe overheating C2H2
(Acetylene) for arcing, CH4 (Methane) for sparking, CO (Carbon Monoxide) for
overheating cellulose and C2H6 (Ethane) for local overheating. Key Gas method
classifies transformer health condition into 4 conditions as shown in Table 12 [19,
67]. Fig. 9 shows the principal gases for each fault [65].
Table 12: Dissolved Key Gas Concentration Limits in Parts Per Million (ppm) [67]
STATUS H2 CH4 C2H2 C2H4 C2H6 CO CO2 TDCG
Condition 1 100 120 35 50 65 350 2500 720
Condition 2 101-700 121-400 36-50 51-100 66-100 351-570 2500-4000 721-1920
Condition 3 701-1800 401-1000 51-80 101-200 101-150 571-1400 4001-10000 1921-4630
Condition 4 >1800 >1000 >80 >200 >150 >1400 >10000 >4630
Case studies using Key Gas method
The Key Gas method is used to analyse the collected 2000 DGA samples. 20 DGA
samples (analysed using Key Gas method) are given in Table 13.
42
Figure 9. Principal gases for each fault [65]
Table 13: 20 samples for Key Gas
No. of
samples H2 CH4 C2H2 C2H4 C2H6 CO CO2 F key Gas
1 495 1775 2 2438 276 293 2999 Thermal
2 80 619 0 2480 326 268 2952 Thermal
3 21 24 0 98 23 159 917 Thermal
4 231 3997 0 5584 1726 0 2194 Thermal
5 127 24 81 32 0 0 2024 Arcing
6 2 7 0 0 0 0 132 Normal
7 217 286 884 458 14 176 1544 Arcing
8 54 0 0 4 0 106 1303 Normal
9 246 43 53 21 0 218 2069 Arcing
10 9474 4066 12,997 6552 353 553 1156 Thermal
11 507 1053 17 1440 297 22 2562 Thermal
12 416 695 0 867 74 200 14,316 Thermal
13 47 12 0 8 0 115 1113 Normal
14 441 207 261 224 43 161 1123 Arcing
15 18.9 46.9 0 61.54 6.9 371 4257 Thermal
16 116.6 623 2.87 1683.5 416 317 3876 Thermal
17 200 700 1 740 250 N N Thermal
18 300 490 95 360 180 N N Corona
19 56 61 31 32 75 N N Thermal
20 33 26 0.2 5.3 6 N N Normal
43
3.5.3 Doernenburg Ratio Method
In order for the Doernenburg ratio method to be valid, at least one of the gas
concentrations of H2, CH4, and C2H4 must be at least twice the L1 level in given
Table 14 [65]. The 4 ratios used in Doernenburg method along with the associated
faults are shown in Table 15.
Table 12: L1 Limit Concentration for Doernenburg ratio method
KEY GAS
CONCENTRATIONS
L1
[µL/L (ppm)]
Hydrogen (H2) 100
Methane (CH4) 120
Acetylene (C2H2) 35
Ethylene (C2H4) 50
Ethane (C2H6) 65
Carbon Monoxide (CO) 350
Carbon dioxide (CO2) 2500
Table 13: Doernenburg ratio method
Suggested fault
Diagnosis
Ratio1(R1) Ratio2(R2) Ratio 3 (R3) Ratio 4(R4)
CH4/H2 C2H2/C2H4 C2H2/CH4 C2H6/C2H2
Oil Gas
Space Oil
Gas
Space Oil
Gas
Space Oil
Gas
Space
1.Thermal
decomposition >1 >0.1 <0.75 <1 <0.3 <0.1 >0.4 >0.2
2.PartialDischarge(low-
intensity PD) <0.1 <0.01
Not
Significant <0.3 <0.1 >0.4 >0.2
3.Arcing(high-intensity
PD)
>0.1
to<1
>0.01
To<0.1 >0.75 > 1 >0.3 >0.1 <0.4 <0.2
Case studies using Doernenburg Ratio method
The collected 2000 DGA samples were analysed using Doernenburg Ratio method of
which 20 samples are shown in Table 16.
44
Table 14: 20 samples for Doernenburg method.
No. of H2 CH4 C2H2 C2H4 C2H6 CO CO2 RD1 RD2 RD3 RD4 F Doern.
samples
1 495 1775 2 2438 276 293 2999 3.585859 0.00082 0.001127 138 Thermal
2 80 619 0 2480 326 268 2952 7.7375 0 0 #DIV/0! Thermal
3 21 24 0 98 23 159 917 1.142857 0 0 #DIV/0! Thermal
4 231 3997 0 5584 1726 0 2194 17.30303 0 0 #DIV/0! Thermal
5 127 24 81 32 0 0 2024 0.188976 2.53125 3.375 0 Arcing
6 2 7 0 0 0 0 132 3.5 #DIV/0! 0 #DIV/0! Thermal
7 217 286 884 458 14 176 1544 1.317972 1.930131 3.090909 0.015837 Thermal
8 54 0 0 4 0 106 1303 0 0 #DIV/0! #DIV/0! Normal
9 246 43 53 21 0 218 2069 0.174797 2.52381 1.232558 0 Arcing
10 9474 4066 12,997 6552 353 553 1156 0.429175 1.983669 3.196508 0.02716 Arcing
11 507 1053 17 1440 297 22 2562 2.076923 0.011806 0.016144 17.47059 Thermal
12 416 695 0 867 74 200 14,316 1.670673 0 0 #DIV/0! Thermal
13 47 12 0 8 0 115 1113 0.255319 0 0 #DIV/0! Normal
14 441 207 261 224 43 161 1123 0.469388 1.165179 1.26087 0.164751 Arcing
15 18.9 46.9 0 61.54 6.9 371 4257 2.481481 0 0 #DIV/0! Thermal
16 116.6 623 2.87 1683.5 416 317 3876 5.343053 0.001705 0.004607 144.9477 Thermal
17 200 700 1 740 250 N N 3.5 0.001351 0.001429 250 Thermal
18 300 490 95 360 180 N N 1.633333 0.263889 0.193878 1.894737 Arcing
19 56 61 31 32 75 N N 1.089286 0.96875 0.508197 2.419355 Thermal
20 33 26 0.2 5.3 6 N N 0.787879 0.037736 0.007692 30 Normal
3.5.4 Duval Triangle:
This graphical method was developed by Duval and has proven to be accurate and
dependable over many years and is now gaining popularity. Duval proposed the
triangle shown in Fig. 10 for fault diagnosis [30, 38, 39, 69]. A brief explanation for
using this triangle is given below.
Figure 10. Duval Triangle [38].
Legend
PD= Partial Discharge
T1= Thermal Fault Less than 300 C
0
T2= Thermal Fault between 300 C0and
700 C0
T3= Thermal Fault Greater than 700C0
D1= Low energy Discharge (Sparking)
D2= High Energy Discharge (Arcing)
DT= Mix of Thermal and Electrical
Faults
45
Table 15: Limit and generation Rate per month Limits [15]
1. First Determine whether the problem exists using Table 17. At least one
of the hydrocarbon gases or hydrogen (H2) must be in IEEE condition 3,
and increasing at generation rate (G2), before the problem is confirmed. If
there is sudden increase in H2 with only carbon monoxide (CO) and
carbon dioxide (CO2) and little or none of the hydrocarbon gases then use
(CO/CO2 ratio) to determine if the cellulose insulation is being degraded
by overheating; normal ratio for CO/CO2 is about 7 [15, 36].
2. Add the amount of the three gases or rates of increment in ppm.
3. Divide each individual gas or its rate of increment by the total number
obtained in 2.
4. Plot the percentage of each gas on the Duval Triangle, beginning on the
side indicated for that particular gas. Draw lines across the triangle for
each gas parallel to the hash marks shown on each side of the triangle. An
example is shown in Figure 11 [30-39].
Gas
L1
Limits
G1 Limits
(ppm per
month)
G2
(ppm per month)
H2
CH4
C2H2
C2H4
C2H6
CO
CO2
100
75
3
75
75
700
7000
10
8
3
8
8
70
700
50
38
3
38
38
350
3500
46
Figure 11. Duval Triangle Diagnostic Example of a reclamation transformer [36].
Case studies using Duval Triangle method
The 2000 samples were analysed using Duval triangle as explained above. 20
samples using Duval Triangle method are shown in Table below.
Table 18: 20 samples for Duval Triangle method
No. of
samples H2 CH4 C2H2 C2H4 C2H6 CO CO2 P1 P2 P3 ZONES F Duval
1 495 1775 2 2438 276 293 2999 42.11151 57.84104 0.04745 T3 Thermal
2 80 619 0 2480 326 268 2952 19.97419 80.02581 0 T3 Thermal
3 21 24 0 98 23 159 917 19.67213 80.32787 0 T3 Thermal
4 231 3997 0 5584 1726 0 2194 41.71798 58.28202 0 T3 Thermal
5 127 24 81 32 0 0 2024 17.51825 23.35766 59.12409 D2 Arcing
6 2 7 0 0 0 0 132 100 0 0 PD Corona
7 217 286 884 458 14 176 1544 17.56757 28.13268 54.29975 D2 Arcing
8 54 0 0 4 0 106 1303 0 100 0 T3 Thermal
9 246 43 53 21 0 218 2069 36.75214 17.94872 45.29915 D1 Arcing
10 9474 4066 12,997 6552 353 553 1156 17.21787 27.74508 55.03705 D2 Arcing
11 507 1053 17 1440 297 22 2562 41.95219 57.37052 0.677291 T3 Thermal
12 416 695 0 867 74 200 14,316 44 56 0 T3 Thermal
13 47 12 0 8 0 115 1113 60 40 0 T2 Thermal
14 441 207 261 224 43 161 1123 29.91329 32.36994 37.71676 D2 Arcing
15 18.9 46.9 0 61.54 6.9 371 4257 43.24972 56.75028 0 T3 Thermal
16 116.6 623 2.87 1683.5 416 317 3876 26.97705 72.89867 0.124276 T3 Thermal
17 200 700 1 740 250 N N 48.57738 51.35323 0.069396 T3 Thermal
18 300 490 95 360 180 N N 51.85185 38.09524 10.05291 DT Corona
19 56 61 31 32 75 N N 49.19355 25.80645 25 D2 Arcing
20 33 26 0.2 5.3 6 N N 82.53968 16.8254 0.634921 T1 Thermal
47
3.6 RESULTS DISCUSSIONS AND OBSERVATIONS
The 5 DGA interpretation techniques are used to analyse the collected 2000 DGA
samples as shown above. Table 19 shows the results of each technique when applied
on the same 20 samples. Based on this analysis, one can observe the following:
Key Gas method is very strict and cannot precisely identify the fault
type. However, it is the best method that can identify whether the oil
sample is faulty or not.
