power transformers

C I R E DC I R E DC I R E DC I R E D 20th International Conference on Electricity Distribution Prague, 8-11 June 2009

Paper 0944-

CIRED2007 Session X Paper No 0944 / 5

A FUZZY-LOGIC APPROACH TO PREVENTIVE MAINTENANCE

OF CRITICAL POWER TRANSFORMERS

Teo BRESCIA Sergio BRUNO Massimo LA SCALA

Polimeri Europa SpA – Italy Politecnico di Bari - Italy Politecnico di Bari - Italy

[email protected] [email protected] [email protected]

Silvia LAMONACA Giuseppe ROTONDO Ugo STECCHI

Politecnico di Bari – Italy Politecnico di Bari – Italy Politecnico di Bari – Italy

[email protected] [email protected] [email protected]

ABSTRACT

In this paper, a predictive method for the diagnosis of

power transformer failures is presented. The approach is

based on the interpretation of Dissolved Gas Analysis

(DGA) data using Fuzzy Logic.

INTRODUCTION

This paper focuses on methods that are based on the consideration that insulating organic compounds (cellulose paper and oil) produce gas when subjected to thermal and electric stress. Most gases originate in insulating oil and their composition is influenced by temperature. In particular, in this paper, a predictive method for the diagnosis of power transformer failures is presented. The approach is based on the interpretation of DGA data using Fuzzy Logic (FL). The proposed diagnostic method adopts indicators related to the ratios C2H4/C2H6, C2H2/C2H4 CH4/H2 and to the concentration of specific gases such as hydrogen, carbon monoxide, methane, ethane, ethylene, acetylene [1][6].

DISSOLVED GAS ANALYSIS

Dissolved Gas Analysis (DGA) is one of the most commonly adopted diagnostic method for detecting and evaluating faults in electric devices. Related technical guides, derived by practical experiences and examination of a large machine number, can be found in [7]. These guides concern electric devices filled with insulating oil and solid insulation (paper or cellulose paperboard) and describe how dissolved or free gas concentration can be interpreted in order to diagnose machines conditions.

Faults

Through the data collected during internal inspections of a large number of machines, the IEC Technical Committee 10 (TC-10) has classified the faults that can be detected by gas analysis [11]:

- partial discharges (PD) that connect conductors only partially through insulation system (low temperature plasma discharges);

- low-energy discharge (D1) in oil and/or paper due to the flow of electricity through the disrupted insulation;

- high-energy discharge (D2) in oil and/or paper with

high current level; this fault is usually accompanied by large disruptions, burns and device shutdown;

- thermal faults due to overheating of the insulation.

Typical values of gas concentrations

Gassing is a ordinary phenomenon due to aging of any transformer. If gas concentration is under a certain limit, it is considered as a “typical gas pattern”. For each gas, the typical value represents the limit value under which it is reasonably possible to exclude the presence of anomalies. These values are based on operating practices and are usually evaluated considering that only the 10% of transformer population exceeds such without exhibiting anomalies. These typical values for the 90% of the transformer population are defined as 90 percentile typical values. These values are usually adopted for deciding if a more frequent monitoring of gas is necessary in order to prevent a possible fault. Table I represents the 90 percentile typical values for industrial transformer and for generic power transformers.

GAS Industrial

transformer [ppm] Power transformer

[ppm] H2 200 60-150 CO 800 400-850 CO2 6000 5300-12000 CH4 200 35-130 C2H6 150 50-70 C2H4 200 110-250 C2H2 < 1 80-270

Table I: Typical gas concentrations for industrial or

power transformers

Indicators

On the basis of the above mentioned processes and

experimental results collected by several researches, it is

possible to give an interpretation of gas concentration

measurements. In particular, the IEC Technical Committee

10 (TC 10) has based its statistics on data collected all over

the world (about 10,000 transformers, operating in 15

different power grids). By following an approximate

approach, as shown in Table II, it is possible to associate the

presence of certain gases with the onset of a specific fault.

European and American norms, after decades of

observations, have identified certain ratios that permit to

give a more reliable interpretation of fault nature [13-14].


