American Journal of Electrical Power and Energy Systems

2021; 10(5): 74-81

http://www.sciencepublishinggroup.com/j/epes
doi: 10.11648/j.epes.20211005.11
ISSN: 2326-912X (Print); ISSN: 2326-9200 (Online)

An Advanced Fuzzy Logic Based Method for Power

Transformers Assessment
Si Huy Cuong Nguyen1, *, Thanh Phong Mai2
1
Technical Department, Electric Testing Company, Ho Chi Minh City, Vietnam
2
Academic Affairs Department, Long An College, Tan An City, Vietnam

Email address:
*
Corresponding author

To cite this article:

Si Huy Cuong Nguyen, Thanh Phong Mai. An Advanced Fuzzy Logic Based Method for Power Transformers Assessment. American Journal
of Electrical Power and Energy Systems. Vol. 10, No. 5, 2021, pp. 74-81. doi: 10.11648/j.epes.20211005.11

Received: August 28, 2021; Accepted: September 11, 2021; Published: September 26, 2021

Abstract: Dissolved Gas Analysis is an effective method for detecting faulty power transformers in their early stages.
However, technical interpretation of results can be complex and highly dependent on the experience of experts. This paper
presents an attempt to detect power transformer incipient fault via gas concentrations obtained from oil sampling and Dissolved
Gas Analysis. The proposed method uses a sophisticated fuzzy logic system to perform fault type classification. Ratios and
relative percentages of 5 key gases (Hydrogen, Methane, Ethane, Ethylene, and Acetylene) are taken as input variables, then the
fuzzy system will try to generate an output vector that indicates six basic fault types, including partial, low, and high energy
discharges as well as three ranges of thermal fault. This method can be easily implemented in any environment that supports
basic mathematical operators. To demonstrate how the proposed fuzzy logic method works, the authors developed an offline
MATLAB script and an online web-based application that can provide multiple assessments by various methods simultaneously.
The set of membership functions and fuzzy rules presented in this paper allows the detection of multiple faults at once.
Performance tests on many actual data sets show that the proposed method achieves better accuracy than the traditional ratio
codes, even on a par with state-of-the-art graphical-based tools such as the Duval triangle or pentagon.
Keywords: Power Transformer, Incipient Fault Detection, Dissolved Gas Analysis, Fuzzy Logic

techniques must be adopted to improve the reliability of the

1. Introduction equipment and to avoid any catastrophic failure. Among
In-service power transformers are frequently subjected to existing techniques, dissolved gas in oil analysis (DGA) is a
both potential internal defects and external stresses. Thermal powerful method to detect power transformer incipient faults
stress caused by local overheating accelerates the aging [1-3].
process of oil and paper insulation. Electrical and mechanical Conventional DGA interpretation methods such as key gas
stresses from external sources such as lightning strikes or inspection or gas ratios based methods [4-7] have been
short-circuit current greatly contribute to reducing the widely used, but they still have some limits and sometimes
remaining lifetime of power transformers. Those stresses cannot give a proper diagnosis. Recently, the introduction of
cause material decomposition and generate dissolved the Duval triangles and pentagons [8-10] solved the problem
combustible gases in insulating oil, some of which are oxides of unidentified faults. However, the analysis is not always
of carbon, hydrogen, and hydrocarbons. Internal defects straightforward as there may be more than one fault present
generate a particular amount of characteristic gases dissolved at the same time. Precise DGA interpretation is still a hot
in insulating oil that can be used for early fault identification. topic in the power transformer fault diagnosis and condition
Proactive detection of faults helps minimize the risk of assessment research area.
undesirable outage of power transformers from the power In this paper, a fuzzy logic-based method is developed to
system network. Effective monitoring and diagnostic enhance the quality of existing DGA interpretation tools.
While other methods can only detect a single fault, the fuzzy
75 Si Huy Cuong Nguyen and Thanh Phong Mai: An Advanced Fuzzy Logic Based Method for Power Transformers Assessment

