Académique Documents
Professionnel Documents
Culture Documents
Michel Duval
IREQ, Varennes, QC, Canada
issolved gas analysis (DGA) is widely used to detect incipient faults in oil-filled electrical equipment. Several
methods are available for the interpretation of laboratory results,
such as those recommended in IEC Standard 60599 [1] and IEEE
Standard C57.104-1991 [2]. Computational and graphical methods employing gas ratios and proportions have been worked out
for recognizing the characteristic patterns of dissolved gases associated with the main fault types. These diagnosis methods have
been developed and validated using large data sets for equipment
in service, where faults were identified and documented by maintenance experts inspecting the equipment.
As indicated in IEC 60599 and 60567, there is always some
degree of inaccuracy in laboratory dissolved-gas measurements,
especially at low gas concentrations. This inaccuracy affects gas
ratios and other diagnostic calculations. Therefore, the results
based on them may be correspondingly uncertain in some cases.
An early proposal for dealing with the random variability of DGA
data and the consequent ambiguity of diagnosis results, presented
by J. J. Dukarm [3] and implemented in initial versions of a DGA
software product [4], was based on the representation of fault
detection and diagnostic criteria by fuzzy sets [5].
In the present article, the measurement inaccuracy problem is
examined in more detail using recent CIGRE data, and it is shown
how proper allowance for this inaccuracy can improve transformer
condition assessment and diagnosis by DGA.
James Dukarm
Delta-X Research, Victoria, BC, Canada
21
At low gas
concentrations
Best lab
3%
22%
Average
15%
30%
Worst lab
65%
64%
22
At low gas
concentrations
Best lab
1%
7%
Average
7%
27%
Worst lab
15%
65%
23
sis of the DGA history of a transformer, taking all of the combustible gases into account, requires computer assistance with appropriate software such as [4].
A simple rule suggests itself for deciding that a gas concentration or rate has exceeded a particular threshold value. The
value is considered to have been exceeded if the reported concentration or rate exceeds it by more than its accuracy margin.
The effect of this decision rule is to avoid making a positive determination (in favor of a fault of more severity) unless it is very
unlikely that the reported concentration or rate exceeds the threshold value only because of random measurement error.
An active, energetic fault generates so much gas so rapidly
that the accuracy margin is not really an important factor in the
analysis. The existence of a fault and its probable nature are
abundantly clear when there are high gassing rates in the presence of newly elevated high levels of dissolved gas. In cases
where gas concentrations and gassing rates are low, taking into
account the accuracy of laboratory results becomes all the more
important for early detection of incipient faults.
mal faults at temperatures between 300C and 700C (T2), thermal faults at temperatures above 700C (T3), and combinations
of thermal and electrical faults (DT).
The scope of the gas ratios methods (IEC, Dornenburg, Rogers,
and the triangle) is to identify these general types of faults; therefore, they do not use CO and CO2, which are considered separately to determine paper involvement or not in the fault. H2 is
not used in the triangle method because it is a by-product of
practically all faults, and the three gases used (CH4, C2H4, and
C2H2) are sufficient to obtain a reliable diagnosis.
The coordinates of the triangle are calculated as follows: CH4
(%) = m/(m + e + a), C2H4 (%) = e/(m + e + a), and C2H2 (%) =
a/(m + e + a), where m = [CH4], e = [C2H4], and a = [C2H2] in
ppm.
B. Diagnostic Reliability
The accuracy of DGA gas ratios and reliability of diagnoses
are affected by the accuracy of DGA measurement results. This
can be appreciated by considering the results of the three CIGRE
RRTs, two performed at medium concentration levels (Figure 2)
and one performed at low concentration levels (Figure 3). The
three different gas-in-oil standards used, each corresponding to
a (CH4, C2H4, C2H2) point in the triangle, are represented by
crosses in the figures; the results reported by the participating
laboratories are represented by red dots. The laboratory results
farthest from the central prepared value are the least accurate.
Each standard value is shown with a polygonal neighborhood
illustrating the extent of measurement variability to be expected,
corresponding to the average CIGRE laboratory accuracies given
in (1). The boundaries of the polygon are calculated by applying
the CIGRE average accuracy to the CH4, C2H4, and C2H2 concentrations of the prepared RRT standard and using a standard
statistical method for determining the uncertainty of a result cal-
No diagnosis
Wrong diagnosis
Total
(%)
Key gases
58
58
Rogers
33
38
Dornenburg
26
29
IEC
15
23
Triangle
24
outcomes indicated, indicating the approximate relative likelihood of each outcome. In the low-range T2 case of Figure 4 and
Table 4, the CF for T2 is 61.5%, and for DT, it is 38.5%. In the
medium-range T2 case, the CF for T2 is 100%. Table 4 shows the
CF for all cases illustrated in Figure 4.
