SELECTED TOPICS in POWER SYSTEMS and REMOTE SENSING

Fuzzy Logic Application in DGA Methods to Classify Fault Type in Power Transformer
1

R.N.AFIQAH, 2I. MUSIRIN, 3D. JOHARI, 4M.M. OTHMAN, 5T.K.A.RAHMAN, 6 Z.OTHMAN Centre of Electrical Power Engineering Studies (CEPES) Faculty of Electrical Engineering University Technology MARA Shah Alam MALAYSIA

afiqah.romainor@gmail.com, 2ismailbm@salam.uitm.edu.my, 3dalinaj@ yahoo.com, 4 mamat505my@yahoo.com, 5khawa@salam.uitm.edu.my, 6zulkifli_1980@yahoo.co.uk

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 power transformers is crucial. Dissolved gas-in-oil analysis (DGA) is a successful technique to detect these potential faults and it provides wealth of diagnostic information. This project used two DGA methods which are Rogers Ratio and IEC Ratio to interpret the DGA results. However, there are situations of errors and misleading results occurring due to borderline and multiple faults. Fuzzy logic is implemented here as an improved DGA interpretation method that provides higher reliability and precision of fault diagnosis. Key-Words: - DGA methods, Fault diagnosis of transformer, Fuzzy logic, Fuzzy inference, IEC ratio, Rogers ratio

1 Introduction
Power transformers are the most critical and expensive equipments in the operation of modern power system. Since their failure can cause interruptions in the supply to electrical installations, detection of the incipient faults is required. Early detection can minimize damage to the equipment and consequently prevent premature breakdown or failure. It can also lead to huge savings in the 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 faults that normally occur in power transformers are arcing, partial discharge, low-energy sparking, severe overloading, pump motor failure and overheating in the insulation system. These faults can lead to decomposition of the insulating materials and formation of various gases at different concentration. The fault gases are hydrogen (H2), methane (CH4), ethane (C2H6), ethylene (C2H4), acetylene (C2H2), carbon monoxide (CO) and carbon dioxide (CO2) and they can be found dissolved either in the transformer oil, in the gas blanket above the oil or in gas collecting devices.. One way to detect the fault is by evaluating the amount of generated gas present and the continuing rate of 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 of gas generation indicates the severity of the fault. There are several techniques in detecting those fault gases and DGA has been recognized as the most effective method. It involves sampling of small volume of the oil from the transformer to measure the concentration of the dissolved gases and diagnosis of the fault that generates the detected gases. Currently, there are various methods in DGA and the most common are Key Gas, Rogers Ratio, Doernenburg, Logarithmic Nomograph, IEC Ratio and Duval Triangle. Although DGA has been extensively used in the industry, in some cases, the conventional methods fail to diagnose. This normally happens for those transformers which have multiple types of faults. The conventional diagnostic methods are based on the ratio of gases generated from a single fault or from multiple faults but with one dominant nature. When gases from more than one fault in a transformer are collected, the relation between different gases becomes too complicated that they may not match the codes are pre-defined [2]. Fuzzy Logic can be used to diagnose such cases. Fuzzy Logic provides an approximate but effective means of describing the behaviour of systems that are too complex or not easily analyzed [3]. It differs from classical logic in that statements can take on any real value between 0 and 1, representing the degree to which an element belongs to a given set, instead of just either

ISSN: 1792-5088

83

ISBN: 978-960-474-233-2

SELECTED TOPICS in POWER SYSTEMS and REMOTE SENSING

zero or one. With its capability to estimate ambiguity involved in a problem, FL can be used to automatically determine the final diagnosis rules and simultaneously adjust membership functions of the corresponding fuzzy subsets before finally arriving at a definite conclusion [4]. In this paper, two DGA methods were investigated: Rogers Ratio and IEC Ratio. Fuzzy Logic controller was developed using MATLAB to automate the evaluation of both methods. Results obtained from the experiment revealed that Fuzzy technique is a feasible approach in addressing fault classification in a transformer.