method is feasible to address the classification of transformer popular until lately. There are several variations such as the
faults in the case where multiple faults occur at once. Since Dornenburg ratio, the Rogers ratio, and the three gas ratio
fuzzy logic has an advantage in processing unclear states, it methods [3-7].
is also possible to implement the criticality alert in the fuzzy The Rogers ratio method [4] considers two of the four ratios
diagnostic system. At the end of this paper, the authors would CH4/H2, C2H2/C2H4, C2H4/C2H6, and C2H6/CH4. However,
like to compare the efficiency of the fuzzy logic-based later studies showed the ratio of C2H6/CH4 did not correlate
approach to other conventional methods by surveying real well with the faults, and thus it was removed in recent studies.
cases in Vietnam. The three ratio method is now recommended by both the IEEE
and the IEC standards [3, 5]. The interpretation guide of this
2. Literature Review method is shown in tables 1 and 2, in which there are three
ratio codes for each ratio and six fault types [6].
Interpretation of DGA results is not always straightforward, Dividing one small value of a fault gas by another small
as there are several possible causes of the presence of gas in a value of another fault gas will give a significant ratio, but the
transformer. Some of those are related to real fault conditions, magnitudes of the fault gases in such cases are too small. For
others are related to more benign conditions such as stray that reason, ratio methods are only applicable when a
gassing. There is no direct and infallible method using DGA to significant amount of the gas is present; otherwise, they may
obtain an exact evaluation of a transformer’s condition. lead to misdiagnosis. The common weakness of ratio-based
However, it is necessary to have a reliable DGA assessment methods is that they sometimes are not capable of giving a
tool to detect any possible fault that might occur inside a result or may yield an incorrect one in others. Therefore,
power transformer. In this section, the authors wish to provide some researchers attempt to add or modify rules to achieve
a brief review of popular DGA interpretation techniques and better accuracy [11].
point out the reason why a fuzzy logic-based method can
provide a better solution. Table 2. The IEC ratio codes [6].

States
2.1. Key Gas Inspection Ratio
0 1 2
r1 = C2H2/C2H4 < 0.1 0.1 – 3 >3
The key gas method applies some basic rules for finding the r2 = CH4/H2 0.1 - 1 < 0.1 >1
fault pattern based on dominant gases. Hydrogen (H2) is r3 = C2H4/C2H6 <1 1–3 >3
primarily generated from corona partial discharge; Acetylene
(C2H2) is created from arcing in oil or paper at very high Table 3. Fault classification by using the IEC ratio codes [6].
temperatures. Overheating and thermal faults give rise to Fault
Methane (CH4), Ethane (C2H6), and Ethylene (C2H4) as well Fault r1 r2 r3
Code
as Carbon Monoxide (CO), and Carbon Dioxide (CO2) if the Normal N 0 0 0
fault is related to solid insulation decomposition. By Partial discharge PD 0 or 1 1 0
determining which gasses are dominating, one can speculate Low energy discharge D1 1 or 2 0 1 or 2
High energy discharge D2 1 0 2
the existence of internal fault. Thermal fault with t < 150°C 0 0 1
There is one big challenge in using this method, as it T1
Thermal fault with 150°C < t < 300°C 0 2 0
requires the users’ experience. Furthermore, software Thermal fault with 300°C < t < 700°C T2 0 2 1
implementation of the key gas method seems to be a challenge. Thermal fault with t > 700°C T3 0 2 2
Inconclusive or wrong fault identification occurs regularly
even with sophisticated key gas rules. The reason for this 2.3. Graphical Methods
problem is that it is not always clear which is the dominant gas, Several graphical methods have been developed to
or the main gas formed may not be reliable enough for fault overcome the problem of having unidentified cases. Two of
identification. However, observing key gases is essential for the most well-known graphical-based methods are the Duval
building an advanced interpretation system based on fuzzy triangles and pentagons [8-10]. Other approaches such as the
logic or artificial intelligence. Mansour diagnostic pentagon [12] or the heptagon developed
Table 1. Fault identification based on key gasses.
by Gouda et al. [13] were introduced recently. In this section,
the authors shall only briefly summarize the Duval triangles
Fault Key gasses and pentagons, as they are used for comparison later in the
Partial discharge H2
research.
Arcing C2H2
Thermal fault (oil) CH4, C2H4, C2H6
2.3.1. Duval Triangles
Thermal fault (paper) CO, CO2
The original Duval triangle [8] uses a set of three
2.2. Gas Ratio Methods characteristics gases: CH4, C2H4, and C2H2. The sides of the
triangle are expressed in triangular coordinates (x, y, z),
Gas ratio-based methods take correlation of ratio between where x, y, z are the relative percentage of CH4, C2H4, and
some pairs of fault gas concentrations with certain fault types. C2H2, respectively.
These methods were introduced in the 1970s and remain
 American Journal of Electrical Power and Energy Systems 2021; 10(5): 74-81 76

