powrem-12

8/12/2019 POWREM-12

1/6

Fuzzy Logic Application in DGA Methods to Classify Fault Type

in Power Transformer

1R.N.AFIQAH, 2I. MUSIRIN, 3D. JOHARI, 4M.M. OTHMAN, 5T.K.A.RAHMAN,6Z.OTHMAN

Centre of Electrical Power Engineering Studies (CEPES)

Faculty of Electrical Engineering

University Technology MARA

Shah Alam

MALAYSIA

[email protected], [email protected], 3dalinaj@ yahoo.com,

[email protected],[email protected], [email protected]

Abstract: - Assessment of power transformer conditions has become increasingly important in recent years. As an

asset that represents one of the largest investments in a utilitys system, detection of incipient faults in powertransformers is crucial. Dissolved gas-in-oil analysis (DGA) is a successful technique to detect these potential faultsand it provides wealth of diagnostic information.This project used two DGA methods which are Rogers Ratio and IECRatio to interpret the DGA results. However, there are situations of errors and misleading results occurring due toborderline and multiple faults. Fuzzy logic is implemented here as an improved DGA interpretation method thatprovides higher reliability and precision of fault diagnosis.

Key-Words: -DGA methods, Fault diagnosis of transformer, Fuzzy logic, Fuzzy inference, IEC ratio, Rogers ratio

1 IntroductionPower transformers are the most critical and expensive

equipments in the operation of modern power system.Since their failure can cause interruptions in the supplyto electrical installations, detection of the incipient faultsis required. Early detection can minimize damage to theequipment and consequently prevent premature

breakdown or failure. It can also lead to huge savings inthe operation and maintenance costs and improves the

overall system reliability.Faults in oil insulator of transformer occur due to

electrical and thermal stresses [1]. The types of faultsthat normally occur in power transformers are arcing,partial discharge, low-energy sparking, severe

overloading, pump motor failure and overheating in theinsulation system. These faults can lead todecomposition of the insulating materials and formationof various gases at different concentration. The faultgases are hydrogen (H2), methane (CH4), ethane (C2H6),ethylene (C2H4), acetylene (C2H2), carbon monoxide(CO) and carbon dioxide (CO2) and they can be founddissolved either in the transformer oil, in the gas blanketabove the oil or in gas collecting devices..

One way to detect the fault is by evaluating theamount of generated gas present and the continuing rateof generation. Since different types of faults generate

different types of gases, the presence of particular gases

can be used to determine the fault type while the rate ofgas generation indicates the severity of the fault.

There are several techniques in detecting thosefault gases and DGA has been recognized as the mosteffective method. It involves sampling of small volumeof the oil from the transformer to measure theconcentration of the dissolved gases and diagnosis of thefault that generates the detected gases. Currently, thereare various methods in DGA and the most common areKey Gas, Rogers Ratio, Doernenburg, LogarithmicNomograph, IEC Ratio and Duval Triangle.

Although DGA has been extensively used in the

industry, in some cases, the conventional methods fail todiagnose. This normally happens for those transformers

which have multiple types of faults. The conventionaldiagnostic methods are based on the ratio of gasesgenerated from a single fault or from multiple faults but

with one dominant nature. When gases from more thanone fault in a transformer are collected, the relation

between different gases becomes too complicated thatthey may not match the codes are pre-defined [2]. FuzzyLogic can be used to diagnose such cases.

Fuzzy Logic provides an approximate but effectivemeans of describing the behaviour of systems that are

too complex or not easily analyzed [3]. It differs fromclassical logic in that statements can take on any real

value between 0 and 1, representing the degree to whichan element belongs to a given set, instead of just either

SELECTED TOPICS in POWER SYSTEMS and REMOTE SENSING

ISSN: 1792-5088 83 ISBN: 978-960-474-233-2

8/12/2019 POWREM-12

2/6

zero or one. With its capability to estimate ambiguityinvolved in a problem, FL can be used to automatically

determine the final diagnosis rules and simultaneouslyadjust membership functions of the corresponding fuzzysubsets before finally arriving at a definite conclusion[4].

In this paper, two DGA methods were investigated:

Rogers Ratio and IEC Ratio. Fuzzy Logic controller wasdeveloped using MATLAB to automate the evaluation

of both methods. Results obtained from the experimentrevealed that Fuzzy technique is a feasible approach inaddressing fault classification in a transformer.