Duval Triangle method is accurate in identifying various faults
however, it relies on graphical analysis and it doesn’t have normal
zone, then it cannot be used to identify incipient faults.
Ratio-based methods lead to out of code results if the amount of gases
used in the ratios is not significant and DGA samples cannot be
diagnosed using these methods.
Various interpretation techniques do not necessarily lead to the same
conclusion for the same oil sample.
Table 19: Results of 5 DGA methods when applied for the same samples
No. of H2 CH4 C2H2 C2H4 C2H6 CO CO2 F Duval F key Gas F Doern. F IEC F Roger
samples
1 495 1775 2 2438 276 293 2999 Thermal Thermal Thermal Thermal out of
2 80 619 0 2480 326 268 2952 Thermal Thermal Thermal Thermal out of
3 21 24 0 98 23 159 917 Thermal Thermal Thermal Out of Thermal
4 231 3997 0 5584 1726 0 2194 Thermal Thermal Thermal Thermal out of
5 127 24 81 32 0 0 2024 Arcing Arcing Arcing Arcing Arcing
6 2 7 0 0 0 0 132 Corona Normal Thermal Out of out of
7 217 286 884 458 14 176 1544 Arcing Arcing Thermal Arcing out of
8 54 0 0 4 0 106 1303 Thermal Normal Normal Out of out of
9 246 43 53 21 0 218 2069 Arcing Arcing Arcing Arcing Arcing
10 9474 4066 12,997 6552 353 553 1156 Arcing Thermal Arcing Arcing Arcing
11 507 1053 17 1440 297 22 2562 Thermal Thermal Thermal Out of Thermal
12 416 695 0 867 74 200 14,316 Thermal Thermal Thermal Out of Thermal
13 47 12 0 8 0 115 1113 Thermal Normal Normal Out of out of
14 441 207 261 224 43 161 1123 Arcing Arcing Arcing Arcing Arcing
15 18.9 46.9 0 61.54 6.9 371 4257 Thermal Thermal Thermal Out of Thermal
16 116.6 623 2.87 1683.5 416 317 3876 Thermal Thermal Thermal Thermal out of
17 200 700 1 740 250 N N Thermal Thermal Thermal Thermal out of
18 300 490 95 360 180 N N Corona Corona Arcing Arcing Thermal
19 56 61 31 32 75 N N Arcing Thermal Thermal Out of out of
20 33 26 0.2 5.3 6 N N Thermal Normal Normal Normal Normal
To overcome the limitations mentioned above and to reduce the personal’s
48
dependency in DGA interpretation, a fuzzy logic model is developed for each
technique. Consistency and accuracy analyses are performed using the 2000 DGA
samples. Results of consistency analysis are then used to develop comprehensive
fuzzy logic model that incorporates all 5 mentioned methods in one model in such a
way the final decision will be based on all 5 methods as will be illustrated in
Chapters 4 and 5.
49
4 CHAPTER 4
FUZZY LOGIC MODEL
4.1 INTRODUCTION
For the purpose of this research, fuzzy logic technique had been used in developing
fuzzy fault diagnostic system for power transformers. Fuzzy logic had been applied
in various fields such as control system, decision support, fault diagnostic, image
processing, and data analysis[64].The fuzzy logic theory was applied in solving
nonlinear control problems heuristically and modularly along linguistic lines [76,
77]. The advantages of fuzzy logic are that it exhibits the nature of human thinking
and makes decision or judgement using linguistic interpretation. Furthermore, the
control rules, regulations and methods based on the perception, experience and
suggestion of a human expert were encoded in a meaningful way to avoid
mathematical modelling problems [64].
In this chapter, the concept of fuzzy logic theory, methodology of fuzzy control and
decision will be presented. This is followed by, the application of the fuzzy logic
technique to five diagnostic methods namely Fuzzy Rogers Ratio, Fuzzy Key Gas,
Fuzzy IEC method, Fuzzy Duval, Fuzzy Doernenburg, and the design methodology
of the fuzzy diagnostic system for each of the aforementioned method.
It is important to observe that there is an intimate connection between Fuzziness and
Complexity [64]. Fuzzy logic allows intermediate value to be defined between
conventional evaluations like yes/no, and black/white. A fuzzy set allows for the
degree of membership of an item in a set to be any real number between 0 and 1. The
most powerful aspect of fuzzy set is the ability to deal with linguistic quantifiers or
“hedges” (dense). The examples of hedges are “more or less”, “very”, “not very” and
“slightly”. This allows human observations, expressions and expertise to be closely
modelled. Since then, fuzzy logic had been established as a useful alternative
approach for reasoning with imprecision and uncertainty [64].
4.2 FUZZY LOGIC APPLICATIONS
Fuzzy Logic has been used in solving problems in various domains. These include
process control, pattern recognition and classification, management and decision-
50
making, operations research and economics [64, 67]. Fuzzy logic acts as a profitable
tool for controlling of subway system and complex industrial processes, as well as
household, entertainment electronics and diagnostic systems [64]. Other applications
that are using fuzzy logic theory are information retrieval system, decisions support
systems, data analysis, fault diagnostic systems, voice and handwritten language
recognition systems and expert system [64].
Generally, the application of the fuzzy logic technique is appropriate:
For very complex processes, when there is no simple mathematical
model
For highly nonlinear processes
If the processing of (linguistically formulated) expert knowledge is to
be performed
4.3 THE ADVANTAGES OF FUZZY LOGIC
Fuzzy logic technology has emerged as a viable approach in control
engineering as well as decision support system. It offers many
advantages, which distinctly made it favourable to solve many
problems. Below are the 3 significant advantages:
4.3.1 Solution to nonlinear problems
Fuzzy logic is the answer for the problem regarding the unsolved non-
linear and complex problems. Fuzzy logic allows heuristic decisions-
making strategies to be formulated by natural language rules rather
than mathematical models. Thus, complex information can be
represented by simplified rules.
4.3.2 Ability to handle linguistic variables
In many applications, linguistic labels are used to provide meaningful
interpretations of the problems. For example, in decision support
systems or fault diagnostic systems, knowledge or experiences of the
experts is required to be coded into machines. By using fuzzy logic,
the expertise or knowledge is extracted from the experts, which is of
non-crisp nature can be easily modelled. Fuzzy decision system is
more reliable due to the absence of human emotional problems such
51
as bias, boredom and annoyance.
4.3.3 Rule reduction in fuzzy rule base
In a conventional expert system, a huge number of rules are needed to
describe the input-output relation. The number of rules can be
expressed as mn, where n denotes the number of the system variables
and m denotes the number of predicates in the antecedent part of each
rule. The large number of rules will degrade the system performance
in terms of processing speed and storage. However, the number of
rules in a fuzzy logic rule base can be greatly reduced without
degrading the performance. A 10:1 rule reduction can be expected in a
fuzzy rule base as compared to a conventional rule base [64, 70].
4.4 THE DISADVANTAGES OF FUZZY LOGIC
Fuzzy logic technology has been proven to be more effective in solving various kinds
of complex or imprecise problems. However, there are some limitations of fuzzy
logic, which are unavoidable that can be summarized as follows:
4.4.1 Highly dependent on domain expert’s knowledge
The use of the fuzzy logic technique concept is to translate the expert knowledge into
a collection of machine understandable rules. Unlike other artificial intelligent
techniques, such as neural networks and genetic algorithms, problem is solved via
training process. A well-defined knowledge base is needed in fuzzy logic to solve
any kind of problems. Hence the knowledge extraction process is crucial as the
whole fuzzy system is dependent on the domain expert knowledge. If the domain
experts provide wrong information, then the system may not be functioning well as
required. Thus, it is important to acquire correct knowledge for the correct experts.
4.4.2 Lack of information
Fuzzy control can be applied in many processes if there is enough information or
relevant knowledge about the process and its control strategies. Solving a totally
unknown or impossible job that even human experts cannot accomplish is difficult to
be accomplished using fuzzy logic technique.
4.4.3 Insufficient design standard or methodology
52
Fuzzy logic has been applied in various applications by number of researches around
the world. However, most of the researches use their own ways to design their
applications. They usually use heuristic or trial and error approach in selecting the
types of membership functions, inference engine and defuzzification methods. This
approach is time-consuming as the number of the fuzzy partitions and mapping of the
membership functions are the important factors that might affect the performance of
the result. Thus, a standard fuzzy system design guideline or systematic design
methodology is needed in order to obtain satisfactory results for fuzzy systems and
reduce the development time constrains [64, 70].
4.5 FUZZY LOGIC CONTROL SYSTEM
Fuzzy logic control system mainly consists of four major elements, which are
fuzzification unit, a fuzzy inference engine, fuzzy knowledge base and a
defuzzification unit. The typical structure for the fuzzy logic control system is shown
in Figure 12. The input values are normally in the crisp value, thus, the fuzzification
and defuzzification operations are needed to map these values into fuzzy values used
internally by the fuzzy logic control system or to defuzzify it into a crisp value. The
output from the defuzzification unit can be an action for controlling certain machine
or it can be a decision based on the knowledge of the decision maker in fuzzy
decision support system.
The fuzzy logic control system involves three successive processes, namely:
fuzzification, fuzzy inference, and defuzzification. Fuzzification converts a crisp gas
ratio into a fuzzy input membership. A chosen fuzzy inference system (FIS) is
responsible for drawing conclusion from the knowledge-based fuzzy rule set of if-
then linguistic statement. Defuzzification then converts the fuzzy output values back
into crisp output actions [14].
