Together We Power The World
Techniques for Interpretation of Data for DGA From Transformers Lance Lewand, Doble Engineering
Purpose of DGA • To pr provi ovide de a non non-in -intru trusiv sivee mean meanss to to dete determi rmine ne if a transformer incipient fault condition exists or not – To Too o cons conser erva vati tive ve – Too li liberal • To ha have ve a hig high h pro probab babili ility ty th that at wh when en en enter terin ing g an an transformer a problem is apparent • To pr prev even entt an an une unexp xpec ecte ted d out outag agee • To red reduc ucee ris risk k to to the the un unit it an and d the the sy syste stem/c m/comp ompan any y
2006 IEEE Conference
Purpose of DGA • To pr provi ovide de a non non-in -intru trusiv sivee mean meanss to to dete determi rmine ne if a transformer incipient fault condition exists or not – To Too o cons conser erva vati tive ve – Too li liberal • To ha have ve a hig high h pro probab babili ility ty th that at wh when en en enter terin ing g an an transformer a problem is apparent • To pr prev even entt an an une unexp xpec ecte ted d out outag agee • To red reduc ucee ris risk k to to the the un unit it an and d the the sy syste stem/c m/comp ompan any y
2006 IEEE Conference
Interpretation Techniques • Inc Incip ipien ientt Faul Faultt Typ Types, es, Fra Frank nk M. Cl Clark ark,, 193 1933/ 3/19 1962 62 • Dö Dörn rnen enbu burg rg Ra Rati tios os,, E. E. Dör Dörne nenb nbur urg, g, 19 1967 67,, 197 1970 0 • Po Pott tth hoff ff’s ’s Sc Sch hem eme, e, K. Po Pottthof off, f, 19 196 69 • Ab Abso solu lute te lim limit its, s, var vario ious us sou sourc rces es,, earl early y 1970 1970ss • Sh Shan ank’ k’ss Vis Visua uall Cur Curve ve me meth thod od,, 197 1970s 0s • Trilinear Plot Method, 1970s • Ke Key y Gas Gas Me Meth thod od,, Dav David id Pu Pugh gh,, 197 1974 4 • Du Duva vall Tri Trian angl gle, e, Mi Mich chel el Du Duva val, l, 19 1974 74 2006 IEEE Conference
Interpretation Techniques • Rogers Ratios, R.R. Rogers, 1975 • Glass Criterion, R.M Glass, 1977 • Trend Analysis, various sources, early 1980s – total volume per day – ppm per day
• Church Logarithmic Nomograph, J.O. Church, 1980s • Expert System Analysis, Richard Lowe, 1985
2006 IEEE Conference
Interpretation Techniques • Expert System Monitor Program, Karen Barrett, 1989 • Transformer Fingerprinting • IEEE C57.104, Limits, rates and TDCG, 1978/1991 • Artificial Neural Networks (ANNs) and Fuzzy Logic – X. Ding, E. Yao, Y. Liu and Paul Griffin, 1996 – Vladimiro Miranda and Adriana Garcez Castro, 2004 – Donald Lamontagne, 2006 2006 IEEE Conference
Interpretation Techniques • IEC 60599 Ratios, Limits and gassing rates, 1999 • Datamining and Log Transformation, Tony McGrail, 2000 • Vector Algorithm, Nick Dominelli, Mike Lau & David Pugh, 2004
2006 IEEE Conference
Most Commonly Used • • • • • • • • • •
Duval Triangle IEEE C57.104, Limits, rates and TDCG Straight Limits Key Gas Method Dörnenburg Ratios Rogers Ratios IEC 60599 Ratios and Limits Trend Analysis Fingerprints Expert System Analysis 2006 IEEE Conference