2 Methodology2.1 Fault Gas AnalysisMineral oil functions as a cooler of transformer and

provides insulation as well. When mineral oil issubjected to high thermal and electrical stresses, itdecomposes and as a result, gases are generated. Thesegases are considered as fault indicators and can begenerated in certain patterns and amounts depending onthe characteristics of the fault [5].

Dissolved gas-in-oil analysis (DGA) is a sensitiveand reliable technique to diagnose the incipient and

potential faults in power transformers. By using thistechnique, it is possible to distinguish fault in a greatvariety of oil-filled equipment. Table 1 tabulates thefault types addressed in this paper.

Table 1

Fault Type used in Analysis

Fault Type Fault Type Code

Thermal fault at low temperature TF

Overheating and sparking OH

Arching ARC

Partial Discharge and Corona PD

Normal Normal

2.2 DGA Interpretation MethodsThere are many methods in DGA. The ratio methods are

the most widely used techniques. In this paper, two ofthe ratio methods were studied: Rogers Ratio and IECRatio. Each diagnosis method was grouped according tothe faults type code. This is tabulated in Table 2 [4].

Table 2

Grouping for Fault Type Codes [4]

Method TF OH ARC PD Normal

Roger Slightoverheating 3.0

5

0

1

2

j x < 1.0x 1.0

01

k x < 1.0

1.0 x 3.0

x > 3.0

0

1

2

l x < 0.1

0.1 x 3.0

x > 3.0

0

1

2


ISSN: 1792-5088 84 ISBN: 978-960-474-233-2

8/12/2019 POWREM-12

3/6

Table 5

Classification of Faults based on Rogers

Ratio Codes [6]

i j k l Diagnosis

1-2 0 0 0 Slight overheating

3.0

0

12

i x < 0.1

0.1 x 1.0

x > 1.0

1

0

2

k x < 1.0

1.0 x 3.0

x > 3.0

0

1

2

Table 7

Classification of Fault based on IEC

Ratio Codes [6]

l i k Diagnosis0 0 1 Thermal fault

700C

OH_2

1-

2

0 1-

2

Discharge of low

energy

ARC_1

1 0 2 Discharge of high

energy

ARC_2

0 1 0 PDs of low energydensity

PD_1

1 1 0 PDs of high

energy density

PD_2

0 0 0 Normal normal

Although Rogers Ratio and IEC Ratio are useful,the drawback of these ratio methods is that there can besome combinations of gases that do not fit into thespecified range of values. In multiple fault conditions,

for example, gases from different faults are mixedtogether resulting in confusing ratios between different

gas components [2]. These ratios may not match existingcodes and diagnosis of the fault type cannot be given.This problem can be overcomed by using fuzzydiagnosis presented in the next section.

2.3 Fuzzy Logic ApplicationThe fuzzy analysis consists of three parts:

fuzzification, fuzzy inference and defuzzification.

Fuzzification is the process of transforming crisp inputvalues into grades of membership for linguistic terms offuzzy sets. The membership function is used to associate

a grade to each linguistic term. A chosen fuzzy inferencesystem (FIS) is responsible for drawing conclusions

from the knowledge-based fuzzy rule set of if-thenlinguistic statements. Fault types listed in Table 5 andTable 7 form the fuzzy rule set for the diagnosis system.Defuzzification then converts the fuzzy output valuesback into crisp output actions.

2.3.1 Fuzzy Rogers RatioAs shown in Table 4, the four ratio codes are defined asinputs and classified as either Low (Lo), Medium (Med),High (Hi) and Very High (Vhi) [4]. Table 8 shows themembership intervals of each ratio.


ISSN: 1792-5088 85 ISBN: 978-960-474-233-2

8/12/2019 POWREM-12

4/6

Table 8

Membership Intervals for Rogers Ratio

Ratio code Range Code

i x < 0.1

0.1 x 1.01.0 x 3.0

x > 3.0

5 (Lo)

0 (Med)1 (Hi)

2 (Vhi)

j x < 1.0

x 1.0

0 (Lo)

1 (Hi)

k x < 1.0

1.0 x 3.0

x > 3.0

0 (Lo)

1 (Med)

2 (Hi)

l x < 0.1

0.1 x 3.0

x > 3.0

0 (Lo)

1 (Med)

2 (Hi)

The membership boundaries for ratio codes i, j, k and lare represented by a trapezoidal fuzzy-membership

function illustrated in Fig. 1 respectively.

