Transformer fault types and severity class prediction based on neural pattern-recognition techniques
Introduction
The transformer is considered a crucial equipment in power systems. A malfunction and damage of the transformer can be avoided by early transformer fault detection [1]. Dissolved gas analysis (DGA) is a common technique to diagnose the transformer conditions; recently, classical DGA techniques with the aid of artificial intelligent techniques enhance the diagnostic accuracy of transformer faults [2], [3], [4], [5], [6]. Although DGA techniques have a great ability to detect transformer fault types, they cannot evaluate transformer fault severity, which is considered an intrinsic property of transformer faults [7], [8], [9]. Electrical, thermal, and mechanical stresses decompose the insulating oil into hydrocarbons, especially hydrogen (H2), methane (CH4), ethan (C2H6), ethylene (C2H4), acetylene (C2H2), and carbon monoxide (CO). The energy required to form hydrocarbon gases increases in the order CH4<C2H6<=CO<=C2H4<=H2<<C2H2 [8].
In IEEE Standard C57.104 [10], the condition of the transformer is determined based on the total dissolved combustible gases (TDCG), which is the sum of all combustible gases and its change rate. This method does not consider the concentration variation of each individual dissolved gas. A great issue is encountered because the change and change rate of hydrogen concentration differ from those of acetylene concentration, and TDCG cannot be used as a parameter to identify the severity of the fault type [8, 11].
In [12], an artificial neural network (ANN) was used to develop a model to diagnose the transformer fault and study the effect of the energy weight of the dissolved gases on the identification of fault type severity. The model was constructed based on the Duval triangle method [13] to identify the transformer fault type, and n-octane was used as the starting decomposing material to calculate the weighted factor for each dissolved gas.
A fault severity model based on the rules of IEC 60,599 ratio and the thermodynamic approach using eicosane was reported in [11]. A fuzzy logic system was built to develop not only the transformer fault types but also the severity of the faults and the condition of the transformer.
TDCG was used to determine the transformer condition based on the concentration of the dissolved gases in ppm as in [10]. Jakob and Dukarm [8] demonstrated that the DGA technique was used to classify the type and severity of faults in the transformer based on TDCG. The thermodynamic analysis in [8] based on n-octane decomposing material indicated that the energy required to release the dissolved gases increases by a specified manner. The energy weight is related to the enthalpy of reaction of each dissolved gas, and the severity of the faults can then be identified.
A new phenomenology-based diagnostic technique using the thermodynamic decomposition of mineral-insulating oil was reported in [14]. The results in [14] showed a satisfactory ability of temperature, T, and equilibrium amount of carbon compounds, C(s), (T-C) binary to classify the transformer fault type. The diagnostic accuracy of the proposed model is higher than the traditional DGA techniques such as Dornenburg, Rogers’ four-ratio method, and IEC 60,599 ratio, which indicates that the decomposition method and/or concentration ratio method is not the most adequate option. However, its diagnostic accuracy is slightly lower than the Duval triangle accuracy.
Patterns were recognized in several complex problems or relationships using artificial intelligence (AI), where recognition is previously based on expert knowledge [15]. The process of determining patterns is very important in data processing and decision making. A pattern is the discrete sequence of data that can be numeric data, text, or images [16]. When the pattern recognition (PR) system is constructed and data are entered with different features, the accuracy of the system means its ability to distinguish between these data and determine the variance between them.
In literature, two starting decomposing materials were used for the thermodynamic approach to evaluate the severity of transformer faults. These starting decomposing materials are n-octane and eicosane [8, 11, 12]. Several researchers focused on determining fault severity based on a thermodynamic approach that relied on n-octane, and the others used eicosane. However, the severity of transformer faults using these starting decomposing materials is not the same due to the difference of the enthalpy of formation of the chemical reactions that indicate the decomposition of n-octane and eicosane.
The objective of the proposed technique is to develop the diagnostic model based on neural pattern recognition (NPR) technique and not only diagnose the transformer fault types but also identify the fault severity classes. The NPR technique is used with two suggested scenarios to enhance the diagnostic accuracy of transformer fault types. The proposed NPR model is used with different data transformation techniques to select the best one to achieve a high diagnostic accuracy. The proposed NPR model is compared with AI and conventional DGA techniques. Finally, the proposed NPR technique investigates the effect of the starting decomposing material on the severity class of each transformer fault type.