53
Figure 12. Basic structure of Fuzzy logic control system [17].
4.5.1 Fuzzification
Fuzzification is the process of mapping from observed inputs to fuzzy sets. In
process control, the observed data is usually in crisp set and fuzzification is required
to map the observed range of crisp inputs to corresponding fuzzy values for the
system input variables. The mapped data are further converted into suitable linguistic
terms as labels of the fuzzy set defined for system input variables. When the variable
is classified with a membership function, the expected output is the degree of
membership. This process of classifying a variable using membership function and
degree of membership is called fuzzification.
4.5.2 Fuzzy Knowledge Base
A fuzzy knowledge base usually consists of a group of fuzzy rules, which is
extracted from experts. There are no formal standards to follow in constructing the
fuzzy rules. In most engineering control application, the fuzzy rules are expressed as
“IF-THEN” style. For example, “IF x is A THEN y is B. the rule base consists of a
collection of fuzzy control rules based on the control objective and control policy.
The fuzzy control rules are able to infer a properly control action for any input in the
universe of discourse [64].
54
4.5.3 Fuzzy inference system (FIS)
Although the fuzzy rules appear strictly defined, borderline cases with gas ratios on
or near the line between code 0, 1 or 2 allows FIS to (1) interpret membership of
these rules flexibly, and (2) classify these cases under two different fault types with
individual probability of occurrence attached to each type. There are various types in
which the observed input values can be used to identify the most appropriate rules.
The most well-known types are Mamdani’s Max-Min implication method and
Larsen’s Max-product implication method.
4.5.4 Defuzzification
Defuzzification is used to convert the fuzzy linguistic variable to real variable. It is
the process of mapping from a space of inferred fuzzy control actions to a space of
non-Fuzzy (crisp) control actions. A defuzzification strategy is aimed at producing a
non-fuzzy control action that best represents the possibility distribution of the
inferred fuzzy control action [64].
55
Figure 13. Five fuzzy inference systems (2) analyse gas values (1) regarding to the defect condition (3a) of the power transformer, analytical results are defect condition and reliability for all leaves (3c) of the defect condition tree (3b) [17].
ii MM
H2
C2H2
CH4
CO2
CO
C2H6
C2H4
Measure-
ments Cl
os
in
g
th
e
la
ck
of
de
fin
iti
on
s
Duval’s
Triangle
l
Key Gas
Doernen
burg
Ratio
IEC Ratio
Roger
Ratio
M
od
elli
ng
wit
h
fuz
zy
inf
ere
nc
e
sys
te
ms
Sta
nd
ard
iza
tio
n
of
the
rul
e
bas
e
Op
tim
izat
ion
by
trai
nin
g
Co
mp
act
ion
wit
hin
a
def
ect
co
ndi
tio
n
tre
e
Interpretation Components
Defect
Condition.
Highly
Reliable.
Highly
Detailed
1 2 3a
Unknown
Defect
Healthy
thermal Leaky tap
Changer tank
Thermal and
electrical
electrical
Partial
discharge
Circulating
currents discharge
Low
energy
High
energy
With
tracking
Without
tracking
High
energy
Low
energy
In core
And tank
In
winding
With
Continuous
Sparking to
Floating potential
Without
Continuous
Sparking to
Floating potential
Root 3b
>1000°C 700°C-
1000°C
General
Conductor
overheating
<300°C 300°C-
700°C
thermal
Cellulose
degradation
<150°C
Partial
discharge
150°C-
200°C
Thermal and
electrical
discharge
150°C-200°C
Condition: c=
Reliability: R=
3c
56
4.6 INTRODUCTION TO DGA FUZZY DIAGNOSTIC SYSTEM
In the fault detection process, it is hard to determine the relationship between the
phenomena and the reasons for the transformer faults. The faults often show some
form of vagueness or fuzziness. For example, according to Key Gas standard, high
concentration of gas acetylene (C2H2) is related to the fault of arcing and the gas
C2H2 concentration of 35 ppm or less is considered normal. However, sometimes at
35 ppm, the transformer may indicate arcing condition. Due to this difficulty, fuzzy
set theory can be utilized to deal with these uncertainties. Through fuzzy set theory,
the membership grade function can translate uncertain or qualitative information into
quantitative data. Fuzzy logic is known for its capability in handling linguistic
variables. Linguistic labels are used to provide meaningful interpretations of the
problems at hand. Thus, fuzzy logic is popular in many applications; including
problems of complex system diagnostic.
4.7 THE DESIGN METHODOLOGY OF FUZZY DIAGNOSTIC SYSTEM
To implement a reliable fuzzy diagnostic system, there is a general procedure in
designing a fuzzy system. This procedure is shown in Figure 13. The detail of this
procedure is described below [64, 71]:
4.7.1 Identification of the fuzzy input and output variables
Before applying fuzzy logic, it is important to understand the whole system process
and the objective of applying it. Then, a fuzzy system is designed by determining the
fuzzy input and output variables that are required to construct the fuzzy logic model
as shown in Figure 14[19,66].
Figure 14. Fuzzy logic model flow chart [19].
57
Also, in this step the number of fuzzy partitions of the input-output linguistic
variables is to be determined. Each of the fuzzy variables needs to be quantified into
smaller subsets appropriately. The process must be conducted carefully because the
number of fuzzy partitions may affect the performance of the control system.
However, the more partitions do not necessary mean better control performance. An
optimum number of partitions will make the system more efficient.
4.8 ASSIGNMENT OF MEMBERSHIP FUNCTIONS
The usage of membership functions is based on the system variables. Thus, the
appropriate membership functions are needed for the input and output fuzzy
variables. The most commonly used membership functions are the triangular, L-
function, T-function and Trapezoidal. The rule base consists of a collection of fuzzy
control rules based on the control objective and control policy.
4.8.1 Selection of fuzzy compositional operator (inference engine)
There are various ways in which the observed inputs values can be used to identify
the most appropriate rules to infer a suitable fuzzy control action. However, the
Mamdani’s Max-Min method is the most commonly used method and is used in this
project. Defuzzification is used to convert the fuzzy linguistic variable to variable as
shown in Figure 15. For fuzzy diagnostic system or decision support system, Max-
membership defuzzification method is chosen where the element that has the
maximum membership is chosen.
Figure 15. Steps for constructing a fuzzy logic system [17]
58
4.9 FUZZY LOGIC MODELS FOR DGA INTERPRETATION
TECHNIQUES
Fuzzy logic models are developed to aid in standardizing the overall decision of
various DGA interpretation techniques. Each fuzzy logic model is developed in
accordance to fuzzy inference flow chart shown in Figure 14. Input variables to the
model are the 7-key gases in parts per million (ppm). The output of each model is
divided into 5 sets of membership functions comprising all fault conditions that
operating transformers may exhibit along with a membership function for normal
condition (F5) as summarized in Table 20 then to develop the fuzzy logic model by
incorporate the thermal fault at low temperature (cellulose/oil) (F1) with overheated
fault (cellulose and or oil) (F2) to be one thermal fault (F1) and (F4) will represent the
Normal condition, a membership function (F5) is added to represent the “out of code”
condition that ratio methods may lead to for some DGA samples as shown in Table
21.
Table 16: Fault types by considering two thermal faults
Fault Type Fault Type
code
Thermal fault at low
temperature F1
Overheating and sparking F2
Partial Discharge and corona F3
Arcing F4
Normal F5
Out of code F6
Table 17: Fault types by considering one thermal fault.
Fault Type Fault Type
code
Thermal fault F1
Partial Discharge and corona F2
Arcing F3
Normal F4
Out of code F5
Set of fuzzy logic rules in the form of (IF-AND-THEN) statements relating the input
variables to the output were developed based on transformer’s diagnostic and test
data interpretation techniques. Each fuzzy model is built using the graphical user
59
interface tool provided by MATLAB where each input is fuzzified into various sets
of membership functions. Centre-of-gravity which is widely used in fuzzy models
was used for defuzzification method where the desired output z0 is calculated as
below:
Z0 = (1)
Where is the membership function of output.
The output membership functions for all models are shown in Figure 16&17.
Output Criticality
Figure 16. Fuzzy logic models output membership functions (when two thermal faults F1 and F2
are considered).
Output Criticality
Figure 17.Fuzzy logic models output membership functions for one thermal fault.
The cellulosic thermal decomposition produces CO and CO2 at lower temperature
60
than that for oil decomposition and traceable amount of these gases can be found at
normal operating condition. Oil thermal decomposition starts at higher temperature
and at about 350°C production of C2H4 begins. At about 450 °C, H2 production
exceeds all other gases causing low-intensity discharges such as partial discharge and
very low level intermittent arcing. At about 700 °C, more C2H2 is produced causing
high intensity arcing or continuing discharge proportion as shown in Table 22 [23].
Table 18: Fault types for one thermal fault
Method F1 Thermal fault
(cellulose, oil)
F2 Electrical fault
(corona)
F3 Electrical
fault (Arcing)
Roger Thermal fault 150 °C- 700 °C Low energy
electrical discharge
High energy
discharge
IEC
Thermal fault 150 °C- 700 °C
Low energy
electrical discharge
High energy
discharge
Doernen.
Thermal decomposition
Low energy
electrical discharge
High energy
discharge
Duval Thermal fault 150 °C- 700 °C
Low energy
electrical discharge
High energy
discharge
K.Gas Overheated cellulose/oil Low energy
electrical discharge
High energy
discharge
Table 20 is established based on Figure 16 which shows the various types of faults
and the significant gases produced by each fault.
Figure 18.Type of faults and generated gases [19].
The fuzzy models for IEC, Roger, Doerenburg, Key Gas and Duval triangle are
described below.