Paper 0944-


Faults Discharges

in gas vacuole

High energy

Discharges Overheating

Acetylene Hydrogen Ethylene

others

Ethylene Ethane

Methane Hydrogen

Gas Hydrogen Methane Ethane

Carbon oxides (CO e CO2) for cellulose decomposition

Table II: Main faults categories according to more

important gases

These ratios are :

;3;2;142

22

42

22

2

4

HC

HCR

HC

HCR

H

CHR ===

.5;462

42

22

62

HC

HCR

HC

HCR ==

According to the IEC approach, Italian Comitato

Elettrotecnico Italiano C.E.I. (Italian Electrical Committee)

Standards adopts the Rogers ratios (RR) in order to identify

possible faults (R1, R2 and R5). These ratios must be

calculated whenever, for any of monitored gases, there is a

violation of concentration threshold or a steep rising rate.

DIAGNOSTIC MODEL BY MEANS OF FUZZY

LOGIC

The proposed diagnostic method can take advantage of

historical data series obtained by several studies on

transformers. This method uses as indicators hydrogen,

carbon monoxide, carbon dioxide, methane, ethane,

ethylene, acetylene, and the ratios C2H4/C2H6, C2H2/C2H4 e

CH4/H2 [15-17].

Type Description

1 No faults 2 Low-energy partial discharge 3 High-energy partial discharge 4 Low-temperature thermal fault T < 150°C 5 Low-temperature thermal fault 150°C<T<300°C 6 Medium-temperature thermal fault 300°C<T<700°C 7 High-temperature thermal fault T>700°C 8 Low-energy electric discharge 9 High-energy electric discharge

10 Paper involved in fault

Table III: Type of faults

Ten possible events (faults) have been listed in Table III

and collected in the fuzzy set U1={1, 2, 3, 4, 5, 6, 7, 8, 9,

10}.

Table IV shows the indicators that have been selected for

this study. This choice was made considering that such fault

detectors are the ones that show the higher correlation with

the chosen fault events.

The set of states that can be assigned to the three Rogers

ratios of Table IV are collected in the fuzzy set U2={0, 1,

2}, where U2 derives from the application of IEC Standards

The membership functions that associate the Rogers ratio

R1 to the state 0, 1 or 2 is shown in Fig. 1. Other ratios

exhibit a similar behaviour.

Table IV Code Fault Detectors

a H2

b CH4

c C2H4

d C2H6

e C2H2

f CO

g CH4/H2

h C2H4/C2H6

i C2H2/C2H4

Table IV: Type of fault detectors

Fig. 1: Membership function of R1

In order to describe the concentration associated with each

of the six gases assumed as fault detectors, the fuzzy set U3

= {L, M, H} has been defined. The state L, M or H is

referred to low, medium, or high concentration of each gas.

The membership functions that links states L, M, and H to

gas concentration level were developed according to

International and European standards. As an example the

membership functions of hydrogen concentration is shown

in Fig 2. Similar behaviour are exhibited by the other 5

gases.

Figure 3 shows graphically the further step that was made in

the development of the proposed model: for each state (0, 1

or 2) of each Rogers ratio, a characteristic function that

associates such state with the selected faults is defined.

In Fig. 3 (C2H2/C2H4), the Roger ratio R2 is represented.

The state 0 (R0) is characterized by a high membership

degree with respect to faults 1, 2, 4, 5, 6 and 7. The

membership degree is low with respect to fault conditions 3,

8, 9 and 10.

The state 1 (R1) is characterized by high correlation only

with faults 3, 8, and 9, whereas the state 2 (R2) has a strong

correlation only with faults 8 and 9. The same

considerations apply to Rogers ratios R1 and R5 which are

omitted for brevity.


Paper 0944-


The same step is repeated for gas concentration parameters.

For each state L, M or H of each gas a characteristic

function that correlates such state with the selected faults is

defined. Thus, it is possible to take into account key-facts

like corona faults, overheating and discharge arcs

characterized by a high concentration of hydrogen, ethylene

and acetylene or the presence of carbon monoxide which is

a signature of paper involvement in the fault.

Fig. 2: Hydrogen membership function

Fig. 3: Correlation between R2 and fault conditions

In this way, each gas has a function that describes its

correlation with the fault. Usually, ethylene, methane and

ethane play a principal role in detecting thermal fault,

whereas acetylene gives the last word in detecting high

energy discharges. An example of correlation between gas

concentration and fault is shown in Fig. 4. Of course similar

correlations have been defined for the other five gas-based

fault detectors.