This method allows the identification of the six basic types vertices of this pentagon correspond to the maximum relative
of faults mentioned in the last section (PD, D1, D2, T1, T2, concentration of 40%. Inside this pentagon, the zones are
and T3), in addition to mixtures of electrical/ thermal faults in corresponding to the basic types of faults just like that of the
zone DT. Those regions are established through empirical Duval triangle, as well as a stray gassing zone (S). The
inspection of DGA results from a specific liquid type and the percentage of each gas is marked on the appropriate axis
observed equipment gassing source. A fault is identified based drawn from the center of the pentagon to one of the vertices.
on which region the corresponding point (x, y, z) lies on. These points are then connected to form a small polygon in
To this day, there are several versions of the Duval triangle, which the centroid always lies inside the Duval pentagon.
in which the first, fourth and fifth are exclusive to mineral oil. The position of the designated centroid points to one of the
The fourth and fifth triangles take a different set of gases and seven fault zones as one can observe in figure 1.
are only used for thermal fault inspection [9]. The first Both Duval methods use relative gas percentages instead of
triangle, together with the pentagon counterpart, as depicted ratios and thus avoid the problem of unidentified cases. In
in figure 1, are widely used for the diagnosis of high voltage contrast, because the triangles and pentagons always give a
power transformers. diagnostic, they should only be used to identify a fault when a
sufficient amount of combustible gas exists. Moreover, due to
2.3.2. Duval Pentagon the nature of graphical-based interpretation methods, they
The Duval pentagon [10] uses the percentages of five cannot verify the existence of multiple faults at a time.
gases (H2, CH4, C2H6, C2H4, and C2H2) to their sum. The

Figure 1. The Duval triangle (left) and pentagon (right).

curves calculated by the minimum combinations of (1) and

3. Proposed Method (2). In these equations, a, b, c, and d are parameters that
affect the shape and boundary of the curve. They represent
When one or more than one fault occurs in a transformer, the boundary conditions so that the membership functions
multiple key gases with different concentrations exist and cause translate input values into intermittent fuzzy states.
0,
the ratio codes to overlap. Because of that, the relationship

2 ,
between various gases becomes too complicated and may not

; ,
match the predefined values. In multiple-fault conditions, gases

1 2 ,
from different faults are mixed, resulting in confusing ratios (1)

1,
between gas components. This problem can be overcome with
the aid of more sophisticated analysis methods such as the fuzzy
1,
logic presented in this section. The proposed method is called

1 2 ,
!
Fuzzy Ratio and Percentage (FRP in short).

; , !

2 ,
! !
3.1. Fuzzy Logic Based Method (2)
!
0,
In the IEC ratio-based interpretation methods, the ratio
codes 0, 1, or 2 can either be True or False, but not anything
in between. The gas ratio boundary should be fuzzy,
Conditional statements given by the rules in table 2 are
especially when more than one type of fault exists. Between
combinations of conventional logics “AND” and “OR”,
different kinds of faults, the codes should not change sharply
which can be converted into mathematical terms by the
across their boundaries. Therefore, in the proposed fuzzy
“MIN” and “MAX” operators. For example, the last rule is
logic-based method, input variables are transformed into a set
[r1 is 0] AND [r2 is 2] AND [r3 is 2]; this statement is
of states via fuzzy functions. Membership functions in the
translated to min[µs(r1), µz(r2), µz(r3)]. When a condition is
uppermost region are S-shape functions governed by (1),
fulfilled, either fully or partially, certain rules will be
while counterparts in the lowest region are Z-shape curves
triggered. If the output is defined by a vector in which each
represented by (2). The middle region is occupied by π-shape
77 Si Huy Cuong Nguyen and Thanh Phong Mai: An Advanced Fuzzy Logic Based Method for Power Transformers Assessment