Table 2 Grouping for Fault Type Codes [4]

Method Roger TF Slight overheating <150C Overheating 150C-200C Overheating 200C-300C OH Conductor overheating Winding circulating current Core/tank circulating current Thermal fault 300C700C Thermal fault >700C Discharge of low energy Discharge of high energy PDs of low energy density PDs of high energy density Normal ARC Flashover Arching Continuous sparking PD PDs PDs with tracking Normal Normal

IEC

Thermal fault <150C Thermal fault 150C-300C

2 Methodology
2.1 Fault Gas Analysis
Mineral oil functions as a cooler of transformer and provides insulation as well. When mineral oil is subjected to high thermal and electrical stresses, it decomposes and as a result, gases are generated. These gases are considered as fault indicators and can be generated in certain patterns and amounts depending on the characteristics of the fault [5]. Dissolved gas-in-oil analysis (DGA) is a sensitive and reliable technique to diagnose the incipient and potential faults in power transformers. By using this technique, it is possible to distinguish fault in a great variety of oil-filled equipment. Table 1 tabulates the fault types addressed in this paper.
Table 1 Fault Type used in Analysis Fault Type Thermal fault at low temperature Overheating and sparking Arching Partial Discharge and Corona Normal Fault Type Code TF OH ARC PD Normal

2.2.1 Rogers Ratio

The Rogers Ratio method utilizes four gas ratios: CH4/H2, C2H6/CH4, C2H4/C2H6 and C2H2/C2H4. Diagnosis of faults is accomplished via a simple coding scheme based on ranges of the ratios as tabulated in Table 3 and 4 [6].
Table 3 Gas Ratio Codes [6] Gas ratios CH4/H2 C2H6/CH4 C2H4/C2H6 C2H2/C2H4 Ratio codes i j k l

Table 4 Rogers Ratio Codes [6] Ratio code i Range x < 0.1 0.1 x 1.0 1.0 x 3.0 x > 3.0 x < 1.0 x 1.0 x < 1.0 1.0 x 3.0 x > 3.0 x < 0.1 0.1 x 3.0 x > 3.0 Code 5 0 1 2 0 1 0 1 2 0 1 2

2.2 DGA Interpretation Methods

There are many methods in DGA. The ratio methods are the most widely used techniques. In this paper, two of the ratio methods were studied: Rogers Ratio and IEC Ratio. Each diagnosis method was grouped according to the faults type code. This is tabulated in Table 2 [4].
j k

ISSN: 1792-5088

84

ISBN: 978-960-474-233-2

SELECTED TOPICS in POWER SYSTEMS and REMOTE SENSING

Table 5 Classification of Faults based on Rogers Ratio Codes [6] i 1-2 1-2 0 0 1 1 j 0 1 1 0 0 0 k 0 0 0 1 1 2 l 0 0 0 0 0 0 Diagnosis Slight overheating <150C Overheating 150C200C Overheating 200C300C Conductor overheating Winding circulating currents Core/tank circulating currents, overhead joints Flashover without power follow through Arc with power follow through Continuous sparking to floating potential Partial discharge Partial discharge with tracking (note CO) Normal deterioration TF_1 TF_2 TF_3 OH_1 OH_2 OH_3 l 0 0 0 0 12 1 0 ARC_2 1 ARC_3 0 PD_1 PD_2 Normal

Table 7 Classification of Fault based on IEC Ratio Codes [6] i 0 2 2 2 0 0 1 1 0 k 1 0 1 2 12 2 0 0 0 Diagnosis Thermal fault <150C Thermal fault 150C-300C Thermal fault 300C-700C Thermal fault >700C Discharge of low energy Discharge of high energy PDs of low energy density PDs of high energy density Normal TF_1 TF_2 OH_1 OH_2 ARC_1 ARC_2 PD_1 PD_2 normal

0 0 0 5 5 0

0 0 0 0 0 0

0 1-2 2 0 0 0

1 1-2 2 0 1-2 0

ARC_1

2.2.2 IEC Ratio

The IEC Ratio is an improvement over the Rogers method. Instead of using four gas ratios, ratio C2H6/CH4 was dropped because it indicated only a limited temperature range of decomposition [4]. The remaining three gas ratios have different ranges of code compared to Rogers method and their values are tabulated in Table 6 [6]. The fault interpretations can be divided into 9 different types and they are shown in Table 7 [6].
Table 6 IEC Ratio Codes [6] Ratio code l Range x < 0.1 0.1 x 3.0 x > 3.0 x < 0.1 0.1 x 1.0 x > 1.0 x < 1.0 1.0 x 3.0 x > 3.0 Code 0 1 2 1 0 2 0 1 2

Although Rogers Ratio and IEC Ratio are useful, the drawback of these ratio methods is that there can be some combinations of gases that do not fit into the specified range of values. In multiple fault conditions, for example, gases from different faults are mixed together resulting in confusing ratios between different gas components [2]. These ratios may not match existing codes and diagnosis of the fault type cannot be given. This problem can be overcomed by using fuzzy diagnosis presented in the next section.

