• International Journal of Technology (IJTech)
  • Vol 11, No 2 (2020)

Data-driven Fault Diagnosis of Power Transformers using Dissolved Gas Analysis (DGA)

Data-driven Fault Diagnosis of Power Transformers using Dissolved Gas Analysis (DGA)

Title: Data-driven Fault Diagnosis of Power Transformers using Dissolved Gas Analysis (DGA)
Arian Dhini, Akhmad Faqih, Benyamin Kusumoputro, Isti Surjandari, Andrew Kusiak

Corresponding email:

Cite this article as:

Dhini, A., Faqih, A., Kusumoputro, B., Surjandari, I., Kusiak, A., 2020. Data-driven Fault Diagnosis of Power Transformers using Dissolved Gas Analysis (DGA). International Journal of Technology. Volume 11(2), pp. 388399

Arian Dhini Laboratory Industrial Engineering Department, Faculty of Engineering, Universitas Indonesia, Kampus UI Depok, Depok 16424, Indonesia
Akhmad Faqih Department of Electrical Engineering, Faculty of Engineering, Universitas Indonesia, Kampus UI Depok, Depok 16424, Indonesia
Benyamin Kusumoputro Department of Electrical Engineering, Faculty of Engineering, Universitas Indonesia, Kampus UI Depok, Depok 16424, Indonesia
Isti Surjandari Laboratory Industrial Engineering Department, Faculty of Engineering, Universitas Indonesia, Kampus UI Depok, Depok 16424, Indonesia
Andrew Kusiak Department of Industrial and Systems Engineering, College of Engineering, University of Iowa, 4627 Seamans Center for the Engineering Arts and Sciences, Iowa City -Iowa, 52242, USA
Email to Corresponding Author

Data-driven Fault Diagnosis of Power Transformers using Dissolved Gas Analysis (DGA)

A power transformer is a critical piece of equipment in a power plant for distributing electricity, and it experiences thermal and electrical stresses during operation. Dissolved gas analysis (DGA) remains one of the most effective techniques to monitor the health of oil-filled transformers. Some traditional approaches for interpreting DGAs have been introduced. Occasionally, such approaches leave the state of the transformer uncategorized. This study proposed data-driven approaches for a fault diagnosis system based on DGA data using support vector machine (SVM). SVM is known for its robustness, good generalization capability, and unique global optimum solutions, particularly when data is limited. Backpropagation neural networks (BPNN) and extreme learning machine-radial basis function (ELM-RBF), a recent Neural Networks (NN)-based method with extremely fast computation time, were compared to SVM. An advanced technique to overcome the imbalanced data and synthetic minority oversampling technique (SMOTE) was proposed to investigate the effect on classifier performance. The model was trained and tested using IEC TC 10 databases and transformer DGA monitoring data of a thermal power plant in Jakarta. The results indicated that SVM displayed the best performance compared to ELM-RBF and BPNN. It demonstrated extremely high accuracy, while still maintaining fast computation time for all stages in the proposed multistage fault diagnosis system.

Condition monitoring; Dissolved gas analysis; Fault diagnosis; Support vector machine; Transformer


Thermal power plants remain main electricity providers; however, most of them are highly aged. According to Ulum et al. (2017), plant equipment, which degrades over time, leads to electricity production loss. Therefore, the industrial process measurements are imperative to monitor and assure the quality and safety in operations (Kusiak and Song, 2009). Those measurements are the bases to determine the maintenance activities. Such maintenance will raise the reliability and availability of equipment (Pariaman et al., 2017), including the reliability and availability of power transformers, one of the key pieces of equipment  in a  power distributing system.   Despite  the fact  that  faults  in a transformer occur infrequently, once they do occur, their impact is significant in terms of safety, downtime, and equipment loss (Hernandez and Labib, 2017). DGA has been known to be a popular, sensitive, and reliable technique to monitor the insulation condition of transformers (Saha, 2003; Chakravorti et al., 2013). However, conventional DGA interpretation methods include certain drawbacks in terms of accuracy and uncertainty (Shintemirov et al., 2009; Ghoneim and Taha, 2016). These approaches require the manual interpretation of experts, and some measurements may be unidentifiable when using any interpretation method (Lin et al., 1993; Yang et al., 2009; Abu-Siada and Islam, 2012).