The volume calculation for deriving the CF is only practical
using software such as [4]. If areas are used instead of volumes
for determining CF values (e.g., in estimating CF by visual inspection of the overlap areas), in effect treating the error probability distribution as uniform, the CFs of the secondary diagnoses tend to be somewhat higher. The polygon boundary calculations can be simplified considerably using interval arithmetic,
generally producing somewhat larger uncertainty polygons.
Table 5 and Figure 5 illustrate data from three DGA samples
taken from a power transformer with a constant base load. The
first sample shows extremely low gas concentrations, which in
fact had been stable for a long time. The second sample shows
slight increases in CH4 and C2H4, which are still below the typical rates of increase reported by CIGRE [10] for power transformers in service. The third sample shows much higher gas
increases, which are above typical values and clearly are faultrelated.
The data from Table 5 are plotted as three points in Figure 5.
The first, representing the earlier background gas levels, is marked
as a blue cross. Because those levels are so low, being at or near
the laboratory detection limits, the uncertainty region (in blue) is
very large, overlapping five fault zones in this case. The second
sample is shown as a dot in the DT zone, with an uncertainty
region (in red) overlapping the DT and T2 zones. The diagnosis
for this sample is DT (85.1%) and T2 (14.9%). The third sample,
with the much larger gas levels, is a dot in the T3 zone, with a
narrow uncertainty region (in red) lying entirely within the T3
zone near the triangle boundary. The diagnosis for this final
25
Table 4. Examples of confidence factors (CFs) associated with the diagnosis of faults, calculated at various gas concentration levels, using
the default accuracy values of CIGRE laboratories.
Fault 1
CH4
C2H4
C2H2
PD
D1
(ppm)
PD
99
9.9
D1
38
3.8
D2
15
1.5
T2
69
6.9
T3
20
2
0
0
1.2
50
5
30
3
75
7.5
T1
T2
T3
DT
0.1
12
D2
97.5
2.5
5.5
64.3
50
100
100
35
100
3.5
97.5
1
0.1
2.5
100
5
0.5
30.2
61.5
38.5
100
100
1 PD
= corona partial discharges, D1 = electrical discharges of low energy, D2 = electrical discharges of high energy, T1 = thermal faults at
temperatures <300C, T2 = thermal faults at temperatures between 300 and 700C, T3 = thermal faults at temperatures >700C, and DT =
combinations of thermal and electrical faults.
2X
= Cases where the calculation of gas ratios and diagnoses is prohibited by IEC 60567 because all concentrations are below typical values.
sample is T3 (100%). The reason for the gas formation was found
to be deteriorating no-load tap changer contacts.
IEC 60599 points out that background gases can bias or distort diagnostic calculations and should be subtracted before those
calculations are done, provided that gas levels are high enough.
26
If measurement accuracy is known or can be reasonably estimated and if the accuracy is accounted for in all calculations, it is
feasible to do the diagnostic calculations (e.g., for the Duval triangle) at much lower gas levels and obtain useful results, with
any ambiguity being quantified by CF as explained previously.
Although subtraction of background gas levels, when working
with fairly low gas concentrations, can greatly increase the relative uncertainty of the results, it is nevertheless possible, in many
cases, to arrive at a useful tentative result if measurement uncertainty is included in the calculations. For example, after subtracting the background gases of the first sample from the second one (effect of subtraction not shown in Table 5 and Figure
5), the point falls within the T2 zone, with a large uncertainty
region and a diagnosis of DT (51.0%), T2 (41.1%), and T3 (7.2%).
By comparing with the diagnosis for the second sample obtained
without the subtraction, the CF has been markedly reduced for a
fault DT and increased for a fault T2/T3.
The higher the CF, the more reliable is the diagnosis. It is to
be noted that IEC 60599 prohibits the calculation of gas ratios
and diagnoses for cases indicated as X in Table 4, according to
the rule that gas ratios should not be calculated unless at least
one of the gases considered is above typical levels. In Table 4, in
the case of PD at low gas levels, CF values indeed confirm that it
is not possible to get a reliable diagnosis. In the other cases of
faults at low gas levels, however, this is possible with the help of
CFs, a significant improvement over the IEC rule.