Dissolved Gas Acceptable Limits Various Sources H2 *IEEE
**Electra (CIGRE) IEC 60599 Typical Range Manufact.
CO
100 350 101-700 351-570 701-1800 571-1400 >1800 >1400 28.6 289
CH4
C2H6
C2H4
C2H2
CO2
TCG
120 121-400 401-1000 >1000 42.2
65 66-100 101-150 >150 85.6
50 51-100 101-200 >200 74.6
35 36-50 51-80 >80 --
2500 2500-4000 4001-10000 >10000 3771
720 721-1920 1921-4630 >4630 520
60-150
540-900
40-110
50-90
60-280
3-50
5100-13000
200 (250)
500 (1000)
100 (200)
100 (200)
150 (300)
15 (35)
---
1065 1985
*IN THE PROCESS OF BEING REVISED **CORRECTED VALUES 1978 ( ) VALUE 6 – 7 YEARS
2006 IEEE Conference
Key Gases - Arcing 100 90 80
% , s e l b i t s u b m o C
70 60 50 40 30 20 10 0 CO
H2
CH4
C2H6
C2H4
C2H2
2006 IEEE Conference
Key Gases - Overheating, Oil 100 90 80
% , s e l b i t s u b m o C
70 60 50 40 30 20 10 0 CO
H2
CH4
C2H6
C2H4
C2H2
2006 IEEE Conference
Key Gases - Partial Discharge 100 90 80
% , s e l b i t s u b m o C
70 60 50 40 30 20 10 0 CO
H2
CH4
C2H6
C2H4
C2H2
2006 IEEE Conference
Key Gases - Overheating, Paper 100 90 80
% , s e l b i t s u b m o C
70 60 50 40 30 20 10 0 CO
H2
CH4
C2H6
C2H4
C2H2
2006 IEEE Conference
Dörnenburg Ratio Method • Started out as only two ratios – CH4 /H2 – C2H2 /C2H4 – plotted on a log-log scale. The areas corresponded to thermal deterioration, arcing and partial discharge – too many faults missed - went to 4 ratios
• Ratio 1 (R1)=CH4 /H2 • Ratio 2 (R2)=C2H2 /C2H4 • Ratio 3 (R3)=C2H2 /CH4 • Ratio 4 (R4)=C2H6 /C2H2 2006 IEEE Conference
Dörnenburg Ratio Method • Used to determine 3 general fault types – Thermal faults – Electrical Faults, low intensity discharges – Electrical Faults, high intensity arcing
2006 IEEE Conference
Dörnenburg Ratio-Minimum (Dörn rnen enbu burg rg & IEEE Leve Levels) ls) Gas Levels (Dö
Hydrogen
2 00
1 00
Methane
50
12 0
100 0
350
Acetylene
15
35
Ethylene
60
50
Ethane
15
65
Carbon Monoxide
2006 IEEE Conference
Dörrne Dö nenb nbur urg g Ra Rati tio o • Cri ritter eriia fo for app appllic icaati tio on - a fa faul ultt exi exist stss – On Onee Gas Gas > 2 x min minimu imum m leve levell – At lest lest one gas > mini minimum mum le level vel
• De Dete term rmin inee Val Valid idit ity, y, L1 no norm rm te test st – On Onee gas gas in eac each h ratio ratio > mini minimum mum
• Co Comp mpar aree rat ratio ioss to to Fau Fault lt Di Diag agno nosi siss Tab Table le • Al Alll fall fall wit withi hin n one one cond condit itio ionn-va vali lid d diag diagno nosi siss
2006 IEEE Conference
Dörnenburg Ratio-Fault Diagnosis Table, from the oil
R1 CH4/H2
R2 R3 R4 C2H2/C2H4 C2H2/CH4 C2H6/C2H2
1-Thermal Decomp
>1.0
<0.75
<0.3
>0.4
2-Low Intensity PD
<0.1
Not Sig
<0.3
>0.4
>0.1,<1.0
>0.75
>0.3
<0.4
3-Arcing
Valid only if all the ratios for a particular fault type are met.
2006 IEEE Conference
Dörnenburg Flowchart
From IEEE C57.104 - 1991 2006 IEEE Conference
Initial Roger’s Ratios • Took information from Halstead’s thermal equilibrium and Dörnenberg ratios along with information from faulted units • Originally developed four ratios – CH4 /H2 – C2H6 /CH4 – C2H4 /C2H6 – C2H2 /C2H4 • Came up with a 4 number code that identified 11 incipient fault conditions and a normal condition 2006 IEEE Conference
Halstead’s Thermal Equilibrium
2006 IEEE Conference
Initial Roger’s Ratios Ratio
Range
Code
CH4 /H2
≤
0.1 >0.1 <1 ≥1 <3 ≥3 <1 ≥1
5 0 1 2