index corresponds to a fault code, then those rules will give vector consists of 7 indexes correspond to six basic types of
values in the range between 0 and 1 to the indexes based on faults (PD, D1, D2, T1, T2, T3) and a normal state (N).
the trigger condition. This method allows the detection of
multiple faults at once, a feature that neither the conventional Table 4. Gas ratio membership functions.
gas ratio methods nor the Duval methods had. Membership function Type a b c d
µr10(r1) Z-shape - - 0.3 0.7
3.2. Fuzzy Rules Based on Gas Ratio µr11(r1) Π-shape 0.3 0.7 2.8 3.2
µr12(r1) Π-shape 2.8 3.2 9.8 10.2
Initially, membership functions and fuzzy rules were µr13(r1) S-shape 9.8 10.2 - -
designed based on the IEC gas ratio method. However, this µr20(r2) Π-shape 0 0.4 0.8 1.2
approach turned out to be the same as the conventional µr21(r2) Z-shape - - 0 0.4
µr22(r2) Π-shape 0.8 1.2 2.8 3.2
method; that means the weakness of having unidentified
µr23(r2) S-shape 2.8 3.2 - -
cases still exists. After inspecting various samples, the µr30(r3) Z-shape - - 0.8 1.2
authors developed an improved set of ratio codes through µr31(r3) Π-shape 0.8 1.2 2.8 3.2
membership functions shown in table 4. The corresponding µr32(r3) S-shape 2.8 3.2 - -
fuzzy rules are described in table 5, in which Fr is the output

Table 5. Gas ratios fuzzy rules.

Fault Rule
Normal Fr(0) = min[µr10(r1), µr20(r2), µr30(r3)]
PD Fr(1) = min[max[µr10(r1), µr11(r1)], µr21(r2)]
Fr(2) = max[d11, d12, d13, d14]
Where:
d11 = min[µr11(r1), µr21(r2), µr30(r3)]
D1
d12 = min[max[µr10(r1), µr11(r1)], µr21(r2), µr31(r3)]
d13 = min[µr11(r1), max[µr20(r2), µr22(r2), µr23(r2)], max[µr30(r3), µr31(r3)]]
d14 = min[µr12(r1), max[µr20(r2), µr21(r2), µr22(r2)], µr31(r3)]
Fr(3) = max[d21, d22, d23, d24]
Where:
d21 = min[max[µr10(r1), µr11(r1)], µr21(r2), µr32(r3)]
D2
d22 = min[µr11(r1), max[µr20(r2), µr22(r2), µr23(r2)], µr32(r3)]
d23 = min[µr12(r1), max[µr20(r2), µr21(r2), µr22(r2)], max[µr30(r3), µr32(r3)]]
d24 = µr13(r1)
Fr(4) = max[t11, t12, t13]
Where:
T1 t11 = min[µr10(r1), µr20(r2), µr31(r3)]
t12 = min[µr10(r1), max[µr22(r2), µr23(r2)], µr30(r3)]
t13 = min[µr12(r1), µr23(r2), µr30(r3)]
Fr(5) = max[t21, t22]
Where,
T2
t21 = min[µr10(r1), max[µr22(r2), µr23(r2)], µr31(r3)]
t22 = min[µr12(r1), µr23(r2), µr31(r3)]
Fr(6) = max[t31, t32, t33]
Where:
T3 t31 = min[µr10(r1), µr20(r2), µr32(r3)]
t32 = min[µr10(r1), max[µr22(r2), µr23(r2)], µr32(r3)]
t33 = min[µr12(r1), µr23(r2), µr32(r3)]

