You are on page 1of 6

Comparative Study and Analysis of DGA Methods for Transformer Mineral Oil

N.A. Muhamad, B.T. Phung, T.R. Blackburn, K.X Lai

The University of New South Wales, School of Electrical Engineering and


Telecommunications, Sydney 2052, Australia

Abstract – Dissolved gas-in-oil analysis (DGA) is a sensitive C2H4 and C2H2. Table 1 shows the set of data used in this
and reliable technique for the detection of incipient fault paper.
condition within oil-immersed transformers. The presence of
Fault Type Faults Type Number of
certain key gases is monitored and quantified. There are a
Code cases
number of methods developed for analyzing these gases and Thermal fault at low F1 10
interpreting their significance: Key Gas, Rogers Ratio, temperature
Doernenburg, Logarithmic Nomograph, IEC Ratio and Duval Overheating and sparking F2 33
Arcing F3 22
Triangle. This paper investigates the accuracy and consistency
Partial Discharge and Corona F4 14
of these methods in interpreting the transformer condition. F5
Normal 13
The evaluation is carried out on DGA data obtained from the
Table 1: Set of data used in analysis
local power utilities and from published papers. The data
In this paper, MATLAB was used to develop a program to
consists of 92 different cases. The key gases considered are
test each method. This involves some basic coding and
hydrogen, methane, ethane, ethylene and acetylene. A
construction of Simulink block diagrams.
MATLAB program was developed to automate the evaluation
of the methods.
A. DGA Interpretation Methods
I. INTRODUCTION
Mineral oils are mixtures of many different hydrocarbon Insulating oils under abnormal electrical or thermal stress
molecules. They are composed essentially of saturated breakdown to liberate small quantities of gases. The
hydrocarbon called paraffin whose general molecular formula composition of these gases is dependent upon type of fault. By
is CnH2n+2 with n the range of 20 to 40 [1]. When use in means of dissolved gas analysis (DGA), it is possible to
transformers, the oil acts as a dielectric medium and also as a distinguish fault such as partial discharge (corona),
overheating, and arcing in a great variety of oil filled
heat transfer agent. The breakdown of electrical insulating
equipment [6]. Similar to a blood test or scanner examination
materials and related components inside the transformer
of human body, DGA can give early diagnosis and increase
liberates gases within the unit. The distribution of these gases
the chances of finding the appropriate cure. There are many
can be related to the type of electrical fault, and the rate of gas methods in DGA. In this paper six of the more commonly
generation can indicate the severity of the fault. used methods were studied:
The identity of the gases being generated by a particular
i. Roger Ratio Method
unit can be very useful information in any preventive
maintenance program [2]. There are several techniques in The Roger’s method utilizes four gas ratios: CH4/H2,
detecting those fault gases and DGA was recognized as the C2H6/CH4, C2H4/C2H6 and C2H2/C2H4. Diagnosis of faults is
most informative method. This method involves sampling of accomplished via a simple coding scheme based on ranges of
the oil to measure the concentration of the dissolved gases. the ratios as shown in tables 2 and 3 below [3].
The most important aspect of fault gases analysis is the
correct diagnosis of the fault that generated the detected gases.
Currently there are several methods developed to do the
interpretation of the fault type from the dissolved gases data.
In this paper, the six methods of interpretation of the fault
gases are investigated and compared. They are: Key Gas, Table 2: Gas Ratio Codes [3].
Rogers Ratio, Doernenburg, Logarithmic Nomograph, IEC
Ratio and Duval Triangle. The study was done to evaluate the
accuracy of each method in predicting the fault and the
consistency of each method.

II. METHODOLOGY
The six methods are tested to interpret 92 data sets of 5
fault gases [1, 3-5]. These five key gases are H2, CH4, C2H6,
Table 3: Roger’s Ratio Codes [3].

978-1-4244-2190-9/07/$25.00 ©2007 IEEE 45 PowerTech 2007


The combination of the coding gives 12 different types of iii. Doernenburg Ratio Method
transformer faults. The type of faults based on the code is
shown in table 4 below [3]. This method utilizes the gas concentration from ratio of
CH4/H2, C2H2/CH4, C2H4/C2H6 and C2H2/C2H4. The value of
the gases at first must exceed the concentration L1 to ascertain
whether there is really a problem with the unit and then
whether there is sufficient generation of each gas for the ratio
analysis to be applicable [7]. Table 7 shows the key gases and
their concentration L1 [7].