<1 ≥ 1 <3 ≥3 < 0.5 ≥0.5 <3 ≥3
0 1 2
C2H6 /CH4 C2H4 /C2H6 C2H2 /C2H4
0 1
0 1 2 2006 IEEE Conference
Roger’s Fault Diagnosis Table CH4 /H2 C2H6 /CH4 C2H4 /C2H6 C2H2 /C2H4 Diagnosis
0 5 ½ ½ 0 0 1 1
0 0 0 1 1 0 0 0
0 0 0 0 0 1 1 2
0 0 0 0 0 0 0 0
0 0 0 5
0 0 0 0
0 ½ 2 0
1 ½ 2 ½
Normal Partial Discharge Slight Overheating – below 150 °C Slight Overheating –150 °C to 200°C Slight Overheating –200 °C to 300°C General conductor overheating Winding circulating currents Core and tank circulating currents, overheated joints Flashover without power follow through Arc with power follow through Continuous sparking to floating potential Partial discharge with tracking
2006 IEEE Conference
Refined Roger’s Ratio • Three ratios – Ratio 1 (R1)=CH4 /H2 – Ratio 2 (R2)=C2H2 /C2H4 – Ratio 5 (R5)=C2H4 /C2H6
• No minimum levels – suggested when normal levels exceeded
2006 IEEE Conference
Refined Roger’s Ratio-Fault Diagnosis Case
R2 C 2H 2/C 2H 4
R1 CH 4/H 2
R5 C 2H 4/C 2H 6
Fault
0
<0.1
>0.1,<1.0
<1.0
Normal
1
<0.1
<0.1
<1.0
2
0.1-3.0
0.1-1.0
>3.0
Low energy PD Arcing
3
<0.1
>0.1<1.0
1.0-3.0
4
<0.1
>1.0
1.0-3.0
5
<0.1
>1.0
>3.0
Low temp thermal Thermal <700 C Thermal >700 C
2006 IEEE Conference
Roger’s Ratios Flowchart
From IEEE C57.104 - 1991
2006 IEEE Conference
IEC 60599 • Identifies 6 different fault types – PD: Partial Discharge – D1: Discharge of low energy – D2: Discharge of high energy – T1: Thermal fault, t <300°C – T2: Thermal fault, 300°C < t < 700 °C – T3: Thermal fault, t > 700 °C
• Uses a combination of ratios (based on Roger’s Ratios), gas concentrations and rates of gas increase 2006 IEEE Conference
IEC 60599 Ratio-Fault Diagnosis R2 C 2 H 2 /C 2 H 4
R1 C H 4 /H 2
R5 C 2 H 4 /C 2 H 6
Fault
NS
<0.1
<0.2
PD
>1
0.1-0.5
>1
0.6-2.5
0.1-1
>2
NS
>1 (N S)
<1
D 1 -Low energy D 2 –H igh energy T 1 <300C
<0.1
>1
1-4
<0.2
>1
>4
T2 >300C <700 C T herm al >700 C
NS = not significant regardless of value
Concentrations should be 10 x S (MDL) 2006 IEEE Conference
IEC 60599 Rates of gas increase • >10% increase per month above typical levels = active fault • >50% per week or evolving faults of higher energy = serious
2006 IEEE Conference
IEC 60599 Typical Gas Levels
IEC 60599 Typical Range Communi cating OLTC
H2
CO
CH4
C2H6
C2H4
C2H2
CO2
60-150
540-900
40-110
50-90
60-280
3-50
5100-13000
75-150
400-850
35-130
50-70
110-250
80-270
5300-12000
Note in IEC 60599: Typical values are higher in sealed transformers than free breathing transformers
2006 IEEE Conference
Ratio Methods • Advantages – quantitative – independent of oil volume – can be computer programmed
• Disadvantages – don’t always yield an analysis – not always correct – dependence of preservation system – Dornenburg has fallen out of favor because it misses too many incipient faults
2006 IEEE Conference
Ratio Methods • Solid insulation handled separately using carbon monoxide and carbon dioxide ratios
2006 IEEE Conference
Trend Analysis • Historical Information – Has the percent TCG in the gas space risen suddenly? – Has the percent TCG in the oil risen suddenly? – Nameplate information – How old in the transformer?
2006 IEEE Conference
Trend Analysis – Did a bushing fail at some point? – Did the transformer fail previously? – If the unit has been repaired and was the oil filtered or degassed? – Is the unit heavily loaded or overloaded? – Previous dissolved gas-in-oil test?
