Conventional Dissolved Gases Analysis in Power Transformers: Review

Abstract

Transformers insulated with mineral oil tend to form gases, which might be caused by system faults or extended use. Based on an evaluation of the main failure analysis techniques using combustible gases, this study reviewed the conventional techniques for Dissolved Gas Analysis (DGA), present in the norms IEC 60599 and IEEE Std C57.104, and their failure analysis tendency. Furthermore, to illustrate distinct technique performances and failures, the performance of the following techniques was analyzed based on the IEC TC10 database: Dornenburg, Duval Triangle, Duval Pentagon, IEC ratio method, Key Gas, and Rogers. The objective of this work was to present relevant information to support students and professionals who work in failure analysis and/or assist in the development of new tools in the DGA field.

1. Introduction

The insulation of power transformers usually consists of solid and liquid insulation. The solid insulation is made of Kraft paper, manila, and pressboard [1,2]. Ester liquids are used in some types of power transformers, such as wind turbines or solar; however, mineral oil is still by far the liquid most widely used, and their liquid insulation is achieved using mineral oil.

The mineral oil insulation used in electrical equipment is extracted from petroleum, which is a hydrocarbon compound. Their composition and characteristics depend on the environment where the petroleum was extracted, which might be paraffinic or naphthenic [3].

Laboratory studies have shown that these elements, under normal operating conditions, present the formation of gaseous compounds of low molecular weight hydrocarbons, as well as CO and CO₂. This formation is continuous in low volume during the equipment lifetime; however, in failures or other abnormal conditions, an imbalance in the chemical formation and an increase in the amount of these gases can be observed. Hotspots with temperatures above 300 °C predominantly result in the formation of unsaturated hydrocarbons. Furthermore, if the cellulosic system is reached, there is also the formation of abnormal concentrations of CO and CO₂.

The analysis of dissolved gases (DGA) in power transformers is registered as the analysis of the formation of hydrocarbon gases, and the first study regarding this topic was published in February 1919 in “The Electric Journal” [4]. In this study, the first observations regarding the pattern of gas formation were reported, and the gases formed were composed of CO₂ (1.17%), Heavy Hydrocarbons (4.86%), O₂ (1.36%), CO (19.21%), H₂ (59.10%), N₂ (10.10%), and CH₄ (4.2%). The author highlighted the high proportion of H₂ and the influence of temperature on the growth and amount of dissolved gases.

In 1928, based on knowledge of the problem, the gas formation in equipment failure, the Buchholz relay, was developed. Figure 1 illustrates the functions and actuation mechanism of the relay, the purpose of which was to protect the equipment through its action with the passage of gases through the pipeline interconnected to the equipment with an expansion tank [4].

Analyses at that time were already possible in samples of gases and liquids, and in 1959, a field chromatography analyzer was developed by H.H. Wagner [4]; Figure 2 shows a piece of equipment in the field of chromatography.

From the 1960s onwards, gas chromatography (GC) saw its first applications in identifying dissolved gases in transformers from electrical faults.

Chromatographic analysis of gases dissolved in oil is carried out in three steps. First, the oil sample is collected, which can be collected with flexible aluminum bottles or using a syringe (Figure 3), and the collected gases can be stored in the Buchholz relay. Second, the gases are extracted from the oil sample [8,9]. Third, the gases extracted from the sample are analyzed using the gas chromatograph [10], which consists of separating the different gases from the mixture, and then identifying and quantifying them.

The advances in dissolved gas analysis equipment in regard to transformers make it possible to know the origin of the gases and understand the type of faults from the different types and levels of gases formed in each fault.

In 1978, Rogers [11] published a graph (Figure 4) showing the formation of dissolved combustible gases as the temperature increased. Figure 4 demonstrates the relationship between gas formation and temperature.

According to [13], the following individual gas components can be identified and determined: hydrogen (H₂), nitrogen (N₂), carbon monoxide (CO), carbon dioxide (CO₂), methane (CH₄), ethane (C₂H₆), ethylene (C₂H₄), acetylene (C₂H₂), propane (C₃H₈), and propylene (C₃H₆). With the exception of oxygen (O₂) and nitrogen (N₂), the other gases are formed from secondary chemical reactions, as a result of the breakdown of hydrocarbon molecules due to electrical or thermal stresses.

From the observation of this behavior, several studies have been developed with the objective of defining the type of failure in the equipment from the analysis of the composition of the gases formed.

These techniques make it possible to predict failures before the need to shut down the equipment, the oil draining process, and access for inspection, which can require days of work. Figure 5 illustrates two practical uses of DGA analysis in failures.

In Figure 5a, the monitoring of the equipment power transformer 315MVA 15.75/400 kV identified a large formation of C₂H₄ gas through analysis, showing a possible high-intensity thermal failure. During the inspection, two flexible windings from the bushing were found heated. Figure 5b presents another example of monitoring the equipment power transformer 55 MVA 110/11 kV, which showed a large formation of C₂H₂. DGA analysis indicated thermal failure coupled with the arc flash. The inspection revealed that the winding coil of a phase was burnt out.

In this article, we present the most popular DGA analysis techniques adopted among international standards and analyzed their performance in more detail. These analyses were made with a modified confusion matrix, with the data and types of failures presented in [15]. In addition to using an accessible failure database [15,16], the current study shows whether the techniques are determining the right hypothesis or not. This gives readers a clear and easy-to-interpret direction when using the analysis or developing new techniques.

Therefore, the combination of the interpretations of these techniques and the performance detailing different types of failure, with more than just a generic indicator of accuracy, makes this study an unprecedented review [17,18,19,20,21,22,23,24]. In addition, a table is provided with a combination of all of the responses to the pattern failure of the techniques adopted in [15], allowing the comparison to be more organized and more specific. Other studies have made unclear comparisons of the results of the adopted techniques that often did not have the same pattern of failures defined in [15].

The detailed analysis of the performance techniques from Rogers [11] and the IEC method [11], with the confusion matrix for standardized failures in [15], resulted in a table that grouping the incipient fault types codes for conventional DGA techniques showed in Section 3. These techniques, as presented in the results of this article, presented an overall performance superior to those seen in Rogers’ IEEE method [12] and the IEC 60599 method [25]. Therefore, this study presents a new approach to the use of these techniques that can contribute to better failure analysis performance.

It should be noted that, in this article, the techniques will not address mineral oil-insulated switching equipment; for example, load tap changers, circuit breakers, reclosers, or others. The analysis of dissolved gases in switches is not easy to interpret, as the formation of gases is very high in this equipment, normally associated with electrical arcs. Despite this, in the specific case of load tap changers, there are scientific documents such as [26] and standards such as [27], which serve as guidelines for interpreting failures.