Table 7: Concentration L1 for Doernenburg Ratio method [7]

According to IEEE Standard C57.104-1991[7], the step-by-


Table 4: Classification based on Roger’s Ratio Codes [3] step procedure to diagnose faults using Doernenburg ratio
method is:
ii. IEC Ratio Method
Step 1. Gas concentrations are obtained by extracting the
This method originated from the Roger’s Ratio method, gases and separating them by chromatograph
except that the ratio C2H6/CH4 was dropped since it only Step 2. If at least one of the gas concentrations (in ppm) for
indicated a limited temperature range of decomposition [3]. H2, CH4, C2H2, and C2H4 exceeds twice the values
Here, the remaining three gas ratios have different ranges of for limit L1 (see table 7) and one of the other three
code as compared to the Roger’s ratio method and they are gases exceeds the values for limit L1, the unit is
shown in table 5. The faults are divided into nine different considered faulty; proceed to Step 3.
types as listed in table 6 [3]. Step 3. Determining validity of ratio procedure: If at least
one of the gases in each ratio CH4/H2, C2H2/CH4,
C2H2/CH4 and C2H6/C2H2 exceeds limit L1, the
ratio procedure is valid. Otherwise, the ratios are
not significant, and the unit should be resample and
investigated by alternative procedures.
Step 4. Assuming that the ratio analysis is valid, each
successive ratio is compared to the values obtained
from table 8 in the order of ratio CH4/H2,
C2H2/CH4, C2H2/CH4 and C2H6/C2H2
Step 5. If all succeeding ratios for a specific fault type fall
within the values (column) given in Table 8 [7], the
suggested diagnosis is valid.
Table 5: IEC Ratio Codes [3]

Table 8: Fault diagnosis for Doernenburg Ratio Method [7]

iv. Duval Triangle Method.

M. Duval developed this method in the 1960s. According to


[8], to determine whether a problem exists at least one of the
Table 6: Classification based on IEC Ratio Codes [3] hydrocarbon gases or hydrogen must be at L1 level or above

46
and the gas generation rate is at least at G2. The L1 level and The principle of the Key Gas method is based on the
the gas generation rate for this method are shown in table 9. quantity of fault gases released from the insulating oil when a
fault occurs which in turn increase the temperature in the
power transformer. The presence of the fault gases depends on
the temperature or energy that will break the link or relation of
the insulating oil chemical structure. This method uses the
individual gas rather than the calculation of gas ratios for
detecting fault. The significant and proportion of the gases are
called “key gases”. Figure 2 indicate these “key gases” and
relative proportions for the four general fault types [7].
Table 9: L1 limits and gas generation rate
for Duval Triangle Method [8]. vi. Nomograph Method

Once a problem has been determined to exist, to obtain The logarithmic nomograph method was developed by J.
diagnosis, calculate the total accumulated amount of the three O. Church [1]. This method combines the fault gas ratio
Duval Triangle gases (CH4, C2H2, C2H4) and divide each gas concept with the Key Gas threshold value in order to improve
by the total to find the percentage of each gas of the total. Plot the accuracy of fault diagnosis. It was intended to provide
the percentages of the total on the triangle (Figure1) to arrive both a graphic presentation of fault-gas data and the means to
at the diagnosis [8]. interpret its significance. The Nomograph consists of a series
of vertical logarithmic scales representing the concentrations
of the individual gases as shown in Figure 3.

With this method, straight lines are drawn between adjacent


scales to connect the points representing the values of the
individual gas concentration. The slopes of these lines are the
diagnostic criteria for determining the type of fault. The key at
the bottom of the chart between the two axes indicates the
fault type for the two axes. A visual comparison of the slopes
of the line segments with the keys given at the bottom of the
Nomograph is all needed to identify the type of fault. The
position of the lines relative to the concentration scales
provides a means of assessing the severity of the fault.

Figure 1: Duval Triangle transformer fault diagnosis [8]

v. Key Gas Method

Figure 3: The Logarithmic Nomograph

(a) Principal Gas: Ethylene (b) Principal Gas: Carbon Monoxide Each vertical scale has a threshold value labeled with an
arrow. For the slope of a line to be considered significant, at
least one of the two tie-points should lie above a threshold
value. If neither tie-point lies above a threshold value then the
fault indication of that slope is not considered significant.