Faults should be diagnosed accurately and in a timely manner, since ignoring them degrades the safety and security of the process, and it could lead to catastrophic failures and loss of material and even life (Gao et al., 2015). Therefore, machine learning methods have been recently applied in studies for transformer fault detection and diagnosis (FDD) using DGA data. Support vector machine (SVM) has become an increasingly popular technique in machine learning. SVM is known for its robustness, good generalization capability, and unique global optimum solution. Shin and Cho (2006) explained that its good generalization capability was due to structural risk minimization, which is employed by SVM, rather than empirical risk minimization, as in NN. SVM also performs well with small samples and high dimension data, while still maintaining short computational times (Lv et al., 2005; Bacha et al., 2012; Sahri and Yusof, 2015; Souza and Ramachandran, 2016). Some studies have compared SVM with other methods, such as kNN, ANN (Shintemirov et al., 2009), the expert system (Lv et al., 2005), fuzzy logic, multilayer perceptron (MLP), and radial basis function neural networks (RBFNN) (Fei et al., 2009; Bacha et al. 2012). These studies found that SVM performed better. Ghoneim and Taha (2016) conducted a fault diagnosis study in the same area, but they only applied limit rules to categorize faults.

In reality, the availability of fault data is usually significantly lower than the normal operating condition (NOC). This situation may lead to a decrease in classifier performance. Fault diagnosis focuses on how to detect a fault, the minority class. Sahri and Yusof (2015) tested some scenarios of input features without considering how to handle imbalanced data. Only a few studies have considered this imbalanced data in FDD. Chawla et al. (2002) proposed the use of the synthetic minority oversampling technique (SMOTE) to overcome the imbalanced data problem by creating new data through an interpolation process. No previous studies have applied SMOTE in the case of transformers for fault diagnosis.

This study aims to develop an accurate and fast transformer fault diagnosis system from DGA data using SVM and SMOTE for data balancing. Other recent classifier methods, extreme learning machine-radial basis function (ELM-RBF), which is known for its extremely fast computation time, is applied. The high computation of ELM-RBF, while still maintains good accuracy, will result in a number advantages, such as fast fault diagnosis and lower hardware investment costs.  In addition, the classical approach of the NN-based technique, backpropagation neural networks (BPNN), was also tested. He and Kusiak (2017) employed SVM, Multi-layer perceptron (MLP) and ELM based methods, which performed satisfactorily, to predict wind power. This study contributes to the possibility of lengthening transformer life by proposing fault diagnosis systems using aforementioned data-driven approaches. The proposed models were trained and tested using the fault and NOC DGA data from IEC TC 10 and NOC DGA data from a thermal power plant in Jakarta. Ghoneim and Taha (2016) and Sahri and Yusof (2015) used IEC TC 10 as a data source as well. Only Ghoneim and Taha (2016) and Sahri and Yusof (2015) used real data, whereas Bacha et al. (2012) used simulated data from the Tunisian company of Electricity and Gas.  

The remainder of this article is organized as follows. Section 2 describes the study methods used in this research, from the proposed research framework to the methods applied. The following section, Section 3, elaborates the results and discusses the fault diagnosis based on certain performance measures. Some concluding remarks and areas of future research are provided in Section 4.


Based on the two performance measures of accuracy and computation time, it can be concluded that the SVM outperforms other NN-based methods, ELM-RBF and BPNN, for the proposed multistage fault diagnosis of power transformers using DGA data. SVM performs satisfactorily with high accuracy and fast computation time. It has proven to be an effective method in terms of classification with a better generalization. Despite its poorer results compared to SVM, ELM-RBF also performed the best in terms of computation time. In terms of accuracy, BPNN exhibited the best accuracy. In the future, research to improve accuracy, data quantity, and data quality should be enhanced by adding more training data, selecting optimum features, and combining other types of input data, such as gas ratio. More comparisons with other data-driven methods, such as decision trees, kNN, or even deep learning, and applying optimization in parameter selections, represent additional options.