Faults
Electrical Thermal
Arcing
H2, C2H2
Partial
discharge
H2, CH4
Paper
CO, CO2
Oil
C2H4, C2H6
61
4.9.1 Fuzzy Logic for Roger ratio method
A fuzzy system for Roger’s Ratio method consists of 4 ratio codes as inputs. The 4
ratios are classified as either Low (Lo), Medium (Med), High (Hi) or Very High
(Vhi) according to membership intervals as defined below:
RG1= {Lo, Med, Hi, Vhi}
RG2= {Lo, Hi}
RG3= {Lo, Med, Hi}
RG4= {Lo, Med, Hi}
Fuzzy inference consists of two components which are the antecedent (IF part) and
the consequent (THEN part). Here, the fuzzy inference rules are based on the fault
interpretations given in Table 20. The following are some examples of the fuzzy
rules:
Rules 1: IF RG1= Med AND RG2= Lo AND RG3= Lo AND RG4= Lo THEN Faults
(1)
Rules 3: IF RG1= Hi, AND RG2= Lo AND RG3= Lo AND RG4= Lo THEN Faults (3)
Rules 4: IF RG1= Vhi AND RG2= Lo AND RG3= Lo AND RG4= Lo THEN Faults
(3).
The output of the fuzzy inference can be obtained using the Mamdani’s Max-Min
composition technique. Here the logical “AND” is replaced with the minimization
operator and the logical “OR” is replaced with the maximization operator [72-76].
The developed set of fuzzy rules relates the input and the output variables for Roger
ratio method is shown in the 3D surface graphs (Figure 19).
62
Figure 19. Surface graphs for the Roger ratio method
The model is tested with inputs, RG1 (2.5), RG2 (2.5), RG3 (2.5), RG4 (2.73) as
detected in one of the transformer oil samples results using DGA. The fuzzy logic
model numerical output is 11 as shown in Figure 20. This is corresponding to F6 (out
of code fault).
0 5 0 5 0 5 0 5 0 1
Figure 20. Roger ratio fuzzy rules.
63
4.9.2 Fuzzy logic for IEC method
This system has 3 ratios as inputs. The 3 ratios are simplified and classified as either
Low (Lo), Medium (Med) or High (Hi) according to membership intervals as defined
below:
Ri1= C2H2/ C2H4 = {Lo, Med, Hi}
Ri2= CH4/ H2 = {Lo, Med, Hi}
Ri3= C2H4/ C2H6 = {Lo, Med, Hi}
The types of fuzzy membership functions used are the same as in the previous
method.
The fuzzy rules relate the input and the output variables for IEC ratio method is
shown in the 3D surface graphs (Figure 21). The model is tested with inputs,
C2H2/C2H4 (2.5), CH4/H2 (2.5) and C2H4/C2H6 (2.5) as detected in one of the
transformer oil samples results using DGA. The fuzzy logic model numerical output
is 7.07 as shown in Figure 22. This is corresponding to F3 (Arcing fault) in Figure
16.
Figure 21. Surface graphs for IEC ratio method
64
Figure 22. IEC fuzzy rules
4.9.3 Fuzzy logic for Doernenburg
In this method, the concentration of the hydrocarbon gases was firstly checked
against the limits L1. If all fault gases are within the limits then the diagnosis would
be “normal”. Otherwise, a fault condition is indicated, and the 4 ratios are then
calculated based on the gas concentrations and used to classify the fault. The input
ratios are classified as Low, Medium or High according to membership intervals as
defined in the following:
RD1 = CH4 / H2
RD2 = C2H2 / C2H4
RD3 = C2H2 / CH4
RD4 = C2H6 / C2H2
R11 = {Lo, Med, Hi} R12= {Lo, Med, Hi}
R21 = {Lo, Hi} R22 = {Lo, Hi}
R31 = {Lo, Hi} R32 = {Lo, Hi}
R41 = {Lo, Hi} R42 = {Lo, Hi}
The fuzzifying membership functions are the same types used in the last methods.
65
The fuzzy inference rules are based on the fault interpretation. The model is tested
with inputs, CH4/H2 (1), C2H2/C2H4 (1), C2H2/CH4 (0.5), C2H6 /C2H2 (0.5) as
detected in one of the transformer oil samples results using DGA. The fuzzy logic
model numerical output is 6.05 as shown in Figure 23. Figure 24 shown the surface
graphs for Doernenburg Ratio method:
Figure 23. Doernenburg fuzzy rules.
Figure 24. Surface graphs for Doernenburg method
66
4.9.4 Fuzzy logic for Duval method
The Duval Triangle method fuzzy system consists of 3 fault gas percentages as the
inputs and the 7 regions in the Duval triangle as the outputs. Similar to the approach
used in the previous method, at least one of the gas values must exceed a specified
limit (L1) in order to be considered as having a fault. The types of fuzzy membership
functions used for this method are the same as for previous methods. Based on
membership functions of the inputs, this system can have up to 294 rules (7x6x7).
However as the total % CH4 + % C2H4 + % C2H2 must be 100, some rules are not
used. The rule components and the output inferences of this method were derived
using the same technique as the previous methods. This model is tested with inputs,
P1 (34.3), P2 (80.1) and P3 (50) as detected in one of the transformer oil samples
results using DGA. The fuzzy logic model numerical output is 5.04 as shown in
Figure 25. This is corresponding to F2 (corona fault) in Figure 16. Figure 26 has
shown the surface graphs for Duval Triangle method.
Figure 25. Duval fuzzy rules.
67
Figure 26. Surface graph (P1-P2) for Duval Triangle method
4.9.5 Fuzzy logic for Key Gas method
This method is based on the individual values of the fault gases. Here, all five key
gases were used as inputs and the output is the 5 fault types. The membership of the
fuzzy set “Lo”, “ Med” or “Hi” was used for each fault gas. The model is tested with
inputs, H2 (75), CH4 (75), C2H6 (50), C2H4 (50), C2H2 (50), CO (200), CO2 (1500) as
detected in one of the transformer oil samples results using DGA. The fuzzy logic
model numerical output is (5) as shown in Figure 27. The surface graph of key gas
method is shown in Fig. 28.
The collected 2000 DGA samples were re-analysed using the developed fuzzy logic
models of the 5 DGA interpretation techniques; Table 23 shows 20 samples results
using developed fuzzy logic models. Results were found typical to those obtained
using conventional manual method shown in the previous chapter (Table 19). Fuzzy
logic models improved the results obtained from ratio-based methods (Roger, IEC
and Dorenenburg) as it reduced the number of out-of codes cases. In the next
chapter, developed fuzzy logic models are used to perform consistency and accuracy
analysis of the 5 DGA interpretation techniques. Results of these analyses are then
used to establish a comprehensive fuzzy logic model that incorporates the key
features of the 5 DGA interpretation techniques.
68
Figure 27. Key gas fuzzy rules
Figure 28. Surface graphs (H2-CH4) for Key Gas method
69
Table 19: 20 samples results using Fuzzy logic model
No. of H2 CH4 C2H2 C2H4 C2H6 CO CO2 F Duval F key Gas F Doern. F IEC F Roger
samples by model by model by model by model by model
1 495 1775 2 2438 276 293 2999 Thermal Thermal Thermal Thermal out of
2 80 619 0 2480 326 268 2952 Thermal Thermal Thermal Thermal out of
3 21 24 0 98 23 159 917 Thermal Thermal Thermal Out of Thermal
4 231 3997 0 5584 1726 0 2194 Thermal Thermal Thermal Thermal out of
5 127 24 81 32 0 0 2024 Arcing Arcing Arcing Arcing Arcing
6 2 7 0 0 0 0 132 Corona Normal Thermal Out of out of
7 217 286 884 458 14 176 1544 Arcing Arcing Thermal Arcing out of
8 54 0 0 4 0 106 1303 Thermal Normal Normal Out of out of
9 246 43 53 21 0 218 2069 Arcing Arcing Arcing Arcing Arcing
10 9474 4066 12,997 6552 353 553 1156 Arcing Thermal Arcing Arcing Arcing
11 507 1053 17 1440 297 22 2562 Thermal Thermal Thermal Out of Thermal
12 416 695 0 867 74 200 14,316 Thermal Thermal Thermal Out of Thermal
13 47 12 0 8 0 115 1113 Thermal Normal Normal Out of out of
14 441 207 261 224 43 161 1123 Arcing Arcing Arcing Arcing Arcing
15 18.9 46.9 0 61.54 6.9 371 4257 Thermal Thermal Thermal Out of Thermal
16 116.6 623 2.87 1683.5 416 317 3876 Thermal Thermal Thermal Thermal out of
17 200 700 1 740 250 N N Thermal Thermal Thermal Thermal out of
18 300 490 95 360 180 N N Corona Corona Arcing Arcing Thermal
19 56 61 31 32 75 N N Arcing Thermal Thermal Out of out of
20 33 26 0.2 5.3 6 N N Thermal Normal Normal Normal Normal
70
5 CHAPTER 5
CONSISTENCY, ACCURACY ANALYSES AND
PROPOSED FUZZY LOGIC MODEL
5.1 INTRODUCTION
The DGA methodology has gained worldwide acceptance during the past 20 years as
a diagnostic tool for fault detecting incipient fault in oil-filled power transformers.
This acceptance reflects a unique ability of the DGA to detect faults at the earlier
possible stage and to distinguish different kinds of faults that occur inside the
transformer such as arcing, heating and partial discharge. However, the DGA
diagnostic methods that were invented by different experts from all around the world
have their own unique advantages in fault diagnosis of power transformer. All the
diagnostic methods discussed in the previous chapters have their own unique features
and advantages that contributed to the transformer fault detection. In order to build a
reliable fault diagnostic system, all the 5 DGA diagnostic methods discussed above
will be utilized to form the combined intelligent fault finding model. Combining all
the 5 DGA methods to construct the fuzzy logic model improves the reliability of the
fault diagnostic system [77].
The accuracy and consistencies of the DGA interpretation methods are investigated
in this chapter. DGA interpretation methods are compared together to get the best
optimized method that can predict the condition of power transformers. The data
consists of 2000 samples from different transformers located in Western Australia
and outside Australia. The transformer ages range from 5 to 40 years. These samples
cover wide range of transformer rating from 6 to 490 MVA. For comparing the
accuracy and consistency, the testing method should be similar for each DGA
interpretation technique. For comparison, each diagnosis method is categorized
according to the faults shown in Table 24. In this table F1 represents thermal fault,
F2-F3 (Electrical faults; Corona and arcing), F4 (Normal) when there is no fault.