B. Testing Method

(c) Principal Gas: Hydrogen (d) Principal Gas: Acetylene The testing method should be the same for each DGA
interpretation method in order to compare their accuracy and
Figure 2: Key Gases Diagnosis consistency. Each method diagnosis was grouped according to

47
the faults type code for comparison. This is shown in table 10 Figure 4: Example of Simulink block diagram used for testing.
below.
III. RESULTS

Method F1 F2 F3 F4 F5
Each method was tested against all the 92 cases in the data
Roger Slight Conductor Flashover. PDs Normal set. The percentages of successful prediction and consistency
overheating overheating are calculated using the following formulas:
<150oC Arcing PDs with
Winding tracking
Overheating circulating Continuous RFn
150oC-200oC current sparking. S Fn u 100 (1)
Number of cases of Fn
Overheating Core/tank
¦
n 5
200oC-300oC circulating n 1
S Fn
current. CFn (2)
IEC Thermal fault Thermal fault Discharge of PDs of low Normal Number of fault types
<150oC 300oC-700oC low energy energy density where:
Thermal fault Thermal fault Discharge of PDs of high Fn = type fault code (n=1,2,3,4,5)
150oC-300oC > 700oC high energy energy density
Nomograph Heating Heating and Arcing Arcing, Normal
Discharge heating and (<L1) The results are summarized in table 11. It can be seen that
Arcing and discharge
heating
the Duval Triangle method is the most consistent method
Arcing and followed by the Key Gas, Nomograph, IEC Ratio, Roger
discharge Ratio and lastly the Doernenburg method. Note the low
Doernenburg Thermal Thermal Arcing Corona Normal
decomposition decomposition (< L1) consistency value (<50%) with some of the methods. We also
with very high find that those methods that take into account the limit value
ratio 4
Duval Thermal Thermal fault Low energy PDs Normal
of fault gases before doing diagnosis have better success in
fault <300oC 300oC-700oC discharge (< L1) predicting the normal condition and methods that have no
Mix thermal limit value of faults gases always fail to predict the normal
Thermal fault > High energy and electrical
750oC discharge faults condition. This affects the consistency result.
Key Gas Principal Principal gas: Principal Principal Normal
gas: CH4 and C2H4 gas: C2H2 gas: H2 (< L1)
C2H6 Method Faults Number of Number of % Successful Consistency
Code predictions correct prediction (C)
Table 10: Grouping for faults type code. (P) predictions (S)
(R)
In this paper, the comparison was done using MATLAB. A Roger F1 10 5 50% 45%
program was developed to run the test based on each method F2 13 13 39%
rules in diagnosing the faults. This involved several coding F3 13 12 55%
F4 9 8 57%
and Simulink block diagram to run the test. Figure 4 shows an
F5 4 3 23%
example of block diagrams used as testing method. Generally
IEC F1 6 5 50% 60%
the diagrams consist of three main sections. The first section is F2 26 26 79%
for checking the limit value of the faults gases if applicable. F3 19 18 82%
The second section is for calculating the ratio and finding the F4 9 9 64%
ratio coding if applicable. The last section provides the F5 6 3 23%
diagnosis based on the ratio coding sequence or ratio value or Nomograph F1 15 2 20% 74%
fault gases value. F2 24 23 70%
F3 19 18 82%
F4 20 14 100%
F5 14 13 100%
Doernenburg F1 3 2 20% 40%
F2 15 15 45%
F3 9 8 36%
F4 7 6 43%
F5 8 7 54%
Duval F1 10 10 100% 88%
F2 32 30 91%
F3 26 22 100%
F4 10 7 50%
F5 14 13 100%
Key Gas F1 11 10 100% 78%
F2 46 33 100%
F3 11 10 45%
F4 9 7 50%
F5 13 2 92%
Table 11: Result analysis for each type of faults.