The relative percentages of H2, CH4, C2H6, C2H4, C2H2 in a

3.3. Fuzzy Rules Based on Gas Percentage sample are denoted as p1, p2, p3, p4, p5. They are described in
Performance test on the gas ratio-based fuzzy system 3 levels: “Low”, “Medium, and “High” by the Z, π, and S
shows an improvement in diagnostic accuracy. However, in functions, just like their gas ratio counterparts. The boundary
some partial discharge cases, the Hydrogen contents are values of those functions the gas percentage fuzzy rules are
dominant, while other gases are insignificant. In those cases, mathematically described in the next two tables.
the “Normal” rule is triggered instead of “PD”, regardless of 3.4. Output Calculation
high H2 concentrations. It is not uncommon to find increased
levels of H2 or C2H4 when C2H2 is detected, leading to a By observation during performance tests, the authors
fuzzy boundary between low and high energy discharge realized that the gas ratio fuzzy system was more sensitive to
faults. In that situation, the gas percentage method may be thermal faults, while the gas percentage fuzzy system was
more effective and therefore, should be adopted to support more reliable in detecting partial discharge and low energy
the fuzzy gas ratio. discharge faults. Therefore, the total fault vector should be
 American Journal of Electrical Power and Energy Systems 2021; 10(5): 74-81 78

calculated by taking the normalized average of the gas ratio The ratios are translated to fuzzy states (illustrated in
and the gas percentage fuzzy outputs by (3). figure 2):

"#$# &' ∑, + &1 ∑,

() (2
1) [µr10(r1), µr11(r1), µr12(r1), µr13(r1)] = [0,0,1,0];
%
-./ () + -./ (2 +
(3) 2) [µr20(r2), µr21(r2), µr22(r2), µr23(r2)] = [0.97,0.03,0,0];
3) [µr30(r3), µr31(r3), µr32(r3) = [1,0,0].
Table 6. Gas percentage membership functions. This combination results in Fr(3) equal to 0.97 and
triggers the D2 fault rule. In this case, the fuzzy gas ratio
Membership function Type a b c d
rules in table 5 give a ratio diagnosis vector Fr = [0, 0, 0,
µL(gas %) Z-shape - - 0 10
µM(gas %) Π-shape 0 10 15 25
0.97, 0, 0, 0]. Similarly, the gas percentages are p1 = 33.7%,
µH(gas %) S-shape 15 25 - -
p2 = 11.76%, p3 = 8.02%, p4 = 5.88%, p5 = 40.64%. They
are translated to:
Table 7. Gas percentage fuzzy rules. 1) [µL(p1), µM(p2), µH(p1)] = [0,0,1];
2) [µL(p2), µM(p2), µH(p2)] = [0,1,0];
Fault Rule
3) [µL(p3), µM(p3), µH(p3) = [0.08,0.92,0];
Normal Fp(0) = min[µL(p1), µL(p2), µL(p3), µL(p4), µL(p5)]
4) [µL(p4), µM(p4), µH(p4) = [0.08,0.66,0];
PD Fp(1) = min[µH(p1), max[µL(p3), µM(p3)], µL(p4), µL(p5)]
5) [µL(p5), µM(p5), µH(p5)] = [0,0,1].
D1 Fp(2) = min[max[µM(p1), µH(p1)], µM(p5)]
The condition of D2 is fulfilled and thus Fp(3) equal to 1.
D2 Fp(3) = µH(p5)
T1 Fp(4) = min[µH(p3), µL(p4), µL(p5)]
The gas percentage laws in table 7 result in Fp = [0, 0, 0, 1, 0,
T2 Fp(5) = min[max[µM(p4), µH(p4)], µL(p5)]
0, 0]. By using (3), the total output vector is [0, 0, 0, 1, 0, 0,
T3 Fp(6) = min[µL(p3), µH(p4), µL(p5)] 0], which indicates occurrence of high energy discharge (D2)
with 100% certainty.
* p1, p2, p3, p4, p5 are the percentages of H2, CH4, C2H6, C2H4 and C2H2