Fig.1: Membership function for ratio codes

i, j, k and l respectively

Fuzzy inference uses IF-THEN rule-based system,given by, IF antecedent and THEN consequent. Here, afuzzy rule set is then used to form judgement on thefuzzy inputs derived from the 4 gas ratios.18 fuzzy rules

can be derived and here are some examples of the fuzzyrules based on the fault types listed in Table 5:

Rules 1: IF i is high AND j is lowAND k is low AND l is low THENFaults is TF_1

Rules 6: IF i is medium AND j islow AND k is medium AND l is low

THEN Faults is OH_1

When fuzzy rule has multiple antecedents, the

fuzzy operator AND for minimization operator and ORfor maximization operator is used to obtain a single

number that represents the result of the antecedentevaluation. Therefore, the following equations areproduced based on Rogers Ratio rules to diagnose

different fault:

Fault(TF_1) = max{ min[i=Hi,j=Lo, k=Lo, l=Lo], min[i=Vhi,

j=Lo, k=Lo, l=Lo]}

Fault(OH_1) = min[i=Med, j=Lo,k=Med, l=Lo]

Although the fuzzy rules appear strictly defined,borderline cases with gas ratio on or near the linebetween linguistic values (low, medium, high and very

high) allows FIS to interpret membership of these rulesflexibly and classify these cases under two different fault

types with individual probability of occurrence attachedto each type [3].

FIS involves the operations between input fuzzysets, as illustrated graphically in Fig. 2. It is based onfuzzy inference described previously.

Fig.2: FIS analysis


ISSN: 1792-5088 86 ISBN: 978-960-474-233-2

8/12/2019 POWREM-12

5/6

As illustrated in Fig. 2, each rule is a row of plotsand each column is a variable. The first four columns of

plots (yellow) show the membership functionsreferenced by the antecedent, or the if-part of each rule.The fifth column of plots (blue) shows the membershipfunctions referenced by the consequent, or the then-partof each rule.

2.3.2 Fuzzy IEC RatioThe three gas ratio codes are defined as inputs andclassified as either Low (Lo), Medium (Med) and High(Hi) according to membership intervals as tabulated inTable 9. Fig. 3 is the fuzzy membership function for thismethod.

Table 9

Membership Intervals for IEC Ratio

Ratio code Range Code

l x < 0.1

0.1 x 3.0

x > 3.0

0 (Lo)

1 (Med)

2 (Hi)

i x < 0.1

0.1 x 1.0x > 1.0

1 (Lo)

0 (Med)2 (Hi)

k x < 1.0

1.0 x 3.0

x > 3.0

0 (Lo)

1 (Med)

2 (Hi)

Fig.3: Membership function for ratio

codes l, i and k respectively

12 fuzzy rules can be defined based on the faulttypes listed in Table 6. The fuzzy analysis of this method

was developed using the same technique as described inthe previous method.

3 Results and DiscussionIn order to evaluate the performance of fuzzy logic toclassify the fault types of the transformer, 15 oil sampleswere used and they were tabulated in Table 10 [7].

Table 10

DGA Samples [7]

No. H2 CH4 C2H6 C2H4 C2H2

1 200 700 250 740 1

2 56 61 75 32 31

3 33 26 6 5.3 0.2

4 176 205.9 47.7 75.7 68.7

5 70.4 69.5 28.9 241.2 10.4

6 345 112.25 27.5 51.5 58.75

7 172.9 334.1 172.9 812.5 37.7

8 2587.2 7.882 4.704 1.4 0

9 1678 652.9 80.7 1005.9 419.1

10 206 198.9 74 612.7 15.1

11 180 175 75 50 4

12 106 24 4 28 37

13 180.85 0.574 0.234 0.188 0

14 27 90 24 63 0.2

15 138.8 52.2 6.77 62.8 9.55

Table 11 tabulates actual fault and result of type of faultsfor each oil sample using both DGA methods, RogersRatio and IEC Ratio by applying the fuzzy logic. The

results between the actual inspection of transformer andDGA methods diagnosis using fuzzy logic are

approximately the same. From Table 11, it was foundthat the Rogers Ratio with fuzzy logic made correctdiagnosis in 12 samples out of 15 and IEC Ratio withfuzzy logic can correctly diagnose 13 of the 15 samples.The accuracy of IEC Ratio method is nearly 87% which

is higher compared to Rogers Ratio method with 80%

accuracy. In some samples, however, both DGAmethods with fuzzy logic failed to fit the actual fault.The reasons are unclear.