48
one of the expert systems that can be used to diagnose the
In this analysis, we also found that the best methods in faults because of its ability in storing knowledge and using it
predicting fault types F1 and F2 are the Duval Triangle and to make decision [5]. Here, the final diagnosis rules are
Key Gas method. The Duval Triangle method also is the best automatically determined and the membership functions of the
method for predicting fault types F3 and F5. Others than corresponding fuzzy subsets are simultaneously adjusted. This
Duval Triangle, the Nomograph also is the best method in can give better judgment on diagnosis the transformer faults.
predicting F5 and F4.
A Fuzzy Logic controller was developed using MATLAB
Roger IEC Nomograph Doernenburg Duval Key Gas
to implement the Roger Ratio method. The same data set was
Total cases, 92 92 92 92 92 92
TC
used in the testing. It was found that by using this system, the
No 47 26 0 50 0 0 number of cases with no prediction was decreased by 17
predictions, percent. This gives a better accuracy based on total cases
TNP (52% as compared to 45%) and consistency (51% as
Number of 45 66 92 42 92 92
predictions,
compared to 45%). Details of this work on Fuzzy Logic
TP systems will be discussed in another publication.
Correct 41 61 70 38 82 72
predictions, IV. CONCLUSION
TR
Incorrect 4 5 22 4 10 20 As a conclusion, it was found that those methods using
predictions, specific codes in their interpretation are more accurate if they
TW make a prediction. This is because they are more precise in
Accuracy 91% 92% 76% 90% 89% 78% interpreting the data for each case into the code. However,
( predicted
cases), AP whenever the data does not match with the available codes,
Accuracy 45% 66% 76% 41% 89% 78% these methods are not able to give their prediction. This
( total cases), results in their lower level of consistency in predicting the
AT fault and less accurate based on total case prediction. These
Table 12: Comparison of each method accuracies value. apply to the Roger’s and IEC Ratio method.

Besides using consistency to evaluate the performance of In contrast, those methods that use direct value of fault
each method, the accuracy of each method has also been gases in their interpretation give higher consistency and same
calculated. Here, the accuracy is divided into two categories: value of accuracies as they attempt to provide predictions for
the accuracy (Ap) when considering only the predicted cases all cases. But as they have all the interpretations, the
(Tp) and the accuracy (AT) based on the total number of cases prediction is likely to be incorrect for certain cases. This is
(TC). Their formulas are: because, the diagnosis is generally based on one specific value
of faults gases as indicator, and this tends to widen the range
TR of one type of faults.
AP u 100 (3)
TP
TR VIII. REFERENCES
AT u 100 (4)
TC
1. DiGiorgio, J.B. (2005) Dissolved Gas Analysis of Mineral Oil
Insulating Fluids. DGA Expert System: A Leader in Quality,
Table 12 shows the accuracy results. The calculation of Value and Experience 1, 1-17
accuracy based on the predicted cases show that all methods 2. Chu, D. and A. Lux, On-line monitoring of power transformers
and components: a review of key parameters. Electrical Insulation
have accuracy more than 70 percent. The most accurate is the Conference and Electrical Manufacturing & Coil Winding
IEC Ratio method followed by the Roger Ratio, Doernenburg, Conference, 1999. Proceedings, 1999: p. 669-675.
Duval Triangle, Nomograph and Key Gas method. As seen in 3. Siva Sarma, D.V.S.S. and G.N.S. Kalyani, ANN Approach for
Condition Monitoring of Power Transformers using DGA. 2004
table 12, the methods that used specific code in their diagnosis IEEE Region 10 Conference, TENCON 2004., 2004. C: p. 444-
have high accuracy (>90%). On the other hand, methods that 447.
use direct interpretation based on each value of fault gases are 4. Yang, F. and Z. Liang, Comprehensive method detecting the status
of the transformer based on the artificial intelligence. 2004
less accurate. However, the accuracy based on the total International Conference on Power System Technology, 2004.
number of cases shows different trend. Because of the high PowerCon 2004. , 2004. 2: p. 1638-1643.
value of cases with no prediction, the accuracy drops 5. Hongzhong, M., et al., Diagnosis of power transformer faults on
significantly (<70%) for methods that used specific codes in fuzzy three-ratio method. The 7th International Power Engineering
Conference, 2005. IPEC 2005., 2005.
the diagnosis. 6. Wang, M., A.J. Vandermaar, and K.D. Srivastava, Review of
Condition Assessment of Power Transformers In Service, in IEEE
A possibility to improve the accuracy of DGA diagnosis is Electrical Insulation Magazine. 2002. p. 12-25.
7. C57.104.1991, I., IEEE Guide for Interpretation of Gases
via the use of expert systems [9]. Fuzzy Logic is known as Generated in Oil-Immersed Transformer, I. The Institute of

49
Electrical and Electronic Engineers, Editor. 1992, The Institute of
Electrical and Electronic Engineers, Inc p. 27.
8. FIST3-31, Facilities Instructions, Standards and Techniques
Volume 3-31 Transformer Diagnostics. 2003, Bureu of
Reclamation Hydroelectric Research and Technical Services Group
Denver. p. 5-13.
9. Q.Su, et al., A Fuzzy Dissolved Gas Analysis Method for The
Diagnosis of Multiple Incipient Faults in a Transformer. IEEE
Transaction On Power System, 2000. 15(2): p. 593-597.

50

You might also like