Section snippets
Proposed fault type and severity classification approach
Detection of transformer faults in an early stage is very important to avoid undesired outage. The identification of the fault is based on the dissolved gases due to the decomposition of the insulating oil under different stresses (electrical and thermal). Each fault type can be identified based on the concentration of the dissolved gases and the relations amongst them (gas ratios). The correct fault type detection and recognition are very valuable to maintain the continuous operation of the
n-octane (C8H18) starting decomposing material
Thermodynamic theory is based on the starting decomposing materials as n-octane. The n-octane compound is removed from the crude oil in production to formulate transformer oil [8, 17]. In addition, it considers a paraffin compound as the oil components that generate the fault gases when the oil is subjected to thermal or electrical stresses.
The thermodynamic model based on n-octane can be illustrated as follows [17]:
- •
Hydrocarbon decomposition is an endothermic reaction and then absorbs energy
NPR for transformer fault diagnosis based on DGA
The PR method is one of the easy methods which is available as an application on the MATLAB/Software, and it can be used to diagnose transformer faults based on DGA. Using neural network has several merits such as adaptive-learning, self-organizing, and fault-tolerance capabilities, and neural network can be used with the PR applications. The PR is a very easy method that must be used to recognize an item. It includes three procedures, namely, filtering, feature extraction, and classification.
Suggested scenarios
The identification of transformer fault types is obtained based on NPR using two suggested scenarios. The first one is two-stage scenario, whereas the second is three-stage scenario. The first scenario consists of three NPR models, whereas the second scenario has five NPR models. The transformer fault types are diagnosed in two stages in the first scenario, but they are detected in three stages in the second scenario. Thus, the first one is called two-stage scenario, whereas the second is
Conclusions
The present work presented a proposed DGA model using NPR techniques to enhance the diagnostic accuracy of the transformer fault types based on the DGA approach. Two scenarios were proposed to predict the transformer fault types. The first scenario diagnosed the transformer fault types in two stages, whereas the second scenario diagnosed the transformer fault types in three stages. A total of 446 dataset samples were used for training and testing to measure the accuracy of the proposed NPR
CRediT authorship contribution statement
Ibrahim B.M. Taha: Conceptualization, Methodology, Software, Writing - original draft, Writing - review & editing. Sobhy S. Dessouky: Visualization, Investigation, Supervision, Writing - review & editing. Sherif S.M. Ghoneim: Conceptualization, Methodology, Software, Writing - original draft, Writing - review & editing.
Declaration of Competing Interest
The authors confirmed that there is no any conflict of interest of the submitted paper under the title “Transformer Fault Types and Severity Class Prediction Based on Neural Pattern-Recognition Techniques”.
Acknowledgements
The authors would like to acknowledge the financial support received from Taif University Researchers Supporting Project Number (TURSP-2020/61), Taif University, Taif, Saudi Arabia.
The authors thank the electricity authority of Egypt for supplying the data samples of the DGA test.
References (34)
- et al.
Multiple incipient fault classification approach for enhancing the accuracy of dissolved gas analysis (DGA)
IET Sci. Measur. Technol.
(2019) - et al.
Integrated ANN-based proactive fault diagnostic scheme for power transformers using dissolved gas analysis
IEEE Trans. Dielectric. Electric. Insulat.
(2016) - et al.
Proposed diagnostic methodology using the cross-correlation coefficient factor technique for power transformer fault identification
IET Electric Power Appl.
(2017) - et al.
Fuzzy logic approach for transformer remnant life prediction and asset management decision
IEEE Trans. Dielectric. Electric. Insulat.
(2016) - et al.
Optimal dissolved gas ratios selected by genetic algorithm for power transformer fault diagnosis based on support vector machine
IEEE Trans. Dielectric. Electric. Insulat.
(2016) - et al.
DGALab: an extensible software implementation for DGA
IET Generat. Trans. Distribut.
(2018) - S. Wani, U. Farooque, S. Khan, D. Gupta, Md.A. Khan, “Fault severity determination in transformers using Dissolved Gas...
- et al.
Thermodynamic estimation of transformer fault severity
IEEE Trans. Power Del.
(2015) - et al.
Determination of partial discharge severity in power transformers based on the starting decomposing material
Int. J. Appl. Energy Syst.
(2019) - IEEE Standard C57.104-2008: ‘IEEE guide for the interpretation of gases generated in oil-immersed transformers’,...