TOWARDS COST-EFFECTIVE MAINTENANCE OF POWER TRANSFORMER BY ACCURATELY PREDICTING ITS INSULATION CONDITION

Refat Atef Ghunem1, Khaled Bashir Shaban2, Ayman Hassan El-Hag3, and Khaled Assaleh3
1 2 3

Electrical and Computer Engineering Department, University of Waterloo Computer Science and Engineering Department, Qatar University Electrical Engineering Department, University of University of Sharjah

INTRODUCTION
Transformer asset management has been a priority to power utilities: Improving condition monitoring and diagnosis methods Preparing solid asset maintenance plans Maintenance costs are found to be heavy burden on utilities Developing a well constructed determining process for transformer health index

MEGGER TEST
A classical time-based preventive maintenance test Easy, fast, and inexpensive test to be conducted Indicates insulation deterioration Efficient tool for developing trend analysis A diagnostic tool in unplanned outages

MEGGER TEST DRAWBACKS

Temperature effect Rule in practice: factor of two is attributed to megger test readings for every 10 degrees change Moisture Content effects Accelerating deterioration Reducing the dielectric strength Faults are not detected by the insulation resistance parameter

TRADITIONAL MAINTENANCE PRACTICES

Reducing the unplanned outage cost and period for tripped transformer insulation inspection Rule-of-thumb: if the transformer trips due to the operation of differential, restricted earth fault and Buchholz relays , insulation inspection must be conducted before reconnecting to the network Megger test clearance of abnormal situation is not reliable for reconnecting the transformer Supportive tests should be conducted

SUPPORTIVE TESTS
Other tests are conducted to enhance the diagnosis reliability of the megger test: Oil Breakdown Voltage (BDV) Oil water content
Temperature correction should be considered Due to equilibrium process in the oil-paper system

Dissolved-gas analysis (DGA): CO2/CO ratio Total Dissolved Combustible Gases (TDCG)

BUT, costly ($200/sample) and time consuming (few days)

TOWARDS COST-EFFECTIVE MAINTENANCE

Optimizing the number of time- based maintenance tests Predicting oil quality and dissolved gases parameters informs about the critical tests that should be taken Developing well constructed asset maintenance plans with optimized number of tests and periods between them

THE LITERATURE
Authors Model Predictors Output Breakdown voltage(BDV),oil water content, interfacial tension(IF) and acidity & BDV Water content IF BDV & BDV, oil water content and acidity Khaled Shaban, A. H. El-Hag, and Andrei Matveev & Khaled Assaleh & Ayman El- Hag Artificial Neural Network (ANN) & Polynomial Network Accuracy BDV:84% Water content:60% IF:95% Acidity:75% & Water content: could not be estimated IF:93% BDV:84% Comments

2009& 2008

Megger

BDV and IF are indication for oil contamination only

Mohamed Ahmed Abdel Oil acidity , oil water Wahab ANN 2004 content and transformer & & & age Mohamed A.A. Wahab, Polynomial 1999 & M.M. Hamada, and Regression models Service period A.G. Zeitoun, G. Ismail 2009 & 2008 Sheng-wei Fei, Cheng Liang liu, and Yu bin Support vector Miao machine with & genetic algorithm Sheng-Wei Fei, and Yu Sun I N da Silva, A N de Souza, R M C Hossri, and J H C Hossri ANN & J.S.Navamany, and P.S. Ghosh

More than 90% for both

BDV is not comprehensive and & No. of data is very small

DGA

DGA trend

Less than 90% & More than 90%

It is not practical

2000 & 2003

furan, volume of carbon monoxide (CO), Interfacial tension and acidity & CO ,CO2 and furfural

Aging Degree

N/A & Around 90%

It is not cost-effective

2009

Ali Naderian Jahromi, Ray Piercy, Stephen All standard transformer Transformer health Weighted average Cress, Jim R.R. Service tests index andWang Fan

N/A

The weights are not assigned 8 based on quantities correlations & Relatively hard to be practiced

ANN FOR PREDICTION

Artificial neural network (ANN) is used as a modeling technique Multi-layer feed forward ANN with Back propagation algorithm

THE EXPERIMENT
Data from transformer maintenance records of 19 power transformers in Abu Dhabi Transmission and Despatch Company (TRANSCO) are used Voltage rating of 220/33/11KV Rating is in the range of 50-140MVA Range of 7-15 years into service 54 samples are used to predict oil BDV 31 samples are used to predict oil water content 32 samples are used to predict TCG 33 samples are used to predict CO2/CO ratio All the samples are used for training and testing to predict all the available data