Fig. 4: Correlation between H2 and faults

The next step consists in defining a fuzzy vector FC whose

elements are the values assumed by the membership

functions in correspondence with input data (the three ratios

and the six gas concentration levels).

In the formulation of FC(k), wi denote the k-th hypothesized

fault weight given to the FC vector element.

Once Fc has been evaluated on the basis of actual testing

data, this vector needs to be compared with a similar

reference vector Fr(k) which represents the typical values

experienced by the k-th fault selected among the ten major

faults listed in Table III.

=

.µ w,µ w,µw,µ w,µ w,µw,µ w,µ w,µw

,µw,µ w,µw,µ w,µ w,µw,µ w,µ w,µw

,µ w,µ w,µw,µ w,µ w,µw,µ w,µ w,µw

F

H

f9

M

f9

L

f9

H

e8

M

e8

L

e8

H

d7

M

d7

L

d7

H

c6

M

c6

L

c6

H

b5

M

b5

L

b5

H

a4

M

a4

L

a4

2

i3

1

i3

0

i3

2

h2

1

h2

0

h2

2

g1

1

g1

0

g1

C

The characteristic functions Fr(k) have been obtained as a

result of a large consultation of numerous experienced

operators and users and by the reference analysis over 800

test cases. The comparison between Fr(k) and Fc(k) is

performed in terms of dot product and the result is

normalized with regard to the norm of the reference vector.

The procedure can be summarized as follows:

1. Calculate on the base of the test data the characteristic

Rogers ratios CH4/H2, C2H4/C2H6 and C2H2/C2H4 and

input gas concentrations;

2. Define the fault characteristic vector: Fc=[µ1, µ2,……]

adopting the membership functions;

3. Evaluate through the scalar product Fp(k) =

Fr(k)*Fc(k), k = 1, 2,…..,10;

4. Evaluate the proximity of each incipient fault:

F(k)= ( )

( )kT

kF

r

p , k= 1,2,..,10, where )()( kFkT rr = ;

5. Detect the prevalent and/or the most likely fault type

by:

Fd = max{F(1), F(2), ………F(10)}.

TEST CASE

The proposed diagnostic methodology was implemented for

the maintenance scheduling of four power transformers


Paper 0944-


installed in a petrochemical plant.

The transformers under study are four double secondary

winding transformers for converters supplying, rated in the

range 6-18 MVA, with insulating oil (ONAN), and 13.2/6.2

kV voltage rates.

Due to operating conditions, the four power transformers

under investigation are characterized by early ageing.

Failure or breakdown of such transformers may have

catastrophic consequences (i.e. explosion of the

transformer).

In order to apply the proposed diagnostic model,

membership functions were suitable “tuned” taking into

account typical values of gas concentration for this type of

transformer.

The entire fuzzy model was developed with the Matlab®

Fuzzy Logic Toolbox that takes advantage of a simple but

powerful graphic interface. This method was applied to

historical data series, analyzing transformer conditions since

03/03/1997 up to 12/03/2003. Sample data for gas

concentration and Rogers ratios are referred to a single

transformer and a testing campaign performed in 04-19-

2002. The implementation of such data in the described

fuzzy model permitted to evaluate the elements of the FC

vector, shown in Table V.

Table V # Ratios µ0H µ1H µ2H

1 C2H2/C2H4=0,04 0,998 0 0

2 CH4/H2=7,47 0 0 1

3 C2H4/C2H6=0,04 1 0 0

# Gas (ppm) µLH µMH µAH

4 H2=34 1 0 0

5 CH4=254 0 0 1

6 C2H4=28 1 0 0

7 C2H6=726 0 0 1

8 C2H2=<1 1 0 0

9 CO=623 0 0.99 0

Table V. Fc vector

Applying the proposed procedure we obtain:

Fd = max{0.1, 0.225, 0.237, 0.4, 0.5, 0.4, 0.4, 0.09, 0.06,

0.3} = 0.5;

indicating the presence of a thermal faults with temperatures

ranging in the interval 150-300°C. The value of F(10) can

also suggest that cellulose can be involved in the fault.