2. International Standard Developed Techniques

Since the first observations of the formation of dissolved gases in transformers, some techniques have been developed and disseminated both in the academic environment, by equipment manufacturers, and by companies in the electrical sector responsible for the maintenance and operation of this equipment.

Conventional techniques can be defined by analysis of the concentration of gases, by the relationship between the proportions of gases, or by the graphical method of classifying the faults, among which we highlighted [17]:

• Incipient Fault Types, Frank M. Clark, 1933–1962;
• Dörnenburg Ratios, E. Dörnenburg, 1967, 1970;
• Potthoff’s Scheme, K. Potthoff, 1969;
• Absolute limits, various sources, early 1970;
• Shank’s Visual Curve method, 1970s;
• Trilinear Plot Method, 1970;
• Key Gas Method, David Pugh, 1974;
• Duval’s Triangle, Michel Duval, 1974;
• Rogers Ratios, R.R. Rogers, 1975;
• Method MSS, 1975;
• Glass Criterion, R.M Glass, 1977;
• Trend Analysis, various sources, early 1980s;
• Church Logarithmic Nomograph, J.O. Church, 1980;
• IEEE C57.104, Limits, rates and total dissolved combustible gas (TDCG), 1978–1991;
• IEC 60599 Ratios, Limits and gassing rates, 1999;
• CIGRE TF 15.01.01, 1999;
• Duval’s Pentagon, Michel Duval, 2014;
• CIGRE TB 771-Advances in DGA interpretation, 2019;
• IEEE EI Mag-New sub-zones for arcing in paper and in oil in Pentagons and Triangle 1, 2022;
• IEEE EI Mag-New sub-zones for carbonization of paper C in Pentagon 2 and Triangle 5, 2023.

In addition to the techniques previously mentioned, other methodologies have been developed over time to meet regional definitions, specific demands, or due to the evolutionary process of product companies or electric energy concessionaires, among them:

• Amount of Key Gases, California State University (CSUS);
• RD 153-34.0-46.302-00 procedural guidelines Russian;
• GATRON method Germany;
• ETRA square Japan;
• Total Combustible Gases, Doble Engineering.

The following subsections present the concepts developed by the main analysis techniques [5,12,18,19,20,21,22,23,24,25,26,28,29,30] agreed on by the IEEE [12] and IEC [25] standards.

2.1. Key Gas

The key gas method, developed by Doble Engineering [4], is based on the fact that when a fault occurs, there is the formation of gases that exceed the normal values of insulation degradation. When the gas that characterizes the incipient failure type, called the key gas, is predominant among other gases from typical formations in failure, the condition is considered abnormal. In the following subsections, failures with key gas are shown.

Failure type: Arcing. The formation of faults with arcing can reach temperatures between 2000 and 3000 °C [20], and large amounts of H₂ and C₂H₂ and small amounts of CH₄ and C₂H₄ are produced, as shown in Figure 6. CO and CO₂ can be produced if the failure involves cellulose. Key gas: C₂H₂.

Failure type: Partial discharges—Low-energy electrical discharges produce H₂ and CH₄, with a small amount of C₂H₆ and C₂H₄, as shown in Figure 7. Comparable amounts of CO and CO₂ can result from discharges into the pulp. Key gas: H₂.

Failure type: Overtemperature in oil—Overtemperatures in oil with values of at least 300 °C have the characteristic formation of C₂H₆ and CH₄. Decomposition products also include CH₄. Traces of C₂H₂ can also form if the fault is severe or if there are electrical contacts, as shown in Figure 8. Key gas: C₂H₄.

Failure type: Overheated Cellulose—Overheating insulation paper produces a large amount of CO and CO₂, but the latter is not a combustible gas, emphasizing that the analyses pertain to combustible gases. Gaseous hydrocarbons, such as CH₄ and C₂H₄, are also present if the fault consists of an oil-impregnated structure, as shown in Figure 9. Key gases: CO and CO₂.

2.2. Dornenburg

A report published as part of the CIGRE International Conference on Large High Tension Electric Systems in 1970 [31] was one of the first failure analysis methods based on the ratio of CH₄, C₂H₂, C₂H₄, and H₂ gases. Figure 10 presents an image of the early publication of the relationship between gases by Dornenburg [4].

In this methodology, if any of the gases exceed twice the value of the reference in

Table 1. Reference value gas concentrations; values may differ according to criteria defined by standards.

Gas Type	Concetration (ppm)
Hydrogen (H2)	100
Methane (CH4)	120
Carbon Monoxide (CO)	350
Carbon Dioxide (CO2)	2500
Acetylene (C2H2)	1
Ethylene (C2H4)	50
Ethane (C2H6)	65

or if one of the CO or C₂H₆ gases is greater than this, it will be considered a failure and then this type of failure will be analyzed.

The types of failure are evaluated in a generic diagnosis based on the proportion between the gases, according to

Table 2. Relationships of the Dornenburg method [12].

Suggested Fault Diagnosis	CH4 H2	C2H2 C2H4	C2H2 CH4	C2H6 C2H2
Thermal Decomposition	>1	<0.75	<0.3	>0.4
Partial Discharge (Low intensity)	<0.1	Not significant	<0.3	>0.4
Arcing (High intensity)	0.1 to 1	>0.75	>0.3	<0.4

, which allows the identification of three types of failure: hot spot, electrical discharge (arc), and partial discharges.

2.3. Rogers

Based on the Dornenburg method, the Rogers method proposed the evaluation of failures by the relationship between the gases C₂H₂, C₂H₄, C₂H₆, CH₄, and H₂, originally published according to

Table 3. Rogers method relationships [11].

Fault	CH4 H2	C2H6 CH4	C2H4 C2H6	C2H2 C2H4
Normal	≥0.1 and <1	<1	<1	<0.5
Partial discharge	≤0.1	<1	<1	<0.5
Overheating < 150 °C	≥1 and <3 or ≥3	<1	<1	<0.5
Overheating 150 °C to 200 °C	≥1 and <3 or ≥3	≥1	<1	<0.5
Overheating 200 °C to 300 °C	>0.1 and <1	≥1	<1	<0.5
General conductor overheating	>0.1 and <1	<1	≥1 and <3	<0.5
Winding circulating currents	≥1 and <3	<1	≥1 and <3	<0.5
Core and tank circulating currents	≥1 and <3	<1	≥3	<0.5
Flashover without power follow through	>0.1 and <1	<1	<1	≥0.5 and <3
Arc with power follow through	>0.1 and <1	<1	≥1 and <3 or ≥3	≥0.5 and <3 or ≥3
Continuous sparking to floating potential	>0.1 and <1	<1	≥3	≥3
Partial discharge with tracking	≤0.1	<1	<1	≥0.5 and <3 or ≥3

.