2006 IEEE Conference
Transformer Fingerprints
GAS (PPM) Hydrogen Methane Carbon Monoxide Ethane Carbon Dioxide Ethylene Acetylene
Initial 350 44 670 26 3000 9 --
3 260 61 650 25 1900 5 --
Initial 3 110 210 11 13 520 630 3 4 5000 3900 8 10 ---
2006 IEEE Conference
Transformer Fingerprints GAS (PPM) Hydrogen Methane Carbon Monoxide Ethane Carbon Dioxide Ethylene Acetylene
Initial 0 92 370 2300 6000 180 0
3 1 69 400 2300 6800 180 0
Initial 3 0 0 15 18 33 57 560 520 1800 2200 9 6 0 0
2006 IEEE Conference
Carbon Oxide Gases and Ratios Cellulose Insulation • Shell form > CO2 than core form - due to mass • Accidental CO2 • CO2/CO : 3 -14:1 • CO2 /CO Avg. 7:1 • Approach 1 high temperature faults • High CO2 with low CO-lack of cooling/general overheating 2006 IEEE Conference
Pitfalls • Gases produced not as a result of incipient fault condition – Leaking between tap changers and main tank – lower voltage transformers having higher CO and CO2 values as a result of non-vacuum Hitreatment – Welding producing acetylene and other gases – Out-gassing of paints and gaskets, usually CO and CO2 – Stray gassing characteristics 2006 IEEE Conference
Pitfalls • Incipient Faults not really covered – production of hydrogen from overheated oil thin films on core laminations (>140°C) – Oxidation and thermal heating of the oil causing the production of CO and CO 2
• Gases produced not as a result of incipient fault condition – Leaking between the tap changer and main tank 2006 IEEE Conference
Pitfalls – Galvanic reactions (steel + water + O2 = hydrogen production) – lower voltage transformers having higher CO and CO2 values as a result of non-vacuum treatment, oxygen + heat – Welding producing acetylene and other gases – Out-gassing of paints, gaskets & polymers, usually CO and CO 2
2006 IEEE Conference
Pitfalls – Stray gassing characteristics (highly refined oils ⇒ H2) – Contaminants produce gases – Decomposition of additives such as passivators can produce gases as well (H 2 and CO2)
2006 IEEE Conference
In Reality - Expert Systems are Used • History • Key gases • Ratios • Fingerprints - similar populations • Trend analysis • Internal databases • Total combustible gas • Rate of gas generation • A human expert
Use the tools in the toolbox, not just one!!! 2006 IEEE Conference
THANK YOU FOR YOUR ATTENTION
2006 IEEE Conference
IEEE/PES Transformer Committee Montreal, Canada Tuesday, October 24, 2006
Dissolved gas analysis and the Duval Triangle
by Michel Duval
-DGA is for Dissolved Gas Analysis. -DGA is probably the most powerful tool for detecting faults in electrical equipment in service. -Over one million DGA analyses are performed each year by more than 400 laboratories worldwide.
-Gases in oil always result from the decomposition of electrical insulation materials (oil or paper), as a result of faults or chemical reactions in the equipment. -for example, oil is a molecule of hydrocarbons, i.e., containing hydrogen and carbon atoms, linked by chemical bonds (C-H, C-C).
-some of these bonds may break and form H*, CH3*, CH2* and CH* radicals.
All these radicals then recombine to form the fault gases observed in oil:
-in addition to these gases, the decomposition of paper produces CO2, CO and H2O, because of the presence of oxygen atoms in the molecule of cellulose:
The main gases analyzed by DGA Hydrogen
H2
Methane
CH4
Ethane
C2H6
Ethylene
C2H4
Acetylene
C2H2
Carbon monoxide CO Carbon dioxide
CO2
Oxygen
O2
Nitrogen
N2
-some of these gases will be formed in larger or smaller quantities depending on the energy content of the fault. -for example, low energy faults such as corona partial discharges in gas bubbles, or low temperature hot spots, will form mainly H 2 and CH4.
-faults of higher temperatures are necessary to form large quantities of C2H4. -and finally, it takes faults with a very high energy content, such as in electrical arcs, to form large amounts of C2H2. -by looking at the relative proportion of gases in the DGA results it is possible to identify the type of fault occurring in a transformer in service.
Gas formation patterns -are related only to the materials used and faults involved. -are the same in all equipment where these materials are used (e.g., sealed or air-breathing power transformers, reactors, instrument transformers, LTCs, etc).
Standards/ Guides for the interpretation of DGA: -IEC Publication 60599 (1999). -IEEE Guide C57.104 (1991) (under revision).
Other useful information in: -IEEE EI.Mag., Apr. 2001, June 2002, Aug. 2005. -CIGRE Brochure # 296 (2006).
6 basic types of faults detectable by DGA have been defined by the IEC: 1.Partial discharges of the corona-type (PD). -typical examples: discharges in gas bubbles or voids trapped in paper, as a result of poor drying or poor oil-impregnation.
2.Discharges of low energy (D1) -typical examples: partial discharges of the sparking-type, inducing carbonized punctures in paper. -or low-energy arcing, inducing surface tracking of paper and carbon particles in oil.
3.Discharges of high energy (D2) -typical examples: high energy arcing, flashovers and short circuits with power followthrough, resulting in extensive damage to paper, large formation of carbon particles in oil, metal fusion, tripping of the equipment or gas alarms .
4.Thermal faults of temperatures < 300 °C (T1) Faults T1 are evidenced by paper turning: -brown (> 200 °C). -black or carbonized (> 300 °C). Typical examples: overloading, blocked oil ducts
5.Thermal faults of temperatures between 300 and 700°C (T2) Faults T2 are evidenced by : -carbonization of paper. -formation of carbon particles in oil. Typical examples: defective contacts, defective welds, circulating currents.