The same kind of treatment is applicable to the Rogers/IEC
ratios, and results obtained are similar to those illustrated in Table
4 and Figure 4 for the triangle.
CH4
C2H4
C2H2
PD
D1
D2
65.0
7.2
(ppm)
T1
T2
T3
DT
(% for a fault)
Mar 2000
Aug 2000
Feb 2001
95
158
0.1
0.3
27.3
14.9
85.1
100
1 PD
= corona partial discharges, D1 = electrical discharges of low energy, D2 = electrical discharges of high energy, T1 = thermal faults at
temperatures <300C, T2 = thermal faults at temperatures between 300 and 700C, T3 = thermal faults at temperatures >700C, and DT =
combinations of thermal and electrical faults.
Conclusions
Some inaccuracy is always associated with laboratory DGA
measurements of transformer oil, which may affect the gas ratios, concentration differences, and other calculations upon which
transformer condition assessment and fault diagnosis by DGA
are based. Interval calculations or simple statistical calculations
provide a practical indication of the extent and nature of this effect and improve the accuracy and reliability of diagnoses by
associating a CF with each diagnosis provided. Thus, reliable
diagnoses may be given even at low gas concentrations for the
detection of incipient faults.
Although these calculations in principle can be performed
manually, the number and complexity of them in an analysis incorporating the results from several samples requires computer
assistance, with appropriate software such as [4]. The DGA accuracy values provided by each laboratory should preferably be
used; if they are unavailable, the average CIGRE accuracy values can be used as defaults.
References
[1] Mineral Oil-Impregnated Electrical Equipment in ServiceGuide
to the Interpretation of Dissolved and Free Gases Analysis, IEC
Publication 60599 (1999-03), Mar. 1999.
[2] Guide for the Interpretation of Gases Generated in Oil-Immersed
Transformers, IEEE Std C57. 104-1991, 1991.
[3] J. J. Dukarm, Transformer oil diagnosis using fuzzy logic and
neural networks, in Proc. IEEE Can. Conf. Elect. Comp. Eng.,
1993, pp. 329332.
[4] Transformer Oil Analyst Software. Delta-X Research, Victoria BC
Canada, 19942004.
[5] G. J. Klir and B. Yuan, Fuzzy Sets and Fuzzy Logic: Theory and
Applications. Upper Saddle, NJ: Prentice Hall PTR, 1995.
[6] Guide for the Sampling of Gases and of Oil from Oil-Filled Electrical Equipment and for the Analysis of Free and Dissolved Gases,
IEC Publication 60567 (1992-07), 1992. (A revised version will be
available in 2005).
[7] M. Duval, Using gas-in-oil standards to improve accuracy of dissolved gas analysis results and diagnoses, Electricity Today, vol.
14, no. 6, p.16, 2002.
[8] M. Duval, New techniques for dissolved gas in oil analysis, IEEE
Elect. Insul. Mag., vol.19, no.2, pp. 615, 2003.
[9] M. Duval, et al., Dissolved gas analysis: It can save your transformer, IEEE Elect. Insul. Mag., vol. 5, no. 6, pp. 2227, 1989.
Michel Duval graduated from the University of Toulouse in 1966 with a B.Sc. in
Chemical Engineering, and received a Ph.D.
in Polymer Chemistry from the University of
Paris in 1970. Since then, he has worked for
Hydro Quebecs Institute of Research (IREQ)
on electrical insulating oils, dissolved gas
analysis, and lithium polymer batteries. A senior member of IEEE, he holds 12 patents;
has authored over 60 scientific papers, book chapters, or international standards; and is very active in several CIGRE and IEC
working groups. He may be reached at IREQ, 1800 boul.Lionel
Boulet, Varennes, J3X 1S1, Canada or at duvalm@ireq.ca.
James J. Dukarm has been professionally involved in scientific computing, robotics, and artificial intelligence software applications for many years, starting in 1965 as a
FORTRAN programmer doing statistical
analysis of Peace Corps databases. Since
1988, he has developed widely-used software
products for electric power industry applications, including factory electrical system
monitoring, insulation power factor testing, and dissolved gas
analysis. A founder and partner in Delta-X Research, he is a
member of IEEE, the IEEE Power Engineering Society, and
CIGRE. He has an M.S. in Mathematics from St. Marys University (San Antonio TX) and a Ph.D. in Mathematical Logic
from Simon Fraser University (Burnaby BC). He can be reached
via e-mail at jim@deltaxresearch.com.
27