Supplementary Material
R1-IE-3625-20200207193846.jpg Figure 1
R1-IE-3625-20200207193907.jpg Figure 2
R1-IE-3625-20200207193931.jpg Figure 3 (revised)

Abu-Siada, A., Islam, S., 2012. A New Approach to Identify Power Transformer Criticality and Asset Management Decision based on Dissolved Gas-in-oil Analysis. IEEE Transactions on Dielectrics and Electrical Insulation, Volume 19(3), pp. 10071012

API, 2014. Imbalanced-Learn Documentation: User Guide. Available Online at: https://imbalanced-learn.org/stable/over_sampling.html#smote-variants. Accessed on April 18, 2020

Bacha, K., Souahlia, S., Gossa, M., 2012. Power Transformer Fault Diagnosis based on Dissolved Gas Analysis by Support Vector Machine. Electric power systems research, Volume 83(1), pp. 7379

Broomhead, D.S., Lowe, D., 1988. Radial Basis Functions, Multi-variable Functional Interpolation and Adaptive Networks. Royal Signals and Radar Establishment. Memorandum 4148

Chakravorti, S., Dey, D., Chatterjee, B., 2013. Recent Trends in the Condition Monitoring of Transformers. Power Systems. Springer-Verlag, London, United Kingdom

Chawla, N.V., Bowyer, K.W., Hall, L.O., Kegelmeyer, W.P., 2002. SMOTE: Synthetic Minority Over-sampling Technique. Journal of Artificial Intelligent Research, Volume 16, pp. 321357

Dhini, A., Surjandari, I., Faqih, A., Kusumoputro. 2018. Intelligent Fault Diagnosis for Power Transformer Based on DGA Data Using Support Vector Machine (SVM). In: The 3rd International Conference on System Reliability and Safety, pp. 294298

Dong, M., Xu, D., Li, M., Yan, Z., 2004. Fault Diagnosis Model for Power Transformer based on Statistical Learning Theory and Dissolved Gas Analysis. In: Conference Record of the 2004 IEEE International Symposium on Electrical Insulation, pp. 8588

Duval, M., 1989. Dissolved Gas Analysis: It Can Save Your Transformer. IEEE Electrical Insulation Magazine, Volume 5(6), pp. 2227

Duval, M., dePablo, A., 2001. Interpretation of Gas-in-oil Analysis using New IEC Publication 60599 and IEC TC 10 Databases. IEEE Electrical Insulation Magazine, Volume 17(2), pp. 3141

Fei, S.-w., Liu, C.-l., Miao, Y.-b., 2009. Support Vector Machine with Genetic Algorithm for Forecasting of Key-gas Ratios in Oil-immersed Transformer. Expert Systems with Applications, Volume 36(3) Part 2, pp. 63266331

Gao, Z., Ding, S.X., Cecati, C., 2015. Real-time Fault Diagnosis and Fault-tolerant Control. IEEE Transactions on Industrial Electronics, Volume 62, pp. 37523756

Ghoneim, S.S., Taha, I.B., 2016. A New Approach of DGA Interpretation Technique for Transformer Fault Diagnosis. International Journal of Electrical Power & Energy Systems, Volume 81, pp. 265274

Griffin, P.J., 1988. Criteria for the Interpretation of Data for Dissolved Gases in Oil from Transformers (A Review). Electrical Insulating Oils, ASTM International, pp. 89107

Guardado, J.L., Naredo, J.L., Moreno, P., Fuerte, C.R., 2001. A Comparative Study of Neural Network Efficiency in Power Transformers Diagnosis using Dissolved Gas Analysis. IEEE Transactions on Power Delivery, Volume 16(4), pp. 643647

Han, H., Wang, W.-Y. Mao, B.-H., 2005. Borderline-SMOTE: A New Over-sampling Method in Imbalanced Data Sets Learning. In: International Conference on Intelligent Computing, pp. 878887

He, Y., Kusiak, A., 2017. Performance Assessment of Wind Turbines: Data-Derived Quantitative Metrics. IEE Transactions on Sustainable Energy, Volume 9(1), pp. 6573