4. Results and Discussion

The authors built a script in the MATLAB software
environment to perform DGA interpretations of several
datasets. The code consists of multiple methods, including
Rogers ratio, IEC ratio, Duval triangle and pentagon, and the
proposed fuzzy logic system. Results from all of those
methods are compared with one another to evaluate their
efficiency in incipient fault classification.
4.1. Cases Study

To test the performance of the proposed Fuzzy Ratio and

Percentage method, the dataset obtained from [14] was used.
There are 20 samples described in table 9; results obtained
from multiple analyses are also compared. The proposed
method outperforms traditional ratio codes in fault
diagnosis capability. With this dataset, the proposed method
generally agrees with the Duval triangle and pentagon, even
achieves better accuracy in the tricky cases of partial
discharge.
Another sample dataset was obtained from the Long An
Power Company in Vietnam to investigate the performance
of the proposed method. The performances of multiple
methods are compared in table 9. One can observe that with
such low gas concentrations, fault classification, in this
case, would be tricky. With this dataset, all method
generally agrees with one another. There is one tricky case
Figure 2. Fuzzy gas ratio membership functions of the given example. with sample number 4, which has dominant H2 and CH4
contents. This is an obvious sign of partial discharge.
This example illustrates in detail how the proposed However, except for the proposed method, none of the
method works. Consider a case where gas contents are: H2 others can classify this fault.
= 63 ppm, CH4 = 22 ppm, C2H6 = 15 ppm, C2H4 = 11 ppm Multiple tests on different datasets [11, 15] were also
and C2H2 = 76 ppm. The gas ratios are r1 = 6.9, r2 = 0.35, performed but not fully show in this paper. In general, the
and r3 = 0.73. This combination results in a ratio code of proposed method generated highly reliable conclusions that
“200” that does not belong to the original IEC guideline. agreed well with actual faults.
79 Si Huy Cuong Nguyen and Thanh Phong Mai: An Advanced Fuzzy Logic Based Method for Power Transformers Assessment

Table 8. Dataset used for performance check [14] (gas contents in ppm).

Known IEC Rogers Duval triangle Duval pentagon

Sample H2 CH4 C2H6 C2H4 C2H2 Proposed method
fault [14] [5] [4] [8] [10]
T2: 84%
1 200 700 250 740 1 T2 T2 - T3 T3
T3: 16%
D1: 38%
2 300 490 180 360 95 T2 - T2 DT T3
T2: 62%
3 56 61 75 32 31 D1 - - D2 T1 D1: 100%
N: 46%
4 33 26 6 5.3 0.2 N N N T1 T1
T2: 44%
5 176 205.9 47.7 75.7 68.7 D1 - - D1 T1 D1: 100%
T2: 37%
6 70.4 69.5 28.9 241.2 10.4 T3 - - T3 T3
T3: 58%
PD: 21%
7 162 35 5.6 30 44 D2 D2 D2 D2 D2 D1: 49%
D2: 30%
8 345 112.25 27.5 51.5 58.75 D1 D1 D2 D2 D1 D1: 96%
9 181 262 210 528 0 T2 T2 T2 T3 T3 T2: 100%
T2: 44%
10 172.9 334.1 172.9 812.5 37.7 T3 T3 T3 T3 T3
T3: 50%
11 2587.2 7.882 4.704 1.4 0 PD PD PD T1 S PD: 100%
D1: 50%
12 1678 652.9 80.7 1005.9 419.1 D2 D2 - DT D2
T3: 41%
T2: 39%
13 206 198.9 74 612.7 15.1 T3 - - T3 T3
T3: 59%
N: 31%
14 180 175 75 50 4 T1 N N T2 T1 T1: 19%
T2: 49%
D1: 49%
15 34.45 21.92 3.19 44.96 19.62 D2 D2 - DT D2 D2: 12%
T3: 38%
16 51.2 37.6 5.1 52.8 51.6 D2 D2 D2 D2 D2 D2: 100%
PD: 19%
17 106 24 4 28 37 D2 D2 D2 D2 D2 D1: 37%
D2: 44%
18 180.85 0.574 0.234 0.188 0 PD PD PD T2 PD PD: 100%
19 27 90 42 63 0.2 T2 T2 - T2 T2 T2: 100%
T2: 21%
20 138.8 52.2 6.77 62.8 9.55 D2 D2 - T3 D2
T3: 71%

Table 9. Dataset from Long An PC (gas contents in ppm).