2.3 Fuzzy Logic Application

The fuzzy analysis consists of three parts: fuzzification, fuzzy inference and defuzzification. Fuzzification is the process of transforming crisp input values into grades of membership for linguistic terms of fuzzy sets. The membership function is used to associate a grade to each linguistic term. A chosen fuzzy inference system (FIS) is responsible for drawing conclusions from the knowledge-based fuzzy rule set of if-then linguistic statements. Fault types listed in Table 5 and Table 7 form the fuzzy rule set for the diagnosis system. Defuzzification then converts the fuzzy output values back into crisp output actions.

2.3.1 Fuzzy Rogers Ratio

As shown in Table 4, the four ratio codes are defined as inputs and classified as either Low (Lo), Medium (Med), High (Hi) and Very High (Vhi) [4]. Table 8 shows the membership intervals of each ratio.

ISSN: 1792-5088

85

ISBN: 978-960-474-233-2

SELECTED TOPICS in POWER SYSTEMS and REMOTE SENSING

Table 8 Membership Intervals for Rogers Ratio Ratio code i Range x < 0.1 0.1 x 1.0 1.0 x 3.0 x > 3.0 x < 1.0 x 1.0 x < 1.0 1.0 x 3.0 x > 3.0 x < 0.1 0.1 x 3.0 x > 3.0 Code 5 (Lo) 0 (Med) 1 (Hi) 2 (Vhi) 0 (Lo) 1 (Hi) 0 (Lo) 1 (Med) 2 (Hi) 0 (Lo) 1 (Med) 2 (Hi)

can be derived and here are some examples of the fuzzy rules based on the fault types listed in Table 5: Rules 1: IF i is high AND j is low AND k is low AND l is low THEN Faults is TF_1 Rules 6: IF i is medium AND j is low 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 OR for maximization operator is used to obtain a single number that represents the result of the antecedent evaluation. Therefore, the following equations are produced 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 line between linguistic values (low, medium, high and very high) allows FIS to interpret membership of these rules flexibly and classify these cases under two different fault types with individual probability of occurrence attached to each type [3]. FIS involves the operations between input fuzzy sets, as illustrated graphically in Fig. 2. It is based on fuzzy inference described previously.

j k

The membership boundaries for ratio codes i, j, k and l are 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, a fuzzy rule set is then used to form judgement on the fuzzy inputs derived from the 4 gas ratios. 18 fuzzy rules

Fig.2: FIS analysis

ISSN: 1792-5088

86

ISBN: 978-960-474-233-2

SELECTED TOPICS in POWER SYSTEMS and REMOTE SENSING

As illustrated in Fig. 2, each rule is a row of plots and each column is a variable. The first four columns of plots (yellow) show the membership functions referenced by the antecedent, or the if-part of each rule. The fifth column of plots (blue) shows the membership functions referenced by the consequent, or the then-part of each rule.

12 fuzzy rules can be defined based on the fault types listed in Table 6. The fuzzy analysis of this method was developed using the same technique as described in the previous method.

3 Results and Discussion

In order to evaluate the performance of fuzzy logic to classify the fault types of the transformer, 15 oil samples were used and they were tabulated in Table 10 [7].
Table 10 DGA Samples [7] No. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 H2 200 56 33 176 70.4 345 172.9 2587.2 1678 206 180 106 180.85 27 138.8 CH4 700 61 26 205.9 69.5 112.25 334.1 7.882 652.9 198.9 175 24 0.574 90 52.2 C2H6 250 75 6 47.7 28.9 27.5 172.9 4.704 80.7 74 75 4 0.234 24 6.77 C2H4 740 32 5.3 75.7 241.2 51.5 812.5 1.4 1005.9 612.7 50 28 0.188 63 62.8 C2H2 1 31 0.2 68.7 10.4 58.75 37.7 0 419.1 15.1 4 37 0 0.2 9.55