6.Thermal faults of temperatures > 700°C (T3) Faults T3 are evidenced by : -extensive formation of carbon particles in oil. -metal coloration (800 °C) or metal fusion (> 1000 °C). Typical examples: large circulating currents in tank and core, short circuits in laminations.
Several diagnosis methods have been proposed to identify these faults in service. The first one was the Dornenburg method in Switzerland in the late 1960s, then the Rogers method in UK in the mid 1970s. Variations on these methods have later been proposed by the IEC (60599) and IEEE.
All these methods use 3 basic gas ratios: (CH 4 /H2, C2H2 /C2H4 and C2H6 /C2H4). Depending on the values of these gas ratios, codes or zones are defined for each type of fault. One drawback of these methods is that no diagnosis can be given in a significant number of cases, because they fall outside the defined zones.
Another method used by IEEE is the so-called keygas method, which looks at the main gas formed for each fault, e.g, C 2H2 for arcing. One drawback of this method is that it often provides wrong diagnoses.
Finally, there is the Triangle method, which was developed empirically in the early 1970s, and which is used by the IEC. It is based on the use of 3 gases (CH 4, C2H4 and C2H2) corresponding to the increasing energy levels of gas formation. One advantage of this method is that it always provides a diagnosis, with a low percentage of wrong diagnoses.
Comparison of diagnosis methods. % Wrong diagnoses 58
% Total
Key gases
% Unresolved diagnoses 0
Rogers
33
5
38
Dornenburg 26
3
29
IEC
15
8
23
Triangle
0
4
4
58
The triangle representation also allows to easily follow graphically and visually the evolution of faults with time. However, many people are not quite familiar with the use of triangular coordinates, so I will try to explain that in more detail today.
The triangle method.
The triangle method plots the relative % of CH 4, C2H4 and C2H2 on each side of the triangle, from 0% to 100%. The 6 main zones of faults are indicated in the triangle, plus a DT zone (mixture of thermal and electrical faults).
FAQ: How fault zones have been defined in the Triangle ?
Answer: Fault zones are based on a large number of cases of faulty transformers in service which have been inspected visually.
Cases of faults PD and D1
tracking; sparking; small arcing.
Cases of faults D2
Cases of thermal faults in oil only
circulating currents ; laminations ; bad contacts
Cases of thermal faults in paper
brownish paper ; carbonized paper ; not mentioned
FAQ: how corona PDs, which form a lot of H 2, can be identified in the Triangle without using this gas ?
Answer: in such faults, CH 4 is indeed formed in smaller amounts than H2 (typically 10 to 20 times less), but can still be measured easily by DGA.
FAQ: in the Triangle, why not use H 2 rather than CH4 to represent low energy faults ? Answer: because CH4 provides better overall diagnoses for all types of faults. A possible explanation (?): H 2 diffuses much more rapidly than hydrocarbon gases from transformer oil. This will affect gas ratios using H 2 but not those using hydrocarbon gases.
FAQ: So, how to use the triangle ? If for example the DGA lab results are: CH4 = 100 ppm C2H4 = 100 ppm C2H2 = 100 ppm First calculate: CH4 + C2H4 + C2H2 = 300 ppm.
Then calculate the relative % of each gas: relative % of CH4 = 100 / 300 = 33,3 % relative % of C2H4 = 100 / 300 = 33,3 % relative % of C2H2 = 100 / 300 = 33,3 % These values are the triangular coordinates to be used on each side of the triangle. To verify that the calculation was done correctly, the sum of these 3 values should always give 100%, and should correspond to only one point in the triangle.
Each DGA analysis received from the lab will always give only one point in the triangle.
The zone in which the point falls in the Triangle will identify the fault responsible for the DGA results.
The calculation of triangular coordinates can easily be done manually, or with the help of a small algorithm or software. Errors are often made when developing such an algorithm, so check it first with the free algorithm available from me (
[email protected]).
For those familiar with computer graphics, it is also possible to develop a software displaying the point and the fault zones graphically in the triangle. Several software packages are available for DGA interpretation using the triangle method
.
The Triangle, being a graphical method, allows to easily follow the evolution of faults with time, for instance from a thermal fault to a potentially much more severe fault such as D2.
.
The most severe faults: -faults D2 in paper and in oil (high-energy arcing) -faults T2-T3 in paper (>300 °C) -faults D1 in paper (tracking, arcing) -faults T3 in oil (>700 °C)
The less severe faults: -faults PD/ D1 in oil (sparking) -faults T1 in paper (<300 °C) -faults T2 in oil (<700 °C) -are difficult to find by inspection
A fault in paper is generally considered as more serious than a fault in oil only, because paper is often placed in a HV area (windings, barriers). A popular ratio used to detect paper involvement is the CO2 / CO ratio. If the CO2 / CO ratio is < 3, this is a strong indication of a fault in paper, either a hot spot or electrical arcing.