An inspection of the internal parts of the transformer

(windings, oil, paper insulation) confirmed the diagnosis; in

particular, the internal parts of the transformer were

blackened and the insulation paper looked liked a “rusk”.

meaning that the transformer was operated at high

temperatures for a longer time. Thus, the method proved to

be effective in estimating the severity of incipient faults.

CONCLUSIONS

One last remark about proposed diagnostic methodology,

based on results interpretation by means of fuzzy logic. This

diagnostic approach does not replace the important role of

experienced technical operators, but it could be an efficient

tool to improve the capacity to correlate many different

data. The diagnostic tool, suitable for this particular

application, allows to implement procedures for preventive

maintenance, taking into account specific needs of

industries characterized by a critical process.

REFERENCES

[1] J. J. Kelly, “Transformer Fault Diagnosis by Dissolved

Gas Analysis”, IEEE Transactions on Industry

Applications, Vo1.U-16, No.6 pp.777-782, 1980.

[2] M. Duval, “Dissolved Gas Analysis: It Can Save Your

Transformer”, IEEE Electrical Insulation Magazine,

Vo1.5, No.6, pp. 22-27, 1989.

[3] W. D. Halstead, “A thermodynamic assessment of the

formation of gaseous hydrocarbons in faulty

transformers”, Journal of the institute of petroleum,

Vol. 59, pp.239-41, 1973.

[4] S. R. Lindgren, “Transformer condition assessment

experiences using automated on-line dissolved gas

analysis”, Cigré general session 2004, 29 agosto – 3

settembre, 2004, Parigi, Francia.

[5] M. Shirai, S. Shimoji, T. Ishii, “Thermodynamic study

on the thermal decomposition of insulating oil”, IEEE

Transactions On Electrical Insulation, Vol.12, pp.

272-280, August 1977.

[6] M. Shirai; T. Ishii and Y. Makino, “Evolution of

hydrogen from insulating oil in transformers”, IEEE

Transactions On Electrical Insulation, Vol. 12 E.I.,

pp. 266-272, August 1977.

[7] Norma CEI 10, Guida per il controllo e il trattamento

degli oli minerali isolanti in servizio nei trasformatori

e in altre apparecchiature elettriche, 1997-09.

[8] Norma CEI 10-24, Classificazione generale dei liquidi

isolanti, 1997-06.

[9] Norma CEI 10-1, Oli minerali isolanti per

trasformatori e per apparecchiature elettriche, 1997-

09.

[10] Guida IEC 60567, Oil-filled electrical equipment -

Sampling of gases and of oil for analysis of free and

dissolved gases.

[11] Norma CEI EN 60599 (10-10), Guida

all’interpretazione dell’analisi dei gas disciolti e

liberi”, Apparecchiature in servizio impregante in

olio minerale, 2000.

[12] IEEE std C57.104-1991, Guide for the interpretation

of gases generated in oil-immersed transformers.

[13] E Dornenburg, W Strittmater, “Monitoring oil cooling

transformers by gas analysis”, Brown Boveri Review,

Vo1.61, pp.238-247, 1974.


Paper 0944-


[14] R. R. Rogers, “IEEE and IEC Codes to Interpret

Incipient Faults in Transformers Using Gas in Oil

Analysis”, IEEE Transactions On Electrical

Insulation, Vol EI- 13, No.5, October 1978.

[15] K. Tomsovic, M. Tapper, T. Ingvvarsson, “A fuzzy

information approach to integrating different

transformer diagnostic methods”, IEEE Transaction

on Power Delivery, vol. 8, No. 3, pp.1638-1646, July

1993.

[16] James J. Dukarm, “Transformer Oil Diagnosis Using

Fuzzy Logic and Neural Networks”, 1993 Canadian

Conference on Electrical and Computer Engineering,

Vol.1, pp.329-332.

[17] Yann-Chang Huang, Hong-Tzer Yang, Ching-Lien

Huang, “Developing a New Transformer Fault

Diagnosis System through Evolutionary Fuzzy Logic”,

IEEE Trans. on Power Delivery, Vol.12, No.2,

pp.761-767, April 1997.

power transformers

Documents