2.3.2 Fuzzy IEC Ratio

The three gas ratio codes are defined as inputs and classified as either Low (Lo), Medium (Med) and High (Hi) according to membership intervals as tabulated in Table 9. Fig. 3 is the fuzzy membership function for this method.
Table 9 Membership Intervals for IEC Ratio Ratio code l Range x < 0.1 0.1 x 3.0 x > 3.0 x < 0.1 0.1 x 1.0 x > 1.0 x < 1.0 1.0 x 3.0 x > 3.0 Code 0 (Lo) 1 (Med) 2 (Hi) 1 (Lo) 0 (Med) 2 (Hi) 0 (Lo) 1 (Med) 2 (Hi)

Table 11 tabulates actual fault and result of type of faults for each oil sample using both DGA methods, Rogers Ratio and IEC Ratio by applying the fuzzy logic. The results between the actual inspection of transformer and DGA methods diagnosis using fuzzy logic are approximately the same. From Table 11, it was found that the Rogers Ratio with fuzzy logic made correct diagnosis in 12 samples out of 15 and IEC Ratio with fuzzy 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 DGA methods with fuzzy logic failed to fit the actual fault. The reasons are unclear. From the result obtained, it is proved that by applying the fuzzy logic, the incipient fault can be defined and classified effectively.

Fig.3: Membership function for ratio codes l, i and k respectively

ISSN: 1792-5088

87

ISBN: 978-960-474-233-2

SELECTED TOPICS in POWER SYSTEMS and REMOTE SENSING

Table 11 Diagnosis Results by Rogers Ratio Method and IEC Ratio Method
No. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Actual fault Overheating and sparking Partial Discharge and Corona Normal Arching Overheating and sparking Arching Overheating and sparking Partial Discharge and Corona Arching Overheating and sparking Thermal fault at low temperature Arching Partial Discharge and Corona Overheating and sparking Arching Rogers Ratio 2010 1101 0000 1011 0020 0011 1020 5000 0021 0020 0000 0021 5000 2010 0021 Fault classification Overheating and sparking Overheating and sparking* Normal Overheating and sparking* Overheating and sparking Arching Overheating and sparking Partial Discharge and Corona Arching Overheating and sparking Normal* Arching Partial Discharge and Corona Overheating and sparking Arching IEC Ratio 022 120 000 121 002 101 022 010 102 002 000 202 010 021 102 Fault classification Overheating and sparking Arching* Normal Arching Overheating and sparking Arching Overheating and sparking Partial Discharge and Corona Arching Overheating and sparking Normal* Arching Partial Discharge and Corona Overheating and sparking Arching

Note: *no corresponding with actual result

4 Conclusion
In this paper, two methods have been applied for the interpretation of fault types from the DGA data namely Rogers Ratio and IEC Ratio. Fuzzy logic was explored to improve the diagnosis technique. It has been proven that by using fuzzy logic, the fault type of transformer can be obtained efficiently. Fuzzy logic is applied as the practical representation of the relationship between the fault type and the dissolved levels with fuzzy membership functions. Multiple faults can also be diagnosed by applying fuzzy logic approach. By applying fuzzy logic in DGA methods, the lifespan of transformer can be increased while the cost of maintenance can be reduced accordingly. In order to increase the accuracy of this method, more transformer samples should be analyzed in comparison with actual fault. In addition, appropriate membership functions and rules are necessary to obtain acceptable accuracy. Acknowledgement: The authors would like to express their gratitude to all the people who have directly or indirectly contributed towards the successful completion of this technical paper. References: [1] J. Aragon-Patil, M. Fischer, and S. Tenbohlen, Improvement of dissolved gas analysis (DGA) by means 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 Fuzzy Dissolved Gas Analysis Method for the Diagnosis of Multiple Incipient Faults in a Transformer, IEEE Transactions on Power Systems, Vol. 15, No. 2, pp. 593598, May 2000. [3] C. Chang, C. Lim, and Q. Su, Fuzzy-Neural Approach for Dissolved Gas Analysis of Power Transformer Incipient Faults, Australian Universities Power 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. 117123, March 2009. [5] K. M. Gradnik, Physical-Chemical Oil Tests, Monitoring and Diagnostic of Oil-filled Transformers, Proceeding of 14th 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 Power Transformers using DGA, 2004 IEEE Region 10 Conference, TENCON 2004., pp. 444-447, 2004. [7] Hongzhong Ma, Zheng Li, and P. Ju, Diagnosis of Power Transformer Faults Based On Fuzzy ThreeRatio Method, 7th International Power Engineering Conference, 2005. IPEC 2005, 2005.

ISSN: 1792-5088

88

ISBN: 978-960-474-233-2