Duval Duval Proposed

Sample H2 CH4 C2H6 C2H4 C2H2 Known fault IEC [5] Rogers [4]
triangle [8] pentagon [10] method
T3 (30%)
1 23.6 12.4 3.8 50.9 0 T3 N/A T2 T3 T3
T3 (70%)
2 5.6 32.4 10.1 13.1 0 T2 T2 N/A T2 T2 T2 (100%)
3 43.3 50 8.9 11.2 0 T2 T2 T2 T1 T1 T2 (99%)
N (50%)
4 170.7 68.9 8.4 5.7 0 PD N/A N T1 S
PD (44%)
T2 (69%)
5 4.7 14.3 2 6 0 T2 T3 T2 T2 T2
T3 (31%)
T2 (90%)
6 5.7 14.7 2.1 5.8 0 T2 T2 T2 T2 T2
T3 (10%)
T2 (42%)
7 12.1 11.3 3.6 24.7 0 T3 N/A T2 T3 T3
T3 (58%)

Except for the Duval pentagon, none of the others can reach 50%
4.2. Inspection on a Larger Dataset accuracy in this dataset. A noteworthy feature of the proposed
The authors collected 240 samples from local utilities in the method is that it is more sensitive to partial discharge faults
Southern region of Vietnam and performed multiple tests. The than other interpretation methods.
overall results are summarized in Table 10. In general, the An online demo version of the method is also available for use.
proposed method generated highly reliable conclusions that The IEC ratio, Roger ratio, Duval triangle, and pentagon are also
agreed well with actual faults, with an accuracy of over 80%. included in this demo version; all of them are implemented using
Javascript and HTML. However, the algorithm used for
 American Journal of Electrical Power and Energy Systems 2021; 10(5): 74-81 80

developing graphical-based methods in this online version is not better choice would be using a more mathematical-oriented
very accurate when the point lies on the edge of a fault zone. A environment such as MATLAB or OCTAVE.

Table 10. Comparison of various methods over a large transformer fleet.

Number of correct diagnosis

Fault Actual case
IEC [5] Rogers [4] Duval triangle [8] Duval pentagon [10] Proposed method
PD 23 7 3 7 6 20
D1 32 6 1 22 20 24
D2 63 6 19 15 18 44
T1 76 44 68 35 74 65
T2 18 14 1 4 5 14
T3 28 20 5 28 28 28
Total 240 97 97 111 151 195
Percentage 40% 40% 46% 63% 81%