The CO2 / CO ratio, however, is not very accurate, because it is also affected by the background of CO2 and CO coming from oil oxidation.
The amounts of furans in oil may also be used in some cases to confirm paper involvement, however, the interpretation of results is often difficult.
Other useful gas ratios: . -O / N : a decrease of this ratio indicates excessive 2 2
heating (< 0.3 in breathing transformers).
-C2H2 / H2 : a ratio > 3 in the main tank indicates contamination by the LTC compartment
Gassing not related to faults in service: .-Catalytic reactions on metal surfaces: formation
of H2 only. -“Stray” gassing of oil: the “unexpected gassing of oil at relatively low temperatures (80 to 200 °C)”: gassing of the T1 or T2 type.
Minimum gas formation to attempt a diagnosis: -first limit is related to lab accuracy. -second limit to economic reasons.
First limit: lab accuracy The accuracy of the “average” CIGRE /IEC lab is ~ ± 15% at medium (routine) gas concentrations (> 10 ppm for hydrocarbons).
Its accuracy decreases to ~ ± 30% at 6 ppm, and ± 100% near the lab detection limit (2 ppm).
Effect of lab accuracies of 15 and 30% on DGA diagnosis uncertainty (in red and blue).
When an area of uncertainty crosses several fault zones in the triangle, a reliable diagnosis cannot be given. This is particularly true for lab accuracies > 30%.
Diagnosis uncertainty corresponding to lab inaccuracies of ± 15, 30, 50 and 75 %:
This applies not only to the triangle but to all diagnosis methods.
How inaccurate are the laboratories at medium gas concentrations ?
How inaccurate are at low low gas gas co conc ncen entr trat atio ions ns ?
Minimum gas concentrations to attempt a diagnosis. If for example lab accuracy is ±15% at medium gas levels (>10 ppm): If some gases are < 6 ppm, diagnoses will be uncertain, and a calculation of diagnosis uncertainty should be done. Commercial software is available for that purpose.
If lab accuracy is between 15% and 30%, diagnoses will be uncertain at all gas concentrations, and a calculation of diagnosis uncertainty necessary. Above 30% or 50%, diagnoses become too uncertain. Lab and gas monitor accuracies can be obtained by using gas-in-oil standards. Such standards are available commercially.
Second limit: typical values A recommendation of CIGRE and the IEC is that DGA diagnosis should be attempted only if gas concentrations or rates of gas increase in oil are high enough to be considered significant. Low gas levels may be due to contamination or aging of insulation, not necessarily to an actual fault.
Also, there is always a small level of gases in service, and it would not be economically viable to suspect all pieces of equipment. So, it is better to concentrate on the upper percentile of the transformer population with the highest gas levels.
This is the philosophy behind the use of 90% typical concentrations and 90% typical rates of increase, in order to concentrate maintenance efforts on the 10% of the population most at risk. A consensus has been reached at CIGRE on typical values observed in service worldwide (CIGRE Brochure # 296, 2006).
Ranges of 90 % typical concentration values for power transformers, in ppm: C2H2 All transformers
No OLTC
2-20
Communicating OLTC
60-280
H2
CH4
C2H4
C2H6
CO
CO2
50150
30130
60280
2090
400600
380014000
Ranges of 90 % typical rates of gas increase for power transformers, in ppm/year: C2H2 All transformers No OLTC
0-4
Communicating OLTC
21-37
H2
CH4
C2H4
C2H6
CO
CO2
35132
10120
32146
590
2601060
170010,000
90% typical values are within the same range on all networks, with some differences related to individual loading conditions, equipment used, manufacturers, climate, etc. Each individual network therefore should preferably calculate its own specific typical values.
Influence of some parameters on typical values: -Typical values are significantly higher in young equipment (suggesting there are some unstable chemical bonds in new oil and paper ?). -A bit higher in very old equipment. -Significantly lower in instrument transformers. -Higher in shell-type and shunt reactors (operating at higher temperatures ?).
-Typical values are not affected by oil volume (suggesting that larger faults are formed in larger transformers ?). -Typical values are very similar in air-breathing and in sealed or nitrogen blanketed equipment, contrary to a common belief in the US.
90% typical values in California vs. CIGRE values, in ppm: C2H2
H2
CH4 C2H4 C2H6
CO
CO2
CIGRE/ IEC
220
50150
30130
60280
2090
400- 3800600 14000
California
3
96
88
57
79
613
5991
When DGA results in service reach typical values: -a diagnosis may be attempted to identify the fault (if lab accuracy is good enough). -the equipment should not be considered at risk. -however, it should be monitored more frequently by DGA.