71
5.2 CONSISTENCY ANALYSIS
Table 20: Fault Types Identified by various DGA [77].
Meth
od
F1
Thermal fault (Cellulose, Oil)
F2
Electrical fault (Corona)
F3
Electrical fault (Arcing)
Ro
ger
-Thermal fault 150ºC-700°C -Low energy electrical discharge - High energy electrical
discharge
IEC
-Thermal fault 150ºC-700°C - Low energy electrical discharge - High energy electrical
discharge
Do
ernen
.
-Thermal decomposition - Low energy electrical discharge
- High energy electrical
discharge
Du
va
l
-Thermal fault 150ºC-700°C - Low energy electrical discharge - High energy electrical
discharge
K. g
as
-Over heated cellulose/ oil
- Low energy electrical discharge
- High energy electrical
discharge
To get consistency of individual methods, each method has been tested on the data
samples of 2000 cases provided by Western Power as shown in Table 25. All the five
methods have been used to diagnose the fault using their relative ratio table and
charts. The objective of this analysis is to find out which method is more efficient in
predicting the faults [80-91].
The two formulas used to calculate the consistency in percentage are [65]:
Sfn = (2)
Cfn = (3)
Where fn= fault type codes (n= 1, 2, 3, 4, 5), Sfn = Successful prediction, Cfn =
Consistency.
The Consistency of all methods against the 3 fault types is shown in Figure 29
72
Table 21: Consistency table [65]
TYPES O F TYPE O F FAUL
METHO DS
F1 54 54 37 37 69% 69%
F2 89 89 30 30 34% 34%
F3 256 256 246 246 96% 96%
F4 12 12 10 10 83% 83%
F5 10 10 4 4 40% 40%
F6 1579 1579 525 525 33% 33%
F1 86 86 37 37 43% 43%
F2 88 88 50 50 57% 57%
F3 1114 1114 800 800 72% 72%
F4 48 48 24 24 50% 50%
F5 42 42 25 25 60% 60%
F6 622 622 525 525 84% 84%
F1 65 65 25 25 38% 38%
F2 191 191 90 90 47% 47%
F3 168 168 124 124 74% 74%
F4 63 63 40 40 63% 63%
F5 1513 1513 1513 1513 100% 100%
F6
F1 95 95 95 95 100% 100%
F2 143 143 143 143 100% 100%
F3 166 166 162 162 98% 98%
F4 83 83 50 50 60% 60%
F5 1513 1513 1513 1513 100% 100%
F6 0 0 0
F1 57 57 43 43 75% 75%
F2 351 351 244 244 70% 70%
F3 56 56 56 56 100% 100%
F4 23 23 21 21 91% 91%
F5 1513 1513 1513 1513 100% 100%
F6 0 0 0 0 0% 0%
F
NO .of prediction [P]
FSWFFSWFFSWFFSWF
CO NSISTENCY [C]%SUCCESSFUL [S]NO .of correct predict. R
87%87%Key Gas
65% 65%Doernenburg
92%92% Duval
59%59%RO GER'S
61%61%IEC
Figure 29. Consistency of all techniques for various faults.
Fig. 30 shows that consistency of ratio-based methods is less for thermal fault while
73
it is increasing for electrical fault. The overall consistency for Duval and Key Gas is
higher than ratio-based methods.
5.3 ACCURACY ANALYSIS
The accuracy of each method is divided into two groups (Ap) accuracy of prediction.
[64][80-91].
Ap = (4)
Where,
Tr is correct predication
Tp is number of predication
And the accuracy (At) is considered based on the total number of cases:
At = (5)
Where, Tc is the number total cases [65].
Table 26 shows the results of this analysis.
Table 22: Accuracy of DGA methods
TYPES OF
METHODS
WF FS WF FS WF FS WF FS WF FS WF FS WF FS
Incorrect prediction
Tw
Correct prediction
TR
Accuracy predictio
Ap
ACCURACY total
AT
90% 90%
NO. of prediction
Tp
NO.of no prediction
TNP
TOTAL CASES
TC
94%
98% 98% 98%
2000 2000 0 0 2000 2000 1792 90%1792 208 208 90%
2000 1877 1877 123 123Key Gas 2000 2000 00 2000
68% 68% 45% 45%1378 1378 936 442936 442
94% 94% 94%
2000 2000
Doernenburg
Duval 1963 37 98%372000 2000 0 0 1963
ROGER'S
IEC 2000 2000 622
20002000
622
421 3274211579 1579 16%16%78%78%9494327
It is observed that Doernenburg and Key Gas methods have been most successful in
74
diagnosing electrical fault details i.e. partial discharge of low energy and high energy
or continuous sparking. As seen from the table above there is a big difference
between the accuracy based on the predicted cases and total cases for IEC ratio and
Roger ratio. Clearly, this huge difference is because of the high number of cases with
no prediction. Therefore, the accuracy decreased significantly to less than 45% for
these methods. This decrease is due to using specific codes in the detection.
The overall results of Consistency and Accuracy for all methods are shown in Figure
30.
Figure 30. Overall Consistency and Accuracy comparison of all methods.
Fig. 30 reveals that Duval triangle is the most accurate and consistent method
followed by the Key Gas method. Due to the out of code cases that ratio-based
methods may lead to, based ratio methods have less accuracy and consistency.
5.4 PROPOSED APPROACH TO STANDARDIZE DGA
INTERPRETATION
The proposed approach is based on the incorporation of different DGA techniques
into one prototype software model as shown in the flow chart of Figure 31 and the
75
Simulink model shown in Fig. 32. In this Figure, Key Gas method is firstly used to
determine the health condition of the transformer oil sample based on its DGA
results. If the Key Gas method results in normal condition, the model reports normal
condition and no further analysis will be performed. However, if the key gas method
results in abnormal condition, the oil sample will be further analysed using Duval
triangle and ratio methods (IEC, Roger and Doernenburg) to accurately identify the
fault type. Each individual method is used to identify the fault type and the overall
decision (D) is calculated based on the consistency level of each method according to
the following equation:
D = (6)
where Di is the decision of each individual method weighted by its consistency level
Ai. In case any of the ratio methods provides a ratio that does not fit into the
diagnostic codes, the decision value corresponding to this method is set to zero.
Normal condition is only specified by the Key Gas method while in case of faulty
condition, the fault is specified by all methods. To implement the flow chart in
Figure 31, the individual fuzzy logic models for various DGA interpretation
techniques are integrated in one fuzzy model as shown in Figure 32. The inputs to
the overall model are the 7-key gases and the output represents a numerical number
that is corresponding to the failure rank of the transformer. The individual decisions
for all 5 methods are weighted using the consistency level of each method and are
integrated together according to (6) to provide one overall decision (D). The model is
tested for the DGA data shown in Figure 33, which shows that both Roger and IEC
ratio methods provide a value greater than 8 which is corresponding to F5 in Figure
16 (out of code) and hence, their contribution to the overall decision is eliminated.
Figure 32 shows that the model results in a number equivalent to thermal fault. To
determine whether this thermal fault involves cellulose, the ratio CO/CO2 is used and
is shown as 4.177 which reveal cellulose significant deterioration. Based on the
model output, an asset management decision can be taken as proposed in Table 27.
76
Figure 31. Flow chart of the proposed approach [77].
Figure 32. Proposed overall Fuzz Logic Simulink model
77
Table 23: Asset management decision [77].
The model accuracy has been assessed using other set of DGA data of pre-known
fault conditions that were collected from operating transformers or research papers
such as (DiGiorgio, 2005)[63]; the model showed high agreement with the collected
data. To prove the validity of the fuzzy model with 50 oil samples of two thermal
faults and 25 oil samples of one thermal fault with actual fault type and compare the
final outcome of actual and the fuzzy model. Tables 28 and 29 show samples of the
results of the proposed comprehensive fuzzy logic model. In Table 29, two thermal
faults (F1 and F2) were considered for overheating in cellulose and overheating in
78
paper respectively; the model was not capable to distinguish between the two faults
in many DGA cases. To eliminate this issue, the two thermal faults are compound
and the ratio CO/CO2 was used to determine whether overheating involves cellulose
or not; some results using this later modification are shown in Table 29.
Table 24: - Results of 50 samples from 2000 (based on considering two thermal faults, F5 is
normal)
H2 CH4 C2H4 C2H6 C2H2 CO CO2 Keygas F Duval F Doern.F IEC F Roger F Actual F. Model D.