C57.104-2019 (Revision of IEEE Std C57.104-2008), 2019,

doi: 10.1109/IEEESTD.2019.8890040.
5. Conclusion
[4] R. R. Rogers, "IEEE and IEC Codes to Interpret Incipient
In this paper, a new algorithm to detect potential faults Faults in Transformers, Using Gas in Oil Analysis," in IEEE
inside power transformers was introduced. The diagnostic Transactions on Electrical Insulation, vol. EI-13, no. 5, pp.
system is made based on fuzzy logic that process ratios and 349-354, Oct. 1978, doi: 10.1109/TEI.1978.298141.
percentages of key gases obtain from DGA results. In most [5] “Mineral oil-filled electrical equipment in service - Guidance
cases presented throughout the paper, the use of fuzzy logic on the interpretation of dissolved and free gases analysis,” in
overcomes the limitations of traditional gas ratios based IEC 60599.
interpretation with high accuracy in fault diagnosis. Since the [6] S. Chakravorti, D. Dey, and B. Chatterjee, Recent trends in the
method allows the detection of multiple faults in one sample, condition monitoring of transformers theory, implementation
it can provide comprehensive insights into the conditions of a and analysis. London: Springer, 2013.
power transformer. That feature might provide additional
[7] M. Duval and A. dePabla, "Interpretation of gas-in-oil analysis
information and help condition assessment be more reliable. using new IEC publication 60599 and IEC TC 10 databases,"
The proposed diagnosis algorithm is not only efficient but also in IEEE Electrical Insulation Magazine, vol. 17, no. 2, pp.
very simple to implement. In short, the research contributes a 31-41, March-April 2001, doi: 10.1109/57.917529.
useful tool for the condition assessment of power
[8] M. Duval, "A review of faults detectable by gas-in-oil analysis
transformers. in transformers," in IEEE Electrical Insulation Magazine, vol.
Nowadays, alarm concentration values are set by 18, no. 3, pp. 8-17, May-June 2002, doi:
independent experts, based on previous experience with 10.1109/MEI.2002.1014963.
equipment with similar characteristics [5]. Future research on
[9] M. Duval, "The Duval triangle for load tap changers,
this topic should examine the fault criticality to determine the non-mineral oils and low temperature faults in transformers,"
normality percentages and critical concentrations. Again, in IEEE Electrical Insulation Magazine, vol. 24, no. 6, pp.
using fuzzy logic would be a suitable approach to this task, 22-29, November-December 2008, doi:
since gas concentrations can vary from sample to sample. 10.1109/MEI.2008.4665347.
[10] M. Duval and L. Lamarre, "The duval pentagon-a new
Acknowledgements complementary tool for the interpretation of dissolved gas
analysis in transformers," in IEEE Electrical Insulation
The authors would like to express gratitude toward the Oil Magazine, vol. 30, no. 6, pp. 9-12, November-December 2014,
Testing Lab at the Southern Electrical Testing Company for doi: 10.1109/MEI.2014.6943428.
the data they provided. [11] B. M. Taha, S. S. M. Ghoneim and A. S. A. Duaywah,
"Refining DGA methods of IEC Code and Rogers four ratios
for transformer fault diagnosis," 2016 IEEE Power and Energy
Society General Meeting (PESGM), 2016, pp. 1-5, doi:
References 10.1109/PESGM.2016.7741157.
[1] "IEEE Guide for Evaluation and Reconditioning of Liquid [12] D. A. Mansour, "Development of a new graphical technique
Immersed Power Transformers," in IEEE Std C57.140-2017 for dissolved gas analysis in power transformers based on the
(Revision of IEEE Std C57.140-2006), 2017, doi: five combustible gases," in IEEE Transactions on Dielectrics
10.1109/IEEESTD.2017.8106924. and Electrical Insulation, vol. 22, no. 5, pp. 2507-2512,
[2] "IEEE Guide for Diagnostic Field Testing of Fluid-Filled October 2015, doi: 10.1109/TDEI.2015.004999.
Power Transformers, Regulators, and Reactors," in IEEE Std [13] E. Gouda, S. H. El-Hoshy, and H. H. El-Tamaly, "Proposed
C57.152-2013, 2013, doi: 10.1109/IEEESTD.2013.6544533. heptagon graph for DGA interpretation of oil transformers,"
[3] "IEEE Guide for the Interpretation of Gases Generated in IET Generation, Transmission & Distribution, vol. 12, no. 2,
Mineral Oil-Immersed Transformers," in IEEE Std pp. 490–498, 2018.
81 Si Huy Cuong Nguyen and Thanh Phong Mai: An Advanced Fuzzy Logic Based Method for Power Transformers Assessment

[14] Hongzhong Ma, Zheng Li, P. Ju, Jingdong Han and Limin [15] Mang-Hui Wang, "A novel extension method for transformer
Zhang, "Diagnosis of power transformer faults on fuzzy fault diagnosis," in IEEE Transactions on Power Delivery, vol.
three-ratio method," 2005 International Power Engineering 18, no. 1, pp. 164-169, Jan. 2003, doi:
Conference, 2005, pp. 1-456, doi: 10.1109/TPWRD.2002.803838.
10.1109/IPEC.2005.206897.