The publication of the Rogers method through the IEEE Std C57.104 standard [12] did not consider the relationship between C₂H₆ and CH₄, reducing the amount of failure types, as shown in

Table 4. Rogers method relationships IEEE Std C57.104.

Fault	C2H2 C2H4	CH4 H2	C2H4 C2H6
Normal	<0.1	≥0.1 and <1	<1
Low-energy density arcing-PD	<0.1	<0.1	<1
Arcing high-energy discharge	≥1 and ≤3	≥0.1 and ≤1	>3
Low temperature thermal	<0.1	≥0.1 and ≤1	≥1 and ≤3
Thermal < 700 °C	<0.1	>1	≥1 and ≤3
Thermal > 700 °C	<0.1	>1	>3

.

2.4. IEC

This method was published in the IEEE article in 1978 [11], differentiating from the Rogers method by removing the ratio analysis between the C₂H₆/CH₄ gases. The purpose of removing this ratio was to simplify the analysis of failures, since the contribution of this relationship was limited to a small range in terms of the decomposition temperature and was not useful in identifying the failure.

Table 5. IEC ratio method relations [11].

Fault	C2H2 C2H4	CH4 H2	C2H4 C2H6
No fault	<0.1	≥0.1 and <1	<1
Partial discharge of low-energy density	<0.1	<0.1	≥1 and <3
Partial discharge of high-energy density	≥0.1 and <3	<0.1	≥1 and <3
Discharge of low energy	≥0.1 and <3 or ≥3	≥0.1 and <1	<1
Discharge of high energy	≥0.1 and <3	≥0.1 and <1	<1
Thermal fault of temperature < 150 °C	<0.1	≥0.1 and <1	<1
Thermal fault of low temperature range 150 °C to 300 °C	<0.1	≥1	>3
Thermal fault of medium temperature range 300 °C to 700 °C	<0.1	≥1	>3
Thermal fault of high temperature range > 700 °C	<0.1	≥1	>3

presents the first version of the IEC ratio method published in the article “IEE and IEC Codes to Interpret Incipient Faults in Transformers, using Gas in Oil Analysis” [11] and

Table 6. IEC 60599 [25].

Fault	C2H2 C2H4	CH4 H2	C2H4 C2H6
Partial discharges (PD)	NS	<0.1	<0.2
Discharge of low energy (D1)	>1	≥0.1 and ≤0.5	>1
Discharge of high energy (D2)	≥0.6 and ≤2.5	≥0.1 and ≤1	>2
Thermal fault t < 300 °C (T1)	NS	≥1 but NS	<1
Thermal fault 300 °C < t < 700 °C (T2)	<0.1	>1	≥1 and ≤4
Thermal fault t > 700 °C (T3)	<0.2	>1	>4

NS: Non significant whatever the value.

shows the recent codification through the IEC 60599 standard [25].

In order to use the IEC 60599 ratio method, at least one of the gases shown in Table 1 must be above its limit, if these limits are not exceeded, the conditions of the equipment will be considered normal.

The use of the IEC 60599 ratios method on the Cartesian axis, as shown in Figure 11, allows for the visualization of fault types D1 and D2 overlapping (both related to the discharge of energy). Despite illustrating cases of lightning strikes, this intersection area can lead to the wrong interpretation of their intensity.

2.5. Duval Triangle

The Duval Triangle is currently one of the most widely used methods for interpreting the types of failures in electrical equipment. Its first publication was in 1974 [26].

In the same way as the Dornenburg criteria, the Duval criteria present generic diagnoses for identifying four types of faults: hot spots, a high-energy arc, a low-energy arc, and internal discharges.

The Duval method can indicate defects in equipment that present normal operating conditions. For this type of error to not occur, using Duval’s method, it is necessary that some of the gases exceed the limits established in Table 1, if these limits are not exceeded, the conditions of the equipment will be considered normal.

The Duval triangles are presented in seven different types, T1 to T7, with the T1 method being the classic Duval method applied to insulated transformers with insulating mineral oil, and the T4 and T5 methods being a refinement of this to deepen the interpretation of the type of failure.

The other methods are applied as follows: T2 for on-load tap changer compartments insulated with mineral oil; T3 equipment insulated with non-mineral oil, such as Bio Temp, Silicone, Midel, and FR3; T6 is used to gain further information regarding faults identified as low-temperature faults (PD, T1, or T2) by T3 for FR3 oils and T7 to gain further information regarding faults identified as thermal faults (T1, T2, or T3) and by T3 for FR3 oils [28,31].

In the initial failure analysis of equipment insulated with mineral-insulating oil, T1 is used, as shown in Figure 12, to evaluate the CH₄, C₂H₄, and C₂H₂ gases [29] from Equations (1), (2), and (3).

$${\mathrm{\%}\mathrm{C}\mathrm{H}}_4=\frac{{\mathrm{C}\mathrm{H}}_4}{{\mathrm{C}\mathrm{H}}_4+{{\mathrm{C}}_2\mathrm{H}}_4+{{\mathrm{C}}_2\mathrm{H}}_2}$$

(1)

$$\mathrm{\%}{{\mathrm{C}}_2\mathrm{H}}_4=\frac{{{\mathrm{C}}_2\mathrm{H}}_4}{{\mathrm{C}\mathrm{H}}_4+{{\mathrm{C}}_2\mathrm{H}}_4+{{\mathrm{C}}_2\mathrm{H}}_2}$$

(2)

$$\mathrm{\%}{{\mathrm{C}}_2\mathrm{H}}_2=\frac{{{\mathrm{C}}_2\mathrm{H}}_2}{{\mathrm{C}\mathrm{H}}_4+{{\mathrm{C}}_2\mathrm{H}}_4+{{\mathrm{C}}_2\mathrm{H}}_2}$$

(3)

When low energy or temperature faults are identified from Triangle T1 (PD, T1, and T2), more information can be obtained with Triangle T4, as shown in Figure 13, and gases H₂, C₂H₄, and C₂H₆ from Equations (4)–(6).

$${\mathrm{\%}\mathrm{C}\mathrm{H}}_4=\frac{{\mathrm{C}\mathrm{H}}_4}{{\mathrm{C}\mathrm{H}}_4+{{\mathrm{C}}_2\mathrm{H}}_6+{\mathrm{H}}_2}$$

(4)

$$\mathrm{\%}{{\mathrm{C}}_2\mathrm{H}}_6=\frac{{{\mathrm{C}}_2\mathrm{H}}_6}{{\mathrm{C}\mathrm{H}}_4+{{\mathrm{C}}_2\mathrm{H}}_6+{\mathrm{H}}_2}$$