SAMPLE OF THE DATASET

Insulation Resistance (M ) BDV (kV) 71.4 75 74 67 75 Parameters Water Content (ppm) 26.8 30.57 16.46 22.62 17.63 TDCG (ppm) 497 737 672 665 894

HV/E 1 2 3 4 5 300 820 280 560 270

LV/E 240 880 280 580 220

TV/E 400 1100 380 620 390

CO2/CO 8.25 7.98 6.46 6.23 4.85

RESULTS
4 ANNs:
3 neurons in the input layer (HV/E, LV/E and TV/E) 2 hidden layers with three neurons at the first hidden layer and 10 neurons at the second hidden layer 1 neuron at the output layer (BDV, water content, TDCG and CO2/CO)

Prediction rate: Oil BDV - 96% Oil water content - 84% CO2/CO ratio - 91% TDCG - 88%

RESULTS
80 78 76 Actual oil BDV Predicted oil BDV

BDV value (KV)

74 72 70 68 66 64 62 0 5 10 15 20 25 30 35 40 45 50 55 60

Size of testing set (samples) 1 2 3 4 5 6 7 8 9 10 11 12 Average Prediction Accuracy

Prediction Accuracy 96.66 92.79 92.53 93.61 92.75 92.93 93.61 93.52 94.27 93.99 92.29 95.25 93.68

Sample NO.

This result agrees with the results of the models proposed in [1] and [2]. Transformer oil breakdown voltage gives an indication about the dielectric capability of transformer oil to withstand high electrical stresses. Also, BDV can be an indication about the deterioration state of the insulation system in the transformer. This is because the contamination particles resulted from the insulation deterioration in the transformer oil decrease the BDV value. This explains the high correlation presented between IR and oil BDV.

RESULTS
34 32 30 Actual oil water content Predicted oil water content

Size of testing set (samples) 1 2 3 4 5 6 7 8 9 10 11 12 Average Prediction Accuracy

Prediction Accuracy 84.73 80.61 80.52 80.13 78.88 80.16 79.36 74.39 81.33 74.74 76.57 82.91

Water content value (ppm)

28 26 24 22 20 18 16 0 5 10 15 20 25 30 35

Sample NO.

79.53

compared to BDV the prediction accuracy has been reduced between 10-15%. This can be attributed to the relatively wider scatter of the data of water content (17-32 ppm) compared to BDV (63.6-78.8 kV) and the relatively small size of the training set (31 vs. 54). Moreover, the oil temperature effect in altering the value of the measured oil water content was not completely considered as the oil sample was taken between 60-65 C. Such narrow band of temperature variation may still influence the accuracy of the water content measurement. The proposed model to predict oil water content in this study can be considered superior to other proposed models [1-2] where the effect of temperature variation on measured samples was normalized.

RESULTS
11 Actual CO2/CO ratio 10 9 8 7 6 5 4 3 0 5 10 15 20 25 30 35 Predicted CO2/CO ratio

Sample NO.

Size of testing set (samples) 1 2 3 4 5 6 7 8 9 10 11 12 Average Prediction Accuracy

Prediction Accuracy 91.02 82.31 75.60 73.84 77.27 83.76 75.84 67.29 80.46 82.03 79.79 74.68

CO2/CO ratio

78.66

Although CO2/CO ratio is more effective in diagnosing cellulose involvement in incipient faults rather than indicating the deterioration of cellulose, the proposed model shows an acceptable reliability. This agrees with the practice in utilities to use CO2/CO ratio with other insulation quality parameters as indicators for cellulose thermal degradation.

RESULTS
1400 Actual TDCG Predicted TDCG 1200

1000

800

600

400

200 0 5 10 15 20 25 30 35

Sample NO.

Size of testing set (samples) 1 2 3 4 5 6 7 8 9 10 11 12 Average Prediction Accuracy

Prediction Accuracy 88.37 73.69 75.29 74.25 68.59 69.22 75.89 75.70 74.17 70.59 73.82 70.11 74.14

TDCG value (ppm)

Insulation resistance is affected by the deterioration and the abnormal excess of degradation, thus, IR can be an indicative parameter to the TDCG. Although TDCG as a parameter is a reflection of incipient faults rather than a parameter for aging index, an acceptable and reliable correlation is verified in the proposed model.

CONCLUSIONS
Prediction techniques can improve transformer insulation diagnosis and reduce the predictive and corrective maintenance costs. The proposed models suggests a decent potential for predicting oil BDV, water content, TDCG and CO2/CO ratio. It can be more efficient to test the correlations of the transformer insulation parameters in different conditions (wide temperature ranges, service ages and voltage levels, etc.) Increasing data size can enhance the generalization capability of the model