0.9 9 14 12 0.9 30 840 F5 F2 F1 F2 F6 F5 NORMAL F5[1]
0.9 0.9 0.9 0.9 0.9 3 260 F5 F4 F6 F4 F2 F5 NORMAL F5[1]
0.9 2 2 0.9 0.9 10 420 F5 F2 F1 F4 F6 F5 NORMAL F5[1]
0.9 2 3 0.9 0.9 10 419 F5 F3 F1 F4 F6 F5 NORMAL F5[1]
0.9 1 0.9 2 0.9 10 284 F5 F4 F6 F6 F6 F5 NORMAL F5[1]
0.9 2 3 2 0.9 11 424 F5 F3 F1 F4 F2 F5 NORMAL F5[1]
0.9 0.9 3 0.9 0.9 8 384 F5 F3 F1 F4 F6 F5 NORMAL F5[1]
0.9 2 5 1 0.9 9 424 F5 F2 F1 F4 F6 F5 NORMAL F5[1]
0.9 6 16 3 0.9 7 379 F5 F2 F1 F2 F6 F5 NORMAL F5[1]
0.9 7 25 5 0.9 23 478 F5 F2 F1 F2 F6 F5 NORMAL F5[1]
0.9 2 16 4 0.9 5 471 F5 F2 F1 F6 F6 F5 NORMAL F5[1]
0.9 0.9 6 0.9 0.9 9 398 F5 F2 F1 F6 F6 F5 NORMAL F5[1]
9 3 6 2 0.9 10 475 F5 F2 F4 F4 F4 F5 NORMAL F5[1]
9 9 39 9 0.9 5 420 F5 F2 F1 F6 F2 F5 NORMAL F5[1]
9 53 270 55 220 40 1050 F4 F4 F2 F6 F2 F3 CORONA F3[7.321]
9 47 290 56 240 32 1090 F4 F4 F2 F6 F6 F3 CORONA F3[7.319]
0.9 0.9 0.9 0.9 0.9 10 430 F5 F4 F6 F4 F2 F5 NORMAL F5[1]
0.9 0.9 0.9 0.9 0.9 7 380 F5 F4 F6 F4 F2 F5 NORMAL F5[1]
0.9 0.9 2 0.9 0.9 12 422 F5 F3 F1 F4 F2 F5 NORMAL F5[1]
0.9 0.9 2 0.9 0.9 23 485 F5 F3 F1 F4 F2 F5 NORMAL F5[1]
0.9 0.9 0.9 0.9 0.9 12 450 F5 F4 F6 F4 F2 F5 NORMAL F5[1]
0.9 0.9 0.9 0.9 0.9 1 80 F5 F4 F6 F4 F2 F5 NORMAL F5[1]
0.9 0.9 2 0.9 0.9 19 495 F5 F3 F1 F4 F2 F5 NORMAL F5[1]
0.9 1 1 0.9 0.9 9 441 F5 F4 F6 F4 F6 F5 NORMAL F5[1]
0.9 0.9 2 0.9 0.9 31 635 F5 F3 F1 F4 F2 F5 NORMAL F5[1]
0.9 1 1 1 0.9 7 507 F5 F4 F6 F4 F2 F5 NORMAL F5[1]
0.9 0.9 4 0.9 0.9 53 801 F5 F3 F1 F4 F2 F5 NORMAL F5[1]
9 2 10 0.9 0.9 110 1040 F5 F2 F4 F6 F6 F5 NORMAL F5[1]
9 2 11 0.9 0.9 110 1080 F5 F2 F4 F6 F6 F5 NORMAL F5[1]
9 1 11 0.9 0.9 99 905 F5 F2 F4 F6 F6 F5 NORMAL F5[1]
9 2 12 0.9 0.9 130 980 F5 F2 F4 F6 F6 F5 NORMAL F5[1]
15 52 5 34 0.9 460 4040 F2 F1 F1 F6 F6 F1 THERMAL F1[3.6]
27 36 4 26 0.9 410 2710 F2 F1 F1 F6 F6 F1 THERMAL F1[3.7]
27 56 4 43 0.9 450 3090 F2 F1 F1 F6 F6 F1 THERMAL F1[3.6]
25 38 4 34 0.9 414 2783 F2 F2 F1 F6 F6 F2 THERMAL F2[4.24]
22 52 7 39 0.9 432 2738 F2 F1 F1 F6 F6 F1 THERMAL F1[3.631]
26 50 9 33 0.9 568 2795 F2 F1 F1 F4 F2 F2 THERMAL F2[4.4]
27 51 4 19 0.9 607 2436 F2 F1 F1 F6 F6 F1 THERMAL F1[3.7]
27 52 5 34 0.9 496 3320 F2 F1 F1 F6 F6 F1 THERMAL F1[3.01]
15 47 5 31 0.9 500 2841 F2 F1 F1 F6 F6 F1 THERMAL F1[3.23]
20 55 5 33 0.9 601 3338 F2 F1 F1 F6 F6 F1 THERMAL F1[3.12]
17 66 6 37 0.9 641 3881 F2 F1 F1 F6 F6 F1 THERMAL F1[3.32]
0.9 65 5 42 0.9 601 3639 F2 F1 F1 F6 F6 F1 THERMAL F1[3.23]
0.9 1 0.9 2 0.9 13 237 F5 F4 F6 F6 F6 F5 NORMAL F5[1]
0.9 1 0.9 2 0.9 13 232 F5 F4 F6 F6 F6 F5 NORMAL F5[1]
10 25 5 8 2 344 2396 F5 F3 F1 F6 F6 F5 NORMAL F5[1]
32 44 6 14 2 541 2814 F2 F1 F1 F6 F6 F1 THERMAL F1[3.236]
28 40 5 13 1 458 2598 F2 F1 F1 F6 F6 F1 THERMAL F1[3.637]
25 42 5 14 1 496 2632 F2 F1 F1 F6 F6 F1 THERMAL F1[3.7]
79
Table 25: - Results of 25 samples from 2000 (based on considering one thermal fault, F4 is
normal)
No H2 CH4 C2H2 C2H4 C2H6 CO Actule F. Actule F. Model D.
1 2 7 0 0 0 0 F4 Normal (1)F4
2 54 0 0 4 0 106 F4 Normal (1)F4
3 47 12 0 8 0 115 F4 Normal (1) F4
4 80 619 0 2480 326 268 F1 Thermal (3.93) F1
5 231 3997 0 5584 1726 0 F1 Thermal (4.4)F2
6 507 1053 17 1440 297 22 F1 Thermal (3. 4)F1
7 127 24 81 32 0 0 F3 Arcing (7.04)F3
8 441 207 261 224 43 161 F3 Arcing (6.5) F3
9 217 286 884 458 14 176 F3 Arcing (6.1)F3
10 160 10 1 1 3 N F2 Corona (5.044) F2
11 240 20 96 28 5 N F3 Arcing (6.77)F3
12 2587 7.88 0 1.4 4.7 N F2 Corona (4.67)F2
13 23 6 31 23 172 225 F3 Arcing (6.954)F3
14 103 74 0 9 80 754 F1 Thermal (3.426)F1
15 124 166 0 59 87 530 F1 Thermal (3.3) F1
16 53 49.2 31 2824 514 748 F2 Corona (5.5) F2
17 12 325 109 0 2.9 11.8 F3 Arcing (6.96)F3
18 0 19.3 0 0 57.2 140 F4 Normal (1)F4
19 18.9 303 0.8 0 157 432 F2 Corona (5.39) F2
20 0 46.3 0 6.02 16.4 219.2 F1 Thermal (3.04)F1
21 0 18.8 0 47 60 159 F1 Thermal (3.04)F1
22 12 8778 18.7 11990 4834 317 F1 Thermal (3.93)F1
23 0 73.4 0 0 88.2 124 F2 Corona (5.39)F2
24 65 37 3 42 39 50 F4 Normal (1)F4
25 384 388 33 110 55 173 F2 Corona (5.5)F2
80
6 CHAPTER 6
CONCLUSION AND FUTURE WORK
6.1 RESEARCH CONCLUSION
This research introduces a new interpretation approach for dissolved gas analysis
(DGA) of transformer oil based on the integration of the key strength of all existing
interpretation techniques into one powerful expert model. Current traditional
methods are not consistent and they do not necessarily lead to the same conclusion
for the same oil sample. Moreover, significant number of DGA results fall outside
the proposed codes of ratio- based methods. The new proposed approach relies on
incorporating all traditional DGA interpretation techniques into one fuzzy logic
model. All consistency- weighted decisions of individual DGA interpretation
techniques are combined together to provide one overall decision on each DGA
sample. This decision represents the transformer failure ranking and a proper asset
management action based on the model output can be proposed.
The main conclusion of this research can be drawn as below:
DGA interpretation is not an exact science and there is no one
technique globally accepted as a standard way for DGA interpretation.
Various DGA interpretation techniques do not necessarily lead to the
same conclusion for the same oil sample.
Ratio methods may lead to out of code results and in this case, DGA
samples cannot be interpreted using these methods.
The Key Gas method is very conservative and is not widely accepted
as a tool for DGA interpretation. It may reveal condition 4 (imminent
risk) while the transformer can still be considered healthy as far as the
rate of gas evolution remains constant.
Among the 5 DGA interpretation techniques studied in this thesis,
Duval triangle was found to be the most accurate and consistent
method followed by Key Gas method. Ratio-based methods (Roger,
IEC and Doernenburg) are found to be the less consistent and accurate
methods.
The proposed fuzzy logic model relies on incorporating all 5
81
techniques into one model. Decision of each technique is weighted by
its accuracy level found from analysing 2000 DGA samples collected
from various transformers of different ratings and different life spans.
The final decision is based on all methods.
The proposed technique is aimed to standardize DGA interpretation
techniques. The model robustness was tested against another set of
DGA data with pre-known fault conditions and the agreement was
found very high. The proposed technique eliminates the need for
expert personnel and is easy to implement.
6.2 FUTURE RESEARCH RECOMMENDATIONS
It is suggested to develop online hardware tools for interfacing the data into a fuzzy
model as developed in this research. This will provide operators instant access to
view the transformer overall critical assessment and allow a timely decision. There is
a need to develop monitoring systems to activate alarms and protection devices when
monitored inputs exceed threshold limits. The model together with hardware tool
will satisfy an instant need for the cost effective condition assessment program to
manage risks and initiate maintenance programs, accordingly.
82
REFERENCES
[1] V. Sokolov, "Transformer life management," presented at the II Workshop on
Power Transformers-Deregulation and Transformer Technical, Economic and
Strategical Issues, Salvador, Brazil, 29-31 August 2001.
[2] Nemeth, B., S. Laboncz, and I. Kiss. "Condition Monitoring of Power
Transformers Using Dga and Fuzzy Logic." Paper presented at the Electrical
Insulation Conference, 2009. EIC 2009. IEEE, May 31 2009-June 3 2009.
[2] Siva Sarma, D. V. S. S., and G. N. S. Kalyani. "Ann Approach for Condition
Monitoring of Power Transformers Using DGA." Paper presented at the
TENCON 2004. 2004 IEEE Region 10 Conference, 21-24 Nov. 2004.
[3] Weedy. Electric Power Systems / B.M. Weedy.
[4] N. T. T. Laboratory,"Transformer fluid analysis (furan testing)," Northern
Technology & Testing.
[5] R. Blue, D. Uttamchandani, and O. Farish, " Infrared detection of transformer
insulation degradation due to accelerated thermal aging, " Dielectrics and
Electrical Insulation, IEEE Transactions, vol. 5, pp. 165-168, 1998.