(5)

$$\mathrm{\%}{\mathrm{H}}_2=\frac{{\mathrm{H}}_2}{{\mathrm{C}\mathrm{H}}_4+{{\mathrm{C}}_2\mathrm{H}}_6+{\mathrm{H}}_2}$$

(6)

If the analysis from Triangle T1 includes high or extremely high temperatures (T2 and T3), a more detailed analysis can be obtained with Triangle T5, as shown in Figure 14, with the analysis of CH₄, C₂H₄, and C₂H₆ gases from Equations (7)–(9).

$${\mathrm{\%}\mathrm{C}\mathrm{H}}_4=\frac{{\mathrm{C}\mathrm{H}}_4}{{\mathrm{C}\mathrm{H}}_4+{{\mathrm{C}}_2\mathrm{H}}_4+{{\mathrm{C}}_2\mathrm{H}}_6}$$

(7)

$$\mathrm{\%}{{\mathrm{C}}_2\mathrm{H}}_4=\frac{{{\mathrm{C}}_2\mathrm{H}}_4}{{\mathrm{C}\mathrm{H}}_4+{{\mathrm{C}}_2\mathrm{H}}_4+{{\mathrm{C}}_2\mathrm{H}}_6}$$

(8)

$$\mathrm{\%}{{\mathrm{C}}_2\mathrm{H}}_6=\frac{{{\mathrm{C}}_2\mathrm{H}}_6}{{\mathrm{C}\mathrm{H}}_4+{{\mathrm{C}}_2\mathrm{H}}_4+{{\mathrm{C}}_2\mathrm{H}}_6}$$

(9)

2.6. Duval Pentagon

Graphical analysis using a pentagon shape was presented in 2012 using the Mansour Pentagon method [30], to simultaneously analyze the gases CH₄, C₂H₆, C₂H₄, C₂H₂, and H₂ (see Figure 15).

This analysis proposal aimed to overcome a failure of the Duval Triangle method in that it does not have the gases C₂H₆ and H₂ in its analysis, which are relevant in the analysis of failure due to low overtemperature and discharge.

The Duval Pentagon method was published in 2014 [33], presenting fault boundary regions different from the proposal in [30] and divided into Duval Pentagons 1 and 2.

Figure 16 shows the Pentagon Duval 1, which is for the general analysis of fault types.

Duval Pentagon 2, as shown in Figure 17, is equivalent to the refinement presented by Duval Triangles 4 and 5 [33,34].

The Pentagon 2 method is complementary to the Pentagon 1 analysis, suggested for the refinement of the analysis if failures of type T1, T2, or T3 are identified [12].

In both methods, the calculations to map the type of failure with the collected gas data [32,33] start with one of the gases with values above the limits shown in Table 1; if these limits are not exceeded, the conditions of the equipment will be considered normal.

From the analysis that the equipment is faulty, the type of fault is evaluated with the calculation of the proportion between the gases (Equations (10) to (14)), being:

$${\mathrm{\%}\mathrm{C}\mathrm{H}}_4=\frac{{\mathrm{C}\mathrm{H}}_4}{{\mathrm{C}\mathrm{H}}_4+{{\mathrm{C}}_2\mathrm{H}}_4+{{\mathrm{C}}_2\mathrm{H}}_6+{{\mathrm{C}}_2\mathrm{H}}_2+{\mathrm{H}}_2}$$

(10)

$${{\mathrm{\%}\mathrm{C}}_2\mathrm{H}}_4=\frac{{{\mathrm{C}}_2\mathrm{H}}_4}{{\mathrm{C}\mathrm{H}}_4+{{\mathrm{C}}_2\mathrm{H}}_4+{{\mathrm{C}}_2\mathrm{H}}_6+{{\mathrm{C}}_2\mathrm{H}}_2+{\mathrm{H}}_2}$$

(11)

$${{\mathrm{\%}\mathrm{C}}_2\mathrm{H}}_6=\frac{{{\mathrm{C}}_2\mathrm{H}}_6}{{\mathrm{C}\mathrm{H}}_4+{{\mathrm{C}}_2\mathrm{H}}_4+{{\mathrm{C}}_2\mathrm{H}}_6+{{\mathrm{C}}_2\mathrm{H}}_2+{\mathrm{H}}_2}$$

(12)

$${{\mathrm{\%}\mathrm{C}}_2\mathrm{H}}_2=\frac{{{\mathrm{C}}_2\mathrm{H}}_2}{{\mathrm{C}\mathrm{H}}_4+{{\mathrm{C}}_2\mathrm{H}}_4+{{\mathrm{C}}_2\mathrm{H}}_6+{{\mathrm{C}}_2\mathrm{H}}_2+{\mathrm{H}}_2}$$

(13)

$${\mathrm{\%}\mathrm{H}}_2=\frac{{\mathrm{H}}_2}{{\mathrm{C}\mathrm{H}}_4+{{\mathrm{C}}_2\mathrm{H}}_4+{{\mathrm{C}}_2\mathrm{H}}_6+{{\mathrm{C}}_2\mathrm{H}}_2+{\mathrm{H}}_2}$$

(14)

With the percentages of gas concentration, the coordinates of each of these on the Cartesian axis are calculated [33] with Equations (15) to (19).

$${\mathrm{C}\mathrm{o}\mathrm{r}\mathrm{d}\_\mathrm{\%}\mathrm{C}\mathrm{H}}_4=\left(\right(\mathrm{\%}{\mathrm{C}\mathrm{H}}_4\times \cos 254°);(\mathrm{\%}{\mathrm{C}\mathrm{H}}_4\times \cos -164°\left)\right)$$

(15)

$${{\mathrm{C}\mathrm{o}\mathrm{r}\mathrm{d}\_\mathrm{\%}\mathrm{C}}_2\mathrm{H}}_6=(\left(\mathrm{\%}{{\mathrm{C}}_2\mathrm{H}}_6\times \cos 162°\right);(\mathrm{\%}{{\mathrm{C}}_2\mathrm{H}}_6\times \cos -72°))$$

(16)

$${{\mathrm{C}\mathrm{o}\mathrm{r}\mathrm{d}\_\mathrm{\%}\mathrm{C}}_2\mathrm{H}}_4=\left(\right(\mathrm{\%}{{\mathrm{C}}_2\mathrm{H}}_4\times \cos 326°);(\mathrm{\%}{{\mathrm{C}}_2\mathrm{H}}_4\times \cos -236°\left)\right)$$