From the result obtained, it is proved that byapplying the fuzzy logic, the incipient fault can bedefined and classified effectively.


ISSN: 1792-5088 87 ISBN: 978-960-474-233-2

8/12/2019 POWREM-12

6/6

4 ConclusionIn this paper, two methods have been applied for theinterpretation of fault types from the DGA data namelyRogers Ratio and IEC Ratio. Fuzzy logic was exploredto improve the diagnosis technique. It has been proventhat by using fuzzy logic, the fault type of transformercan be obtained efficiently. Fuzzy logic is applied as thepractical representation of the relationship between thefault type and the dissolved levels with fuzzymembership functions. Multiple faults can also bediagnosed by applying fuzzy logic approach. Byapplying fuzzy logic in DGA methods, the lifespan oftransformer can be increased while the cost of

maintenance can be reduced accordingly. In order toincrease the accuracy of this method, more transformersamples should be analyzed in comparison with actualfault. In addition, appropriate membership functions and

rules are necessary to obtain acceptable accuracy.

Acknowledgement:

The authors would like to express their gratitude to allthe people who have directly or indirectly contributedtowards the successful completion of this technical

paper.

References:[1] J. Aragon-Patil, M. Fischer, and S. Tenbohlen,Improvement of dissolved gas analysis (DGA) bymeans of experimental investigations of generated fault

gases and a fuzzy logic based interpretation scheme,2007.

[2] Q. Su, L. L. Lai, and P. Austin, A FuzzyDissolved Gas Analysis Method for the Diagnosis of

Multiple Incipient Faults in a Transformer, IEEETransactions on Power Systems, Vol. 15, No. 2, pp. 593-598, May 2000.

[3] C. Chang, C. Lim, and Q. Su, Fuzzy-Neural

Approach for Dissolved Gas Analysis of PowerTransformer Incipient Faults, Australian UniversitiesPower Engineering Conference (AUPEC 2004),Brisbane, Australia, 26-29 September, 2004.[4] N. A. Muhamad, B. T. Phung, and T. R.Blackburn, Fuzzy Logic Application in Evaluation of

DGA Interpretation Methods, Vol. 2, No. 1, pp. 117-123, March 2009.[5] K. M. Gradnik, Physical-Chemical Oil Tests,Monitoring and Diagnostic of Oil-filled Transformers,Proceeding of 14

th International Conference on

Dielectric Liquids, Austria, July 2002.[6] Siva Sarma, D. V. S. S. and G. N. S. Kalyani,ANN Approach for Condition Monitoring of PowerTransformers using DGA, 2004 IEEE Region 10Conference, TENCON 2004., pp. 444-447, 2004.[7] Hongzhong Ma, Zheng Li, and P. Ju, Diagnosisof Power Transformer Faults Based On Fuzzy Three-

Ratio Method, 7th International Power EngineeringConference, 2005. IPEC 2005, 2005.

Table 11

Diagnosis Results by Rogers Ratio Method and IEC Ratio Method

No. Actual faultRogers

Ratio

Fault

classification

IEC

Ratio

Fault

classification

1 Overheating and sparking 2010 Overheating and sparking 022 Overheating and sparking

2Partial Discharge and

Corona1101 Overheating and sparking* 120 Arching*

3 Normal 0000 Normal 000 Normal4 Arching 1011 Overheating and sparking* 121 Arching5 Overheating and sparking 0020 Overheating and sparking 002 Overheating and sparking6 Arching 0011 Arching 101 Arching7 Overheating and sparking 1020 Overheating and sparking 022 Overheating and sparking


Corona5000

Partial Discharge andCorona


Corona

9 Arching 0021 Arching 102 Arching

10 Overheating and sparking 0020 Overheating and sparking 002 Overheating and sparking

11Thermal fault at low

temperature0000 Normal* 000 Normal*

12 Arching 0021 Arching 202 Arching


Corona5000

Partial Discharge and

Corona010

Partial Discharge and

Corona

14 Overheating and sparking 2010 Overheating and sparking 021 Overheating and sparking15 Arching 0021 Arching 102 Arching

Note: *no corresponding with actual result


ISSN: 1792-5088 88 ISBN: 978-960-474-233-2

powrem-12

Documents