[6] "DGA Diagnostic Methods." Serveron white Paper (2007): 12 pages.
[7] Woolley, T., F. Fetherston, and D. Frost. "Changes in Levels of Dissolved Gas in
Normally Operating Generator Transformers, the Effect of a Nitrogen
Breathing System and the Effects on Interpretation of Dissolved Gas Analysis."
Paper presented at the Properties and Applications of Dielectric Materials,
1994., Proceedings of the 4th International Conference on, 3-8 Jul 1994.
[8] A. M. Emsley and G. C. Stevens, "A reassessment of the low temperature thermal
degradation of cellulose," presented at the 6th
International Conference on
Dielectric Materials, Materials, Measurements and Applications, 1992.
[9] Fu, Yang, and liang Zhang. "Comprehensive Method Detecting the Status of the
Transformer Based on the Artificial Intelligence." Paper presented at the Power
System Technology, 2004. PowerCon 2004. 2004 International Conference on,
21-24 Nov. 2004.
[10] Mei, Denghua. "A New Fuzzy Information Optimization Processing Technique
for Monitoring the Transformer." Paper presented at the Dielectric Materials,
Measurements and Applications, 2000. Eighth International Conference on
(IEE Conf. Publ. No. 473), 2000.
[11] Yann-Chang, Huang, and Sun Huo-Ching. "Dissolved Gas Analysis of Mineral
Oil for Power Transformer Fault Diagnosis Using Fuzzy Logic." Dielectrics
and Electrical Insulation, IEEE Transactions on 20, no. 3 (2013).
[12] Mofizul Islam, S., T. Wu, and G. Ledwich. "A Novel Fuzzy Logic Approach to
Transformer Fault Diagnosis." Dielectrics and Electrical Insulation, IEEE
Transactions on 7, no. 2 (2000): 177-86.
83
[13] Su, Q., L. L. Lai, and P. Austin. "A Fuzzy Dissolved Gas Analysis Method for
the Diagnosis of Multiple Incipient Faults in a Transformer." Paper presented at
the Advances in Power System Control, Operation and Management, 2000.
APSCOM-00. 2000 International Conference on, 30 Oct.-1 Nov. 2000.
[14] CS Chang CW Lim Q Su “Fuzzy – neural approach for dissolved gas analysis of
power transformer incipient faults”
[15] "IEEE Guide for Loading Mineral-Oil-Immersed Transformers." IEEE Std
C57.91-1995 (2012): 1-112.
[16] J. O. Church "Analyze Incipient Faults with Dissolved-Gas Nomograph",
Electrical World, pp.40 -44 1987
[17] http://www.machinerylubrication.com/Read/457/dissolved-gas-analysis .
[18] S. M. Islam, “Power transformer condition monitoring,” Pacific Power
International and University of Newcastle Developments. 24-26 February 1997.
[19] Abu-Siada, A., M. Arshad, and S. Islam. "Fuzzy Logic Approach to Identify
Transformer Criticality Using Dissolved Gas Analysis." Paper presented at the
Power and Energy Society General Meeting, 2010 IEEE, 25-29 July 2010.
[20] F. Fetherston and B. Finlay, "Power Transformer Condition Assessment the
Second Century and Beyond," presented at the Australian Power Engineering
Conference, AUPEC, Curtin University of Technology, Perth. Australia,
September 23-26, 2001.
[21] insulation condition monitoring and reliability centered maintenance of
electrical plant, Powernet’s Experience, Short Course & Workshop on
Transformer Maintenance and Condition Monitoring, Monash University,
Australia 24-26 February 1997.
[22] Y. Du, M. Zahn, B. C. Lesieutre, A. V. Mamishev, and S. R. Lindgren,
“Moisture equilibrium in transformer paper- oil systems,” Electrical Insulation
Magazine, IEEE, vol. 15, pp. 11-20, 1999.
[23] "IEEE Guide for the Interpretation of Gases Generated in Oil-Immersed
Transformers." IEEE Std C57.104-2008 (Revision of IEEE Std C57.104-1991)
(2009): C1-27.
[24] M. Arshad and S. M. Islam, ” Transformer reliability enhancement using online
dissolved gas monitoring and diagnostics,” in Proceedings (CDROM) 2003
International Power Engineering Conference (IPEC), ISBN: 981-04-8705-3, P.
1110.
[25] Su, Q. "A Fuzzy Logic Tool for Transformer Fault Diagnosis." Paper presented
at the Power System Technology, 2000. Proceedings. PowerCon 2000.
International Conference on, 2000.
[26] S. Xu, S. D. Mitchell, and R. H. Middleton, "Partial discharge localization for a
transformer based on frequency spectrum analysis," presented at the Australian
University Power Engineering Conference, AUPEC, Christchurch, New
84
Zealand, 2003.
[27] Standard Test Method for Interfacial Tension of Oil Against Water by the Ring
Method, American National Standards Institute/American Society for Testing
and Materials (ANSI/ASTM) D 971-1991
[28] FIST 3-30, Transformer maintenance (facilities illustrations, standards and
techniques), U.S Department of the Interior Bureau of Reclamation, Denver-
Colorado.
[29] J. Fitch, "The surface tension test – is it worth resurrecting?," in practicing Oil
Analysis Magazine, September 2002.
[30] R.J. Adler, "Gas breakdown pulse power formulary", North Star Research.
[31] Transformer Maintenance Guide, S.D. Myers, Joe Kelly, R.H. Parrish, 1991
[32] E. Robles,"Field diagnostic testing of power generators and transformers using
modern techniques," presented at the HV Testing, Monitoring and Diagnostics
Workshop, Alexandria, Virginia, 2000.
[33] N. T. Testing, " Oil quality (dielectric breakdown voltage), " 8140 Industrial
parkway #8, Sacramento, CA 95824: NTT technical bulletins.
[34] Evaluation of transformer solid insulation, 12388-88th
Avenue Surrey, B. C.
Canada V3W 7R7, Powertech Labs.
[35] B. D. Sparling, "Assessing the life of the transformer," GE Syprotec Inc.
[36] Duval, M. "New Techniques for Dissolved Gas-in-Oil Analysis." Electrical
Insulation Magazine, IEEE 19, no. 2 (2003): 6-15.
[37] Skelly, D. "Photo-Acoustic Spectroscopy for Dissolved Gas Analysis: Benefits
and Experience." Paper presented at the Condition Monitoring and Diagnosis
(CMD), 2012 International Conference on, 23-27 Sept. 2012.
[38] Hmood, S., A. Abu-Siada, M. A. S. Masoum, and S. M. Islam. "Standardization
of DGA Interpretation Techniques Using Fuzzy Logic Approach." Paper
presented at the Condition Monitoring and Diagnosis (CMD), 2012
International Conference on, 23-27 Sept. 2012.
[39] N. Dominelli,"Furanic and non-furanic analysis as a transformer diagnostic, "
presented at the EPRI Substation Equipment Diagnostics Conference IV, New
Orleans, LA, 1996.
[40] M. Webb, "Anticipating failures by dissolved-gas monitoring," Power
Engineering Journal, vol. 1, pp. 295-298, 1987.
[41] O. D. Sparkman, Z. E. Penton, and F. G. Kitson, "Chapter 2 - Gas
Chromatography," in Gas Chromatography and Mass Spectrometry (Second
edition), ed Amsterdam: Academic Press, 2011, pp. 15-83.
[42] R. R. Rogers, "IEEE and IEC Codes to Interpret Incipient Faults in
Transformers, Using Gas in Oil Analysis," Electrical Insulation, IEEE
85
Transactions on, vol. EI-13, pp. 349-354, 1978.
[43] ASTM, "Standard Test Method for Analysis of Gases Dissolved in Electrical
Insulating Oil by Gas Chromatography," ASTM D3612-02 (Reapproved 2009),
2009.
[44] "IEEE Guide for the Detection and Determination of Generated Gases in Oil-
Immersed Transformers and Their Relation to the Serviceability of the
Equipment," ANSI/IEEE Std C57.104-1978, p. 0_1, 1978.
[45] ASTM, "Standard Guide for Sampling, Test Methods, and Specifications for
Electrical Insulating Oils of Petroleum Origin," ASTM D117-10, 2010.
[46] ASTM, "Standard Practice for Sampling Gas from a Transformer Under
Positive Pressure," ASTM D2759-00, 2010.
[47] ASTM, "Standard Practices for Sampling Electrical Insulating Liquids," ASTM
D923-07, 2007.
[48] E. Forgács and T. Cserháti, "CHROMATOGRAPHY | Principles," in
Encyclopedia of Food Sciences and Nutrition (Second Edition), C. Editor-in-
Chief: Benjamin, Ed., ed Oxford: Academic Press, 2003, pp. 1259-1267.
[49] M. Novotny, "Gas Chromatography," in Encyclopedia of Physical Science and
Technology (Third Edition), A. M. Editor-in-Chief: Robert, Ed., ed New York:
Academic Press, 2003, pp. 455-472.
[50] S. Okabe, S. Kaneko, M. Kohtoh, and T. Amimoto, "Analysis results for
insulating oil components in field transformers," Dielectrics and Electrical
Insulation, IEEE Transactions on, vol. 17, pp. 302-311, 2010.
[51] J. Jalbert, R. Gilbert, Y. Denos, and P. Gervais, "Methanol: A Novel Approach
to Power Transformer Asset Management," Power Delivery, IEEE Transactions
on, vol. 27, pp. 514-520, 2012.
[52] J. Jalbert, S. Duchesne, E. Rodriguez-Celis, P. Tetreault, and P. Collin, "Robust
and sensitive analysis of methanol and ethanol from cellulose degradation in
mineral oils," J Chromatogr A, vol. 1256, pp. 240-5, Sep 21 2012.
[53] A. Schaut, S. Autru, and S. Eeckhoudt, "Applicability of methanol as new
marker for paper degradation in power transformers," Dielectrics and Electrical
Insulation, IEEE Transactions on, vol. 18, pp. 533-540, 2011.