(17)

$${{\mathrm{C}\mathrm{o}\mathrm{r}\mathrm{d}\_\mathrm{\%}\mathrm{C}}_2\mathrm{H}}_2=\left(\right(\mathrm{\%}{{\mathrm{C}}_2\mathrm{H}}_2\times \cos 18°);\left(\mathrm{\%}{{\mathrm{C}}_2\mathrm{H}}_2\times \cos 72°\right))$$

(18)

$${\mathrm{C}\mathrm{o}\mathrm{r}\mathrm{d}\_\mathrm{\%}\mathrm{H}}_2=\left(\right(\mathrm{\%}{\mathrm{H}}_2\times \cos 90°);\left(\mathrm{\%}{\mathrm{H}}_2\times \cos 0°\right))$$

(19)

From the coordinates of the gas concentrations, the centroid of these points is calculated with Equations (20)–(22), the location of which will indicate the type of fault.

$${\mathrm{C}}_{\mathrm{x}}=\frac{1}{6\mathrm{A}}{\sum }_{\mathrm{i}=0}^{\mathrm{n}-1}({\mathrm{x}}_{\mathrm{i}}+{\mathrm{x}}_{\mathrm{i}+1})({\mathrm{x}}_{\mathrm{i}}{\mathrm{y}}_{\mathrm{i}+1}-{\mathrm{x}}_{\mathrm{i}+1}{\mathrm{y}}_{\mathrm{i}})$$

(20)

$${\mathrm{C}}_{\mathrm{y}}=\frac{1}{6\mathrm{A}}{\sum }_{\mathrm{i}=0}^{\mathrm{n}-1}({\mathrm{y}}_{\mathrm{i}}+{\mathrm{y}}_{\mathrm{i}+1})({\mathrm{x}}_{\mathrm{i}}{\mathrm{y}}_{\mathrm{i}+1}-{\mathrm{x}}_{\mathrm{i}+1}{\mathrm{y}}_{\mathrm{i}})$$

(21)

$$\mathrm{A}=\frac{1}{2}{\sum }_{\mathrm{i}=0}^{\mathrm{n}-1}({\mathrm{x}}_{\mathrm{i}}{\mathrm{y}}_{\mathrm{i}+1}-{\mathrm{x}}_{\mathrm{i}+1}{\mathrm{y}}_{\mathrm{i}})$$

(22)

In Equations (20)–(22), n is the number of coordinates of the gases under analysis.

2.7. Others Methods

The analysis of dissolved gases in electrical equipment, as it is not an exact science, is undergoing constant evolution.

Ratio analysis methods, such as IEC, Dornenburg, and Rogers, fail to identify failures in about 15 to 20% of cases, as the cases are outside the failure assessment zones. The key gas method, when applied with the aid of computational tools, presents an error in around 50% of the analyzed cases [35].

Duval Triangle and Pentagon graphic methods do not have the limitations of not identifying failures, since, as they are closed methods, they always present an answer for the type of failure. Furthermore, due to the large number of failure cases on which these methods were based, the assertiveness of these methods is very high.

The IEC method allows for the evaluation of the six basic types of faults, T1, T2, T3, PD, D1, and D2, the Duval Triangle, and Pentagon methods. Furthermore, it permits the identification of the types S, O, C, and T3-H, making the information more useful for actions regarding equipment.

Currently, several ways to improve the accuracy of the analysis have been developed using computational methods of artificial intelligence [36,37,38]. However, in addition to the search for this computational improvement, conventional methods are still debated with each new evaluation proposal.

Although the focus of this review is based on conventional methods standardized through the standards [12,25], the following subsections will briefly present some of the recently developed techniques for the conventional analysis of dissolved gases.

2.7.1. Three Rates Technique TRT

The TRT technique, published in 2018 [39], is a proposal that does not graphically operate and presents a new proposal for ratios between the gases CH₄, C₂H₆, C₂H₄, C₂H₂, and H₂. Its goal is to correct the distortions presented through other conventional analyses. The new ratios are shown in Equations (23)–(25):

$${\mathrm{R}}_1=\frac{{{\mathrm{C}}_2\mathrm{H}}_6+{{\mathrm{C}}_2\mathrm{H}}_4}{{\mathrm{H}}_2+{{\mathrm{C}}_2\mathrm{H}}_2}$$

(23)

$${\mathrm{R}}_2=\frac{{{\mathrm{C}}_2\mathrm{H}}_2+{\mathrm{C}\mathrm{H}}_4}{{{\mathrm{C}}_2\mathrm{H}}_4}$$

(24)

$${\mathrm{R}}_3=\frac{{{\mathrm{C}}_2\mathrm{H}}_2}{{{\mathrm{C}}_2\mathrm{H}}_4}$$

(25)

Based on these ratios, a new fault coding is proposed, as shown in

Table 7. Diagnosis coding rule of proposed TRT interpretation technique.

Fault	R1	R2	R3
T3	R1 > 0.05	R2 ≤ 1	R3 < 0.5
T2	R1 > 0.05	1 ≤ R2 ≤ 3.5	R3 < 0.5
T1	R1 > 0.05	R2 > 3.5	R3 < 0.5
T0	0.05 ≤ R1 ≤ 0.9	NS.	R3 < 0.05
PD1	R1 ≤ 0.05	R2 > 1	R3 < 0.5
PD2	R1 ≤ 0.05	R2 > 1	R3 ≥ 0.5
D1	R1 ≥ 0.05	R2 > 3.5	R3 ≥ 0.5
D2	R1 ≤ 0.9	R2 ≤ 3.5	R3 ≥ 0.5
DT	R1 > 0.9	R2 ≤ 3.5	R3 ≥ 0.5

NS, non-significant whatever the value; (T3): Thermal faults of T > 700 °C; (T2): Thermal faults of 300 < T < 700 °C; (T1): Thermal faults 150 < T < 300 °C; (T0): Thermal faults T < 150 °C; (D1): Low-energy discharge; (D2): High-energy discharge; (PD1): Low-energy-corona partial discharge; (PD2): High-energy-corona partial discharge; (DT): Mix of electrical and thermal faults.

.

2.7.2. Heptagon Graph

For a more detailed proposal of the analysis of the failure, the effects on the cellulosic insulation by the formation of CO and CO₂, together with the analysis of the gases C₂H₆ and H₂, the method Heptagon Graph (Figure 18) graphically assesses failures in equipment insulated with mineral oil from CH₄, C₂H₆, C₂H₄, C₂H₂, H₂, CO, and CO₂ gases. This technique was published in 2018 [40] and included the analysis of C₂H₆ and H₂ due to their crucial characteristics in certain types of failures.

For analysis using this method, from the collection and analysis of gas concentrations, it is evaluated if at least one of them exceeds the limits of Table 1.