To evaluate how much at risk a transformer may become above typical values, the probability of failure in service (PFS) has to be examined. PFS has been defined as the number of DGA analyses followed by a failure-related event (e.g., tripping, fault gas alarm, fire, etc), divided by the total number of analyses, at a given gas concentration.
Probability of having a failure-related event ( PFS, % ) vs. the concentration of C2H2 in ppm at HQ 90
98
99 Norm, in %
PFS, in %
100
300
400
ppm
The PFS remains almost constant below and above the 90% typical value, until it reaches an inflexion point on the curve (pre-failure value).
DGA monitoring should be done more and more frequently as gas concentrations increase from typical to pre-failure value.
Pre-failure concentration values were found by CIGRE to be surprisingly close on different networks: H2
CH4 C2H4 C2H6 C2H2 CO
2401320
270460
700990
7501800
310600
9843000
(in ppm)
This suggests that failure occurs when a critical amount of insulation is destroyed.
In-between typical and pre-failure values, specific alarm values can be defined, depending on the tolerance to risk of the maintenance personnel, and on the maintenance budget available. For example, higher alarm values may be used when the maintenance budget is low, and lower alarm values in the case of strategic equipment.
Pre-failure rates of gas increase (slope 3) are in preparation at CIGRE. Concentration
Time
Pre-failure rates of gas increase in power transformers, in ppm/ day C2H2
H2
CH4
C2H4
C2H6
CO
CO2
0.5
3
5
5
11
NS
NS
On-line gas monitors -are best suited for measuring rates of gas increase (trends). -will detect faults between regular oil samplings. -may now also provide on-line diagnosis.
The triangle can also be used to identify faults in tap changers.
: Normal operation; :Severe coking; : Light coking; : strong arcing D2; : Arcing D1
:
“Heating”;
Thanks a lot for your attention.
An Artificial Neural Networks Approach to Transformer Dissolved Gas Analysis and Problem Notification Donald Lamontagne Section Leader T&D Reliability Analysis and Management Arizona Public Service EPRI Substation Equipment Diagnostic Conference XIV Marriott Hotel and Marina San Diego, CA July 17, 2006
Agenda
Events On-Line DGA Monitoring Neural Networks APS TOAN System Conclusions Questions?
Events
Westwing 6/14/2004 and 7/4/2004 Events
6/14/2004
Sustained fault on 230kV Westwing – Liberty line One breaker failed to open Initial fault split between three banks Communication error on breaker status Last fault through one bank only Post event DGA and thermography
Damaged Transformers
Five 500MVA, Single Phase, 525/230/13.8kV Autotransformers w/ LTC Westinghouse 1973 vintage 14,500 gals of oil in the main tank
Damaged Phases
7/5/2004
Deer Valley
7/20/2004 – T928 Type U bushing failure
167MVA, three phase, 230/69kV FPE 1978 vintage Bushing was Doble tested in 2002 with no issues
Replacement T873
167MVA, three phase, 230/69kV Westinghouse 1979 vintage Removed from service 5/2004 for upgrade to 188MVA Returned to service 7/25/2004 to replace failed T928
T873 DGA Results O2
N2
CO2
CO
H2
CH4
C2H6
C2H4
C2H2
3/26/2004
627
59261
2131
17
9
3
3
2
0
8/18/2004
750
4637
1015
54
13
37
36
3
0
3/28/2005
2734
66252
806
41
3922
446
70
617
2635
All gases from the 8/18/2004 sample were below the IEEE C57.104 “Condition 1” levels – indicating the transformer was behaving normally. The 3/28/2004 sample has H2, C2H4, C2H2 and TDCG at “Condition 4” and CH4 at “Condition 3.”
On-Line DGA Monitoring
On-Line DGA Monitoring
Began utilizing in the summer of 2003 Currently using Serveron TrueGas and TM8 models Continuously sample eight gases (hydrogen, acetylene, methane, ethane, ethylene, CO, CO 2, O2) and report every four hours through gas chromatography Currently installed on fifty-two 230kV and above transformers and shunt reactors.