[54] J. L. Kirtley, W. H. Hagman, B. C. Lesieutre, M. J. Boyd, E. P. Warren, H. P.
Chou, and R. D. Tabors, "Monitoring the health of power transformers,"
Computer Applications in Power, IEEE, vol. 9, pp. 18-23, 1996.
[55] "IEEE Guide for the Interpretation of Gases Generated in Oil-Immersed
Transformers," IEEE Std C57.104-1991, p. 0_1, 1992.
[56] G. L. Martin, "The Hydrogen on-line monitor(R) a system for the detection and
monitoring of failure conditions in power transformers," in Monitors and
Condition Assessment Equipment (Digest No. 1996/186), IEE Colloquium on,
1996, pp. 3/1-3/5.
86
[57] D. Jie, I. Khan, Z. D. Wang, and I. Cotton, "Comparison of HYDROGEN ON-
LINE MONITOR and laboratory DGA results for electric faults in ester
transformer fluids," in Electrical Insulation and Dielectric Phenomena, 2007.
CEIDP 2007. Annual Report - Conference on, 2007, pp. 731-734.
[58] W. Fu, C. Weigen, P. Xiaojuan, and S. Jing, "Study on the gas pressure
characteristics of photoacoustic spectroscopy detection for dissolved gases in
transformer oil," in High Voltage Engineering and Application (ICHVE), 2012
International Conference on, 2012, pp. 286-289.
[59] D. Skelly, "Photo-acoustic spectroscopy for dissolved gas analysis: Benefits and
Experience," in Condition Monitoring and Diagnosis (CMD), 2012
International Conference on, 2012, pp. 29-43.
[60] F. J. M. Harren, G. Cotti, J. Oomens, and S. t. L. Hekkert, "Photoacoustic
Spectroscopy in Trace Gas Monitoring," in Encyclopedia of Analytical
Chemistry, R. A. Meyers, Ed., ed Chichester: John Wiley & Sons Ltd, 2000, pp.
2203-2226.
[61] Q. Zhu, Y. Yin, Q. Wang, Z. Wang, and Z. Li, "Study on the Online Dissolved
Gas Analysis Monitor based on the Photoacoustic Spectroscopy," in Condition
Monitoring and Diagnosis (CMD), 2012 International Conference on, 2012, pp.
433-436.
[62] C. Haisch and R. Niessner, "Light and sound-Photoacoustic spectroscopy,"
Spectroscopy Europe, vol. 14/5, 2002.
[63] DiGiorgio, Joseph B. (1997). “Dissolved Gas Analysis of Mineral Oil I
nsulating Fluids.” California:Northern Technology & Testing.
[64] Sivanandam, S. N., S. Sumathi, et al. (2007). Introduction to fuzzy logic using
MATLAB. Berlin, Springer.
[65] Muhamad, N. A., B. T. Phung, and T. R. Blackburn. "Comparative Study and
Analysis of Dga Methods for Mineral Oil Using Fuzzy Logic." Paper presented
at the Power Engineering Conference, 2007. IPEC 2007. International, 3-6 Dec.
2007.
[66] Abu-Siada, A., Pin Lai Sin, and S. Islam. "Remnant Life Estimation of Power
Transformer Using Oil Uv-Vis Spectral Response." Paper presented at the
Power Systems Conference and Exposition, 2009. PSCE '09. IEEE/PES, 15-18
March 2009.
[67] Arshad, M., and S. M. Islam. "A Novel Fuzzy Logic Technique for Power
Transformer Asset Management." Paper presented at the Industry Applications
Conference, 2006. 41st IAS Annual Meeting. Conference Record of the 2006
IEEE, 8-12 Oct. 2006.
[68] Spurgeon, K., W. H. Tang, Q. H. Wu, Z. J. Richardson, and G. Moss.
"Dissolved Gas Analysis Using Evidential Reasoning." Science, Measurement
and Technology, IEE Proceedings 152, no. 3 (2005): 110-17.
[69] Akbari, A., A. Setayeshmehr, H. Borsi, and E. Gockenbach. "A Software
87
Implementation of the Duval Triangle Method." Paper presented at the
Electrical Insulation, 2008. ISEI 2008. Conference Record of the 2008 IEEE
International Symposium on, 9-12 June 2008.
[70] Lin, C. E., J. M. Ling, and C. L. Huang. "An Expert System for Transformer
Fault Diagnosis Using Dissolved Gas Analysis." Power Delivery, IEEE
Transactions on 8, no. 1 (1993): 231-38.
[71] Marzuki Khalid (1999). “Fuzzy Logic Control.” UTM : Lecture
Notes(unpublished).
[72] Isa, S. N. M., Z. Ibrahim, and F. Patkar. "Comparative Study of Fuzzy Logic
Speed Controller in Vector Controlled Pmsm Drive: Minimum Number of
Fuzzy Rule-Base." Paper presented at the Innovative Technologies in Intelligent
Systems and Industrial Applications, 2009. CITISIA 2009, 25-26 July 2009.
[73] Ganjdanesh, Y., Y. S. Manjili, M. Vafaei, E. Zamanizadeh, and E. Jahanshahi.
"Fuzzy Fault Detection and Diagnosis under Severely Noisy Conditions Using
Feature-Based Approaches." Paper presented at the American Control
Conference, 2008, 11-13 June 2008.
[74] Pochampally, K. K., and S. M. Gupta. "A Multiphase Fuzzy Logic Approach to
Strategic Planning of a Reverse Supply Chain Network." Electronics Packaging
Manufacturing, IEEE Transactions on 31, no. 1 (2008): 72-82.]
[75] Haema, J., and R. Phadungthin. "Power Transformer Condition Evaluation by
the Analysis of Dga Methods." Paper presented at the Power and Energy
Engineering Conference (APPEEC), 2012 Asia-Pacific, 27-29 March 2012.
[76] Muhamad, N. A., B. T. Phung, T. R. Blackburn, and K. X. Lai. "Comparative
Study and Analysis of Dga Methods for Transformer Mineral Oil." Paper
presented at the Power Tech, 2007 IEEE Lausanne, 1-5 July 2007.
[77] A. Abu-Siada, S. Hmood and Syed Islam,” A New Approach for Consistent
Interpretation of Dissolved Gas- in- Oil Analysis”, IEEE 2013.
[78] www.westernpower.com.au.
[79]Australian CIGRE panel 12 annual reliability survey, Australia 1995.
[80] Schenker, D. F., and T. M. Khoshgoftaar. "The Application of Fuzzy Enhanced
Case-Based Reasoning for Identifying Fault-Prone Modules." Paper presented
at the High-Assurance Systems Engineering Symposium, 1998. Proceedings.
Third IEEE International, 13-14 Nov 1998.
[81] Sung-wook, Kim, Kim Sung-jik, Seo Hwang-dong, Jung Jae-ryong, Yang Hang-
jun, and M. Duval. "New Methods of DGA Diagnosis Using Iec Tc 10 and
Related Databases Part 1: Application of Gas-Ratio Combinations." Dielectrics
and Electrical Insulation, IEEE Transactions on 20, no. 2 (2013).
[82] Kim, Y. M., S. J. Lee, H. D. Seo, J. R. Jung, and H. J. Yang. "Development of
Dissolved Gas Analysis(DGA) Expert System Using New Diagnostic
Algorithm for Oil-Immersed Transformers." Paper presented at the Condition
88
Monitoring and Diagnosis (CMD), 2012 International Conference on, 23-27
Sept. 2012.
[83] Qun-xiong, Zhu, Tao Ting-ting, and Xu Yuan. "A New Flowchart-Based Device
Monitoring and Management System." Paper presented at the Control and
Decision Conference (CCDC), 2013 25th Chinese, 25-27 May 2013 2013.
[84] Soo-jin, Lee, Kim Young-min, Seo Hwang-dong, Jung Jae-ryong, Yang Hang-
jun, and M. Duval. "New Methods of Dga Diagnosis Using Iec Tc 10 and
Related Databases Part 2: Application of Relative Content of Fault Gases."
Dielectrics and Electrical Insulation, IEEE Transactions on 20, no. 2 (2013).
[85] Mosinski, F., and T. Piotrowski. "New Statistical Methods for Evaluation of
Dga Data." Dielectrics and Electrical Insulation, IEEE Transactions on 10, no.
2 (2003): 260-65.
[86] Hong-Tzer, Yang, and Huang Yann-Chang. "Intelligent Decision Support for
Diagnosis of Incipient Transformer Faults Using Self-Organizing Polynomial
Networks." Power Systems, IEEE Transactions on 13, no. 3 (1998): 946-52.
[87] Guanjun, Zhang, K. Yasuoka, S. Ishii, Yang Li, And Yan Zhang. "Application
Of Fuzzy Equivalent Matrix For Fault Diagnosis Of Oil-Immersed Insulation."
Paper Presented At The Dielectric Liquids, 1999. (Icdl '99) Proceedings Of The
1999 Iee 13th International Conference On, 1999.
[88] Cai, Guo-wei, Chao Pan, Yan-tao Wang, and De-you Yang. "A New Method
Based on Fuzzy Topsis for Transformer Dissolved Gas Analysis." Paper
presented at the Sustainable Power Generation and Supply, 2009. SUPERGEN
'09. International Conference on, 6-7 April 2009.
[89] Tu, Yanming, and Zheng Qian. "Dga Based Insulation Diagnosis of Power
Transformer Via Ann." Paper presented at the Properties and Applications of
Dielectric Materials, 2000. Proceedings of the 6th International Conference on,
2000.
[90]Guardado, J. L., J. L. Naredo, P. Moreno, and C. R. Fuerte. "A Comparative
Study of Neural Network Efficiency in Power Transformers Diagnosis Using
Dissolved Gas Analysis." Power Delivery, IEEE Transactions on 16, no. 4
(2001): 643-47.
[91] Duval, M., and A. dePabla. "Interpretation of Gas-in-Oil Analysis Using New
Iec Publication 60599 and Iec Tc 10 Databases." Electrical Insulation
Magazine, IEEE 17, no. 2 (2001): 31-41.