If any of the gases present values above the limits in Table 1, the evaluation of the type of failure will begin with the conversion of values into proportions, using Equations (26) to (32).

$${\mathrm{\%}\mathrm{C}\mathrm{H}}_4=\frac{2.9167\times {\mathrm{C}\mathrm{H}}_4\times 100}{\begin{array}{c}(2.9167\times {\mathrm{C}\mathrm{H}}_4+5.3846\times {\mathrm{C}}_2{\mathrm{H}}_6+7\times {\mathrm{C}}_2{\mathrm{H}}_4+116.6667\times {\mathrm{C}}_2{\mathrm{H}}_2+3.5\times {\mathrm{H}}_2+CO+0.14\times {\mathrm{C}\mathrm{O}}_2)\end{array}}$$

(26)

$${\mathrm{\%}{\mathrm{C}}_2\mathrm{H}}_6=\frac{5.3846\times {\mathrm{C}}_2{\mathrm{H}}_6\times 100}{(2.9167\times {\mathrm{C}\mathrm{H}}_4+5.3846\times {\mathrm{C}}_2{\mathrm{H}}_6+7\times {\mathrm{C}}_2{\mathrm{H}}_4+116.6667\times {\mathrm{C}}_2{\mathrm{H}}_2+3.5\times {\mathrm{H}}_2+\mathrm{C}\mathrm{O}+0.14\times {\mathrm{C}\mathrm{O}}_2)}$$

(27)

$${\mathrm{\%}{\mathrm{C}}_2\mathrm{H}}_4=\frac{7\times {\mathrm{C}}_2{\mathrm{H}}_4\times 100}{(2.9167\times {\mathrm{C}\mathrm{H}}_4+5.3846\times {\mathrm{C}}_2{\mathrm{H}}_6+7\times {\mathrm{C}}_2{\mathrm{H}}_4+116.6667\times {\mathrm{C}}_2{\mathrm{H}}_2+3.5\times {\mathrm{H}}_2+\mathrm{C}\mathrm{O}+0.14\times {\mathrm{C}\mathrm{O}}_2)}$$

(28)

$${\mathrm{\%}{\mathrm{C}}_2\mathrm{H}}_2=\frac{116.6667\times {\mathrm{C}}_2{\mathrm{H}}_2\times 100}{(2.9167\times {\mathrm{C}\mathrm{H}}_4+5.3846\times {\mathrm{C}}_2{\mathrm{H}}_6+7\times {\mathrm{C}}_2{\mathrm{H}}_4+116.6667\times {\mathrm{C}}_2{\mathrm{H}}_2+3.5\times {\mathrm{H}}_2+\mathrm{C}\mathrm{O}+0.14\times {\mathrm{C}\mathrm{O}}_2)}$$

(29)

$${\mathrm{\%}\mathrm{H}}_2=\frac{3.5\times {\mathrm{H}}_2\times 100}{(2.9167\times {\mathrm{C}\mathrm{H}}_4+5.3846\times {\mathrm{C}}_2{\mathrm{H}}_6+7\times {\mathrm{C}}_2{\mathrm{H}}_4+116.6667\times {\mathrm{C}}_2{\mathrm{H}}_2+3.5\times {\mathrm{H}}_2+\mathrm{C}\mathrm{O}+0.14\times {\mathrm{C}\mathrm{O}}_2)}$$

(30)

$$\mathrm{\%}\mathrm{C}\mathrm{O}=\frac{\mathrm{C}\mathrm{O}\times 100}{(2.9167\times {\mathrm{C}\mathrm{H}}_4+5.3846\times {\mathrm{C}}_2{\mathrm{H}}_6+7\times {\mathrm{C}}_2{\mathrm{H}}_4+116.6667\times {\mathrm{C}}_2{\mathrm{H}}_2+3.5\times {\mathrm{H}}_2+\mathrm{C}\mathrm{O}+0.14\times {\mathrm{C}\mathrm{O}}_2)}$$

(31)

$$\mathrm{\%}{\mathrm{C}\mathrm{O}}_2=\frac{0.14\times {\mathrm{C}\mathrm{O}}_2\times 100}{(2.9167\times {\mathrm{C}\mathrm{H}}_4+5.3846\times {\mathrm{C}}_2{\mathrm{H}}_6+7\times {\mathrm{C}}_2{\mathrm{H}}_4+116.6667\times {\mathrm{C}}_2{\mathrm{H}}_2+3.5\times {\mathrm{H}}_2+\mathrm{C}\mathrm{O}+0.14\times {\mathrm{C}\mathrm{O}}_2)}$$

(32)

With the relative percentages of gas concentrations, the point of failure in the heptagon is calculated through the center of mass of the concentrations.

3. Results

To analyze the performance of the main techniques presented in this study, the failure database, with real cases, available in Appendix A, was used.

We adapted the type of failure of some techniques to the classification pattern of failures presented in [15], enabling the analysis of technique performances.

Table 8. Grouping of the incipient fault type codes for conventional DGA interpretation techniques.

Techinique	Fault Types
	T1	T2	T3	PD	D1	D2
Duval Triangle T1	Thermal fault < 300 °C	Thermal fault 300–700 °C	Thermal fault > 700 °C	Partial Discharge	Low-energy discharge	High-energy discharge
Dornenburg	Thermal decomposition (T)			Partial Discharge (PD)	Energy Discharge–Arcing (D)
Rogers	Thermal fault of low temperature < 150 °C (T1_1)	Winding circulation current (T2)	Core and tank circulation current (T3_1)	Partial discharge with low energy (PD_1)	Continuous Sparking (D1_1)	Arc with power follows through (D2)
	Thermal fault of temperature range 150–200 °C (T1_2)		Insulated conductor overheating (T3_2)	Partial discharge with tracking (PD_2)	Flashover (D1_2)
	Thermal fault of temperature range 200–300 °C (T1_3)
Rogers IEEE	Low temperature thermal (T1)	Thermal < 700 °C (T2)	Thermal > 700 °C (T3)	Low-energy density arcing-(PD)	Arcing-High-energy discharge (D)
IEC ratio method	Thermal fault of temperature < 150 °C (T1_1)	Thermal fault of medium temperature range 300 °C to 700 °C (T2)	Thermal fault of high temperature range > 700 °C (T3)	PD of low-energy density (PD_1)	Discharge with low-energy density (D1)	Discharge with high-energy arcing (D2)
	Thermal fault of low temperature range 150 °C to 300 °C (T1_2)			PD of high-energy density (PD_2)
IEC 60599 ratio method	Thermal fault t < 300 °C (T1)	Thermal fault 300 °C < t < 700 °C (T2)	Thermal fault t > 700 °C (T3)	Partial Discharge (PD)	Discharge of low energy (D1)	Discharge of high energy (D2)
Duval Pentagon 1	Termal fault < 300 °C	Thermal fault 300–700 °C	Thermal fault > 700 °C	Partial Discharge	Low-energy discharge	High-energy discharge
Key Gas	Thermal degradation (TD)			Partial Discharge (PD)	Arcing (D)
	Overheat cellulose (OC)

Note: For analysis of the results, information about non-failure—N and unidentified failure—Not-F will also be used.