Source: www.serveron.com
Laboratory Grade Gas Chromatography Gas Accuracy
Repeatability
Hydrogen H2
±5% or ±3 ppm
<2%
Oxygen O2
±5% or +30/-0 ppm
<1%
Methane CH4
±5% or ±5 ppm
<1%
±5% or ±5 ppm
<2%
±5% or ±5 ppm
<1%
Ethylene C2H4
±5% or ±3 ppm
<1%
3-5,000 ppm
Ethane C2H6
±5% or ±5 ppm
<1%
5-5,000 ppm
±5% or ±1 ppm
<2%
Carbon Monoxide CO Carbon Dioxide CO2
Acetylene C2H2
Range 3-3,000 ppm 30-25,000 ppm 5-7,000 ppm 5-10,000 ppm 5-30,000 ppm
1-3,000 ppm
Artificial Neural Networks
Artificial Neural Networks
A network of nodes and weighted connections, which are loosely analogous to the neurons and synapses in the brain. Each node sums the inputs from several incoming weighted connections and then applies a transfer function to the sum. The transfer function is a smooth, non linear function
logistic function hyperbolic tangent
Neural Networks i1
∑
∑
i2
∑
∑ ∑
in Input Layer
∑
∑
Hidden Layer 1
Hidden Layer 2
Output Layer
Neural Network Training
Underfitting and Overfitting
x x
x
x
xx
x
x x
x x
x
x
xx
x x
x
Underfitting
x
x x
x
x
xx
x x
x
Correct Fit
x
Overfitting
x
APS TOAN (Transformer Oil Analysis and Notification)
Traditional Analysis
Testing accuracy of traditional methods
Diagnosis Methods Dornenberg Ratio
Success 22.9%
Error 65.2%
Not Identifiable 11.9%
Rogers Ratio
24.8%
12.4%
62.9%
IEC 599
42.9%
24.8%
32.4%
APS TOAN
~ 114,000 DGA samples/year Utilizes VT’s ANNEPS engine (w/ modifications) ANN combined with Expert System
Tested at ~ 93% accuracy in predicting fault type Exception based processing system
APS TOAN
Some Modifications to VTs System: Gassing rates Nine vs. eight gases Minimum gas levels Added a Polling Engine Added a Notification Engine
TOAN Provides Answers
Who – Transformer ID When – When the sample was taken? What – What are the gas values and what type of fault is it? How – How severe is it? Where – Where is the fault likely located?
Example Report Who and When Transformer
T629 (Four Corners 350KV) [ Level = 1 : IMMEDIATE ATTENTION ]
Description
FC3 U4 GSU SO. XFMR 345Y/199.186-22KV, 308MVA 1-P
Taken By
SERVERON
Sample ID
Sample Date
Sample Received Date
Days
7412
6/5/2006 5:00:00 PM
6/5/2006 5:58:25 PM
0.167
7410
6/5/2006 1:00:00 PM
6/5/2006 2:58:09 PM
What Gas in Oil Current Sample
Previous Sample
Delta
Rate (ppm/day)
Hydrogen
35.2
34.7
0.5
-13.806
Methane
++ 543.8
542.1
1.7
-21.259
Acetylene
++ 15.7
24.6
-8.9
0.063
Ethylene
++ 1875.4
1888.4
-13.0
-21.445
Ethane
++ 341.5
342.7
-1.2
-0.896
CO
++ 503.8
503.2
0.6
-9.113
CO2
++ 6695.4
6681.3
14.1
0.260
O2
10613.4
10671.6
-58.2
-3.017
N2
59346.8
58284.8
1062.0
37.924
THG
++ 2776.4
2797.8
-21.4
-36.221
TDCG
++ 3315.4
3335.7
-20.3
-52.800
TCG%
1.0
1.0
0.0
-0.093
What Fault Analysis ANN
EPS
Combined
Normal - NR
0.000
0.010
0.000
High Energy Discharge - HEDA
0.000
0.010
0.000
Low Energy Discharge - LED
0.000
0.010
0.000
Overheating - OH
1.000
0.990
1.000
Overheating of Oil - OHO
1.000
0.990
1.000
Cellulose Degradation - CD
0.833
0.990
0.833
Duval Analysis % CH4
% C2H2
% C2H4
Total Gas
22.3
0.6
77.0
2434.9
Duval Diagnosis
T3 - Thermal fault > 700degC
What Diagnosis Final Recommendation
1
Previous Result
1
Recommended Condition
Overall condition needs IMMEDIATE ATTENTION.
Recommended Action
Sample oil daily. Consider removal of unit from service. Advise manufacturer.
Simple Criteria
Unit is ABNORMAL
OH Temperature
Est. temp is above 700 c degrees.
HEDA Severity HEDA Diagnosis LED Diagnosis OH Diagnosis
Possible overheating of oil or cellulose.
OHO Diagnosis
Overheating of oil involved.
CD Diagnosis
Degradation of cellulose involved.
Where Location Fault Location Confidence LTC
Core/Tank
Bushings/Leads
Windings
Other
0.001
1.000
0.000
0.000
0.000
Fault Location
Previous Fault Location
1-core/tank
1-core/tank
Conclusions
Met our goal to build an “exception based” system Although accuracy is good (93%) APS is researching and training improved ANNs ANNs are capable of detecting the underlying, complex patterns of DGA and are a good partner with on -line monitoring