Hernandez, M.D.P.C., Labib, A., 2017. Selecting a Condition Monitoring System for Enhancing Effectiveness of Power Transformer Maintenance. Journal of Quality in Maintenance Engineering, Volume 23(4), pp. 400414

Huang, G.-B., Siew, C.-K. 2005. Extreme Learning Machine with Randomly Assigned RBF Kernels. International Journal of Information Technology, Volume 11(1), pp. 1624

Kusiak, A. Song, Z. 2009. Sensor Fault Detection in Power Plants. Journal of Energy Engineering, Volume 135(4), pp. 127137

Kusumoputro, B., Sutarya, D., Faqih, A., 2016. Performance Analysis of an Automatic Green Pellet Nuclear Fuel Quality Classification using Modified Radial Basis Function Neural Networks. International Journal of Technology. Volume 7(4), pp. 709719

Lin, C.E., Ling, J.-M., Huang, C.-L., 1993. An Expert System for Transformer Fault Diagnosis using Dissolved Gas Analysis. IEEE transactions on Power Delivery, Volume 8(1), pp. 231238

Lv, G., Cheng, H., Zhai, H., Dong, L., 2005. Fault Diagnosis of Power Transformer based on Multi-layer SVM Classifier. Electric Power Systems Research, Volume 75(1), pp. 915

Pariaman, H., Garniwa, I., Surjandari, I., Sugiarto, B., 2017. Availability Analysis of the Integrated Maintenance Technique based on Reliability, Risk, and Condition in Power Plants. International Journal of Technology, Volume 8(3), pp. 497507

Saha, T.K., 2003. Review of Modern Diagnostic Techniques for Assessing Insulation Condition in Aged Transformers. IEEE Transactions on Dielectrics and Electrical Insulation, Volume 10(5), pp. 903917

Sahri, Z., Yusof, R., 2015. Fault Diagnosis of Power Transformer using Optimally Selected DGA Features and SVM. In: IEEE 10th Asian Control Conference (ASCC)

Shin, H., Cho, S., 2006. Response Modeling with Support Vector Machines. Expert Systems with Applications, Volume 30(4), pp. 746760

Shintemirov, A., Tang, W., Wu, Q., 2009. Power Transformer Fault Classification based on Dissolved Gas Analysis by Implementing Bootstrap and Genetic Programming. IEEE Transactions on Systems, Man, and Cybernetics, Part C (Applications and Reviews), Volume 39(1), pp. 6979

Souahlia, S., Bacha, K., Chaari, A., 2012. MLP Neural Network-based Decision for Power Transformers Fault Diagnosis using an Improved Combination of Rogers and Doernenburg Ratios DGA. International Journal of Electrical Power & Energy Systems, Volume 43(1), pp. 1346–1353

Souza, F.R., Ramachandran, B., 2016. Dissolved Gas Analysis to Identify Faults and Improve Reliability in Transformers using Support Vector Machines. In: IEEE Power Systems Conference (PSC), Clemson University

Sun, S., 2017. Imbalanced Binary Classification for Detecting Transcription Factor Binding Sites in H1 Human Embryonic Stem Cells, UCLA Electronic Theses and Dissertations

Tang, Y., Zhang, Y.-Q., Chawla, N.V., Krasser, S., 2009. SVMs Modeling for Highly Imbalanced Classification. IEEE Transactions on Systems, Man, Cybernetics, Part B, Volume 39(1), pp. 281288

Ulum, B., Nurrohman, Ambarita, E., Gaos, Y.S., 2017. Energy and Exergy Analysis of Mount Salak Geothermal Power Plant Unit 1-2-3. International Journal of Technology, Volume 8(7), pp. 12171228

Yang, Z., Tang, W., Shintemirov, A., Wu, Q., 2009. Association Rule Mining-based Dissolved Gas Analysis for Fault Diagnosis of Power Transformers. IEEE Transactions on Systems, Man, and Cybernetics, Part C (Applications and Reviews), Volume 39(6), pp. 597610

Zhang, Y., Ding, X., Liu, Y., Griffin, P., 1996. An Artificial Neural Network Approach to Transformer Fault Diagnosis. IEEE Transactions on Power Delivery, Volume 11(4), pp. 18361841