, based on [41] and adapted according to the observations of this work, presents the summary of the adaptations developed.

One way to evaluate the results was by using confusion matrices. The confusion matrix is a very popular way to evaluate classification tasks, representing counts of predicted and actual values. For this reason, this form of analysis was selected to detail the response of the techniques to the expected result.

The actual classes (types of failures) are vertically distributed, while the predicted classes by the technique are horizontally distributed. The number in the intersection of classes with the same name (actual and predicted classes) means the number of correct identifications for that class. The numbers in the intersection of classes with different names are misclassified failures.

Due to the adaptation of the predicted or expected results presented in Table 8, it was necessary to modify the confusion matrix, since there will not always be predicted or known values in all method scenarios. For example, for the key gas method, the actual failures T1, T2, and T3 are included in the failures TD and OC. Therefore, different from the common confusion matrix, in our analysis, the confusion matrix can be not square.

In addition to using the confusion matrix to analyze the techniques within each type of failure, some performance measures were used and are detailed below:

• Precision: This metric measures how much we can trust a model when it predicts that an example belongs to a certain class:

$$\mathrm{P}\mathrm{r}\mathrm{e}\mathrm{c}\mathrm{i}\mathrm{s}\mathrm{o}\mathrm{n}=\frac{\mathrm{T}\mathrm{P}}{\mathrm{T}\mathrm{P}+\mathrm{F}\mathrm{P}}$$

(33)

• Recall: This metric measures the ratio of failures that were correctly identified for a specific class:

$$\mathrm{R}\mathrm{e}\mathrm{c}\mathrm{a}\mathrm{l}\mathrm{l}=\frac{\mathrm{T}\mathrm{P}}{\mathrm{T}\mathrm{P}+\mathrm{F}\mathrm{N}}$$

(34)

• F1-Score: This metric is the harmonic mean between accuracy and recall:

$$\mathrm{F}1-\mathrm{S}\mathrm{c}\mathrm{o}\mathrm{r}\mathrm{e}=2\times \frac{\mathrm{P}\mathrm{r}\mathrm{e}\mathrm{c}\mathrm{i}\mathrm{s}\mathrm{o}\mathrm{n}\times \mathrm{R}\mathrm{e}\mathrm{c}\mathrm{a}\mathrm{l}\mathrm{l}}{\mathrm{P}\mathrm{r}\mathrm{e}\mathrm{c}\mathrm{i}\mathrm{s}\mathrm{o}\mathrm{n}+\mathrm{R}\mathrm{e}\mathrm{c}\mathrm{a}\mathrm{l}\mathrm{l}}$$

(35)

• Accuracy: This metric is the fraction of predictions that the model got right:

$$\mathrm{A}\mathrm{c}\mathrm{c}\mathrm{u}\mathrm{r}\mathrm{a}\mathrm{c}\mathrm{y}=\frac{\mathrm{T}\mathrm{P}+\mathrm{T}\mathrm{N}}{\mathrm{T}\mathrm{P}+\mathrm{T}\mathrm{N}+\mathrm{F}\mathrm{P}+\mathrm{F}\mathrm{N}}$$

(36)

• Support: This will give the information of the number of cases in analysis.

To solve Equations (33) and (34), the definitions used are:

• True Positive (TP): model correctly predicts the positive class (prediction and actual, both are positive);
• True Negative (TN): model correctly predicts the negative class (prediction and actual, both are negative);
• False Positive (FP): model gives the wrong prediction of the negative class (predicted-positive, actual-negative);
• False Negative (FN): model wrongly predicts the positive class (predicted-negative, actual-positive).

3.1. Key Gas

From the analysis of the confusion matrix, as shown in Figure 19, evaluated without combining the response of the types of failures of this technique, a significant assertiveness of the failures of the T3 type is observed. It is also noteworthy that around 21%, 19 cases, were not identified (Not-F). Furthermore, PD-type failures were not identified by this technique.

To visualize the efficiency of this technique, the types of standard faults [15] were adapted according to the proposed grouping in Table 8, transforming the TD and OC faults into T, with the result of the performance analysis of this technique shown in

Table 9. Performance analysis of the Key Gas method.

Failure	Performance Analysis
	Precision	Recall	F1-Score	Support
D	0.82	0.71	0.76	51
PD	0.00	0.00	0.00	8
T	0.88	0.73	0.80	30
Accuracy			0.66	88

.

The analysis of the Precision, Recall, and F1-score performances shows that the technique presents a good performance for the failures that it identifies.

It is observed that despite a high overall accuracy of 66%, the ability to discriminate the types of faults between thermal and electrical types is lost.

3.2. Dornenburg

The analysis of the Dornenburg method by the confusion matrix, as shown in Figure 20, illustrates a high assertiveness of the types of failure; however, a large number of incorrect results appear in terms of not identifying the type of failure. That is, the incorrect analysis accounts for 17% and, according to

Table 10. Performance analysis of the Dornenburg method.

Failure	Performance Analysis
	Precision	Recall	F1-Score	Support
D	1.00	0.84	0.91	51
PD	0.83	0.71	0.77	7
T	1.00	0.80	0.89	30
Accuracy			0.82	88

, 12.5% of the errors were due to not being able to classify the type of failure (Not-F).

However, it is important to point out that, in the same way as the Key Gas method, the combination of the type of failure to analyze the performance of this method favors the increase in this accuracy.

Despite the good performance of this technique’s accuracy, there is evidence of difficulty in the failure analysis of the PD type, with low Precision and Recall metrics, confirming, respectively, the misclassification of two PD cases and the classification of one T1_T2 case as PD.

3.3. Rogers

The analysis of the Rogers method was divided in two ways. The first one was based on the original methodology published in [11], as shown Figure 21, and adapted considering Table 8. The second stage used the new boundaries defined in [12], as shown in Figure 22, with the default failures of [15] adjusted to this model.

The general analysis between the two versions of the Rogers method highlights the increase in the number of unidentified cases from 13 to 33 cases. Most of these cases were D1 and D2 faults that were not identified by the Rogers IEEE version.

Through

Table 11. Performance analysis of the Rogers method.

Failure	Performance Analysis
	Precision	Recall	F1-Score	Support
D1	1.00	0.29	0.45	17
D2	0.72	0.85	0.78	34
PD	0.83	0.71	0.77	7
T1_T2	0.88	0.54	0.67	13
T3	0.79	0.65	0.71	17
Accuracy			0.65	88

and

Table 12. Performance analysis of the Rogers IEEE method.

Failure	Performance Analysis
	Precision	Recall	F1-Score	Support
D	1.00	0.53	0.69	51
PD	0.80	0.57	0.67	7
T1_T2	0.75	0.46	0.57	13
T3	0.80	0.71	0.75	17
Accuracy			0.56	88

, it is possible to see the performance of the Rogers method. They show a decrease in the accuracy in relation to the original method, adapted in this article, for the grouped form in the IEEE standard.

In addition, even though the failure database was adapted to the Rogers IEEE response type, which would tend to increase accuracy, the opposite result is observed, ranging from 65% to 56%. The main reason for that is the increase in cases not classified (Not-F) and the recall index reduction.

3.4. IEC

The IEC methods were based on the methodologies published in [11,25]. The first [11] did not show the formatting of the types of failures proposed in [15], therefore, the proposed grouping in Table 8 was necessary.

The analysis of the confusion matrix for the IEC method, as shown in Figure 23, and IEC 60599 method, as shown in Figure 24, presents a balance in not identifying the type of failures from 9 to 10.

None of the faults were classified as a thermal fault of temperature < 150 °C (T0) by the IEC method.

For a comparative analysis of the performance of both methodologies, it was necessary to group the IEC methods according to Table 8, observing that both methodologies present similar accuracy, ranging from 74% to 66%.

Despite the increase in Precision in identifying D1, T1_T2, and T3 failures in IEC 60599 compared to the IEC method, a general reduction in the Recall index is also observed, which may explain the difference between the accuracies, see

Table 13. Performance analysis of the IEC method.

Failure	Performance Analysis
	Precision	Recall	F1-Score	Support
D1	0.86	0.71	0.77	17
D2	0.88	0.85	0.87	34
PD	0.83	0.71	0.77	7
T1_T2	0.70	0.54	0.61	13
T3	0.80	0.71	0.75	17
Accuracy			0.74	88

and

Table 14. Performance analysis of the IEC 60599 method.

Failure	Performance Analysis
	Precision	Recall	F1-Score	Support
D1	1.00	0.47	0.64	17
D2	0.87	0.79	0.83	34
PD	0.83	0.71	0.77	7
T1_T2	0.47	0.62	0.53	13
T3	0.83	0.59	0.69	17
Accuracy			0.66	88

.

3.5. Duval Triangle

The Duval Triangle method adopted in the analysis was T1, as it is the base analysis of failures and refinement, as explained in Section 2.

The Duval T1 method presented the best performance with an accuracy of 83%, see

Table 15. Performance analysis of the Duval triangle method.

Failure	Performance Analysis
	Precision	Recall	F1-Score	Support
D1	1.00	0.88	0.94	17
D2	0.94	1.00	0.97	34
PD	0.67	0.57	0.62	7
T1_T2	0.60	0.46	0.52	13
T3	0.82	0.82	0.82	17
accuracy			0.83	88

, and still preserved the characteristics of the type of failure.

Through the confusion matrix, as shown in Figure 25, it is possible to identify that among the observed failures, there was an error in the classification of two failures: T1_T2 as N, three T3 as N, two T1_T2 as PD, two T3 as T2, and three T1_T2 as T3.

With the exception of the PD and T1_T2 failures, there is a balance between the values of the F1-score metric, reflecting the high precision and recall of these indicators.

3.6. Duval Pentagon

As shown in the confusion matrix, seen in Figure 26, the Duval Pentagon method 1 performed similarly to the Duval Triangle T1. The general accuracy of this method, shown in

Table 16. Performance analysis of the Pentagon Duval 1 method.

Failure	Performance analysis
	Precision	Recall	F1-Score	Support
D1	0.84	0.94	0.89	17
D2	1.00	0.88	0.94	34
PD	0.83	0.71	0.77	7
T1_T2	0.78	0.54	0.64	13
T3	0.78	0.82	0.80	17
accuracy			0.82	88

, was 82%, which is considered adequate since it still preserves the description between the failure types.

4. Discussion

The performance analysis between techniques was plotted for each kind of failure accuracy and for each technique, presented in the star graph below. Figure 27 shows the result of this comparison, based on the analysis of the response of each technique for the type of failure standardized in [15].

A comparative performance analysis of the techniques makes it clear that the Duval triangle T1 with Duval Pentagon 1 presents the best performance, and that all techniques present difficulties in identifying T1_T2 and PD failures.

Regarding T1_T2 failures, the technique that presents the best performance is the Dornenburg technique.

With regard to type D2 failures, except for the Duval Triangle T1 method, which presented excellent performance, and Rogers and Rogers IEEE, which had the worst performance, the others presented very similar performances.

The failures of types D1 and T3, in addition to the Duval Triangle and Pentagon methods, highlight the good performance of the Dornenburg and Key Gas methods.

Finally, analysis of the performance of the techniques in the PD type failure highlights the performance of the key gas method.

This kind of analysis, despite the high performance of the Duval Triangle T1 and Duval Pentagon 1 method, emphasizes the importance of analyzing the failure by combining more than one technique to increase the performance in correctly identifying the type of failure. In addition, the visualization of the techniques through a star graph may be a good tool to aid in the selection of which techniques to combine depending on the possible failures in the analysis.

5. Conclusions

Through this article, it was possible to recognize the main steps of the process that involves the science of gas analysis in power transformers insulated with mineral oil, from the knowledge of sampling to the analysis of the results found.

Emphasis was placed on the evolution of the main methods adopted in international standards with general knowledge of the evolution of research in this area. In addition, two recently published techniques, TRT and Pentagon, were presented in an illustrative way, presenting alternative forms of analysis and, in the case of TRT, with great performances.

The way of presenting the results of conventional techniques with confusion and performance matrices allows for a clearer visualization of the strengths and weaknesses of each technique depending on the type of failure.

In addition, a comparative analysis between the star graph techniques brought a form of comparative visualization to the performance of the techniques, helping to choose to combine more than one technique depending on the type of failure under analysis

From the analysis of the results of conventional techniques, it is reaffirmed that the T1 Duval Triangle technique, currently the most popular one, presents the best answers in terms of accuracy associated with discriminating the types of failure. This is crucial when professionals are trying to decide whether to stop the equipment in operation or not.

Finally, in order to support other researchers and other professionals in the field of conventional gas analysis development or even artificial intelligence tools, it is expected that this study will be a supporting document and support the comparison in the analysis of improvements or results found by such methods.