Keywords

1 Introduction

The mineral oil-filled power transformers are considered critical and costly constituent of the power system [1]. The regular power supply is maintained on the cost of transformer’s scheduled condition monitoring. Generally, the liquid and solid materials are used to provide the insulation between core and windings. The solid insulation uses the thermally upgraded kraft (TUK) papers made of cellulose. The performance of the transformers is mostly reliant over the condition of insulating cellulosic paper (solid). The transformers undergo successive aging when it confronted to factors like temperature and moisture [2, 3]. These causes hasten aging, create acids and unwanted sludge components that affect the cellulose insulation straight [4]. A continuous negligence in transformer maintenance not only has negative effects on utilities’ finances, but also lowers the functionality and dependability of the equipment [5]. For maintenance purposes, a thorough inspection of the transformer provides a better understanding of its condition. Utilities must therefore frequently check on the physical state of the power transformers in order to estimate their valuable lifespan.

The status of the transformer can be readily inferred from how much the insulation made of cellulose paper has decomposed. The cellulose used to manufacture insulating paper is the linear polymer comprised glucose (monomer) units that are linked up all together to form a chain structure [6]. Every polymeric chain’s number of glucose units is called the Degree of Polymerization (DP), that is the most reliable parameter to measure the extent of cellulosic insulation decomposition [7]. DP is measured through viscometric technique, and it is an invasive and destructive process of measurement and thereby avoided. The thermal aging of the cellulosic insulation in the company of moisture/water generates few degradation by-products into the transformer oil [8,9,10,11]. They most likely include carbon oxides and furfurals. Furfurals have referred to a family of six heterocyclic compound, that includes 2-Furfuraldehyde (2-FAL), 5-Methyl-2-Furfuraldehyde, Furfuryl Alcohol, 2-Furoic Acid, 5-Hydroxymethyl-2-Furfuraldehyde, and 2-Acetyl Furan [12]. The 2-FAL dissipates in transformer oil in large proportion and it is thermally more stable than the other furanic derivatives and thereby considered as the more reliable and prominent aging indicator of the cellulosic insulation [13].

In accordance with the transformer paper insulation’s DP value, numerous research studies have been conducted to evaluate the transformer health. But, the DP measurement required intrusion into the transformer to cause a severe outage that will not economical near the utilities. On the other hand, several empirical models have been found in the literature to calculate the value of DP based on the concentration of 2-FA, such as Chendong, Stebbin, De-Pablo, and Pahlavanpour. These models take the 2-FAL concentrations (ppb or ppm) and estimate the value of DP, based on which current physical state of cellulosic insulation is identified. However, numerous soft computing techniques, including support vector machines, wavelet networks, fuzzy logic, neural fuzzy systems, and neural networks, have been documented in the literature to evaluate the health of transformers based on different parameters (especially, DP), and 2-FAL computation has not yet been attempted [14,15,16,17,18,19,20,21]. In this work, an Adaptive Neural Fuzzy Inference System (ANFIS) has been exploited to implement the utility of easily measurable parameters to estimate the molecules (in ppm) of 2-FAL in the transformer insulating oil. The system takes temperature and moisture as the inputs to estimate the value of 2-FAL which ultimately give the DP values and hence the status of the cellulose. The proposed method facilitates the utilities for developing a web-based real-time health assessment system. This not only makes the distribution transformers to complete their designed life but also overcomes the revenue losses for the utilities. The system has also an advantage that it avoids the destructive measurement for the transformer’s health assessment and thus the severe outages.

The paper is organized in the chronological order. The first section is the introduction part followed by ANFIS system description in the second section. The proposed methodology and development of the system are detailed thereafter. Finally, the result and discussion and conclusion are included in the last section.

2 ANFIS

The ANFIS employed hybrid input–output mapping technique that utilizes the benefits of neural network’s capability for learning within framework of FIS. The ANFIS’s generalized structure, which consists of nodes connected by directional links, is represented in Fig. 1. ANFIS structure consists of five layers, each having such nodes that can be either static or adjustable. The output obtained at the adaptive square nodes relies at the parameters that are used to create them, whereas spherical nodes responded as the stationary one and their output is reliant over the outcomes of preceding layers. The one-to-one linkage offers conduit the signal without altering its form [22, 23]. Here, Layer 1 is made up of the four nodes A1, A2, B1, and B2. These nodes stand for a particular fuzzy set which maps the input toward the output as per that how firmly it bears a resemblance to the define fuzzy sets. The shape of circular nodes in Layer 2 indicates its fixed nature. These circular-shaped nodes provide an output by multiplying the input signals by the product. Nodes that formed Layer 3 conduct the normalizing function for Layer 2’s output. Nodes in the Layer 4 are of adaptive type, meaning that their output relies on node’s ensuing parameter. The results for the layer, therefore, integrate the results of the prior layer. To create the concluding output, the Layer 5 applies the summing function of incoming signals.

Fig. 1
A schematic consists of nodes connected by directional links. Layer 1 is of 4 nodes A 1, A 2, B 1, and B 2. Layer 2 is of fixed nature nodes pi. Layer 3 is of the nodes of normalizing function for Layer 2 output N. Layer 4 is of nodes of adaptive types. Layer 5 is the concluding output sigma.

Generalize five-layered ANFIS structure

Full size image

3 Proposed Intelligent System for 2-FAL Estimation

The kraft paper is used to provide the insulation inside the transformers and made usually of cellulose (C6H10O5). The cellulose is polymer of glucose monomers in linear structure which breakdowns under successive aging. Due to overloading condition in the transformers, the temperature rises and distresses the cellulosic insulation [24,25,26,27]. The presence of moisture in the transformer accelerates this distress and thereby produces cellulose degradation by-products. The furanic derivatives and carbon oxides are released into transformer oil as the aging (due to decomposition of cellulose) by-products of the cellulosic paper breakdown. Among furanic derivatives, 2-FAL is dissipated in oil in large proportion and thermally stable and thereby predicts the cellulose condition accurately. The ANFIS model proposed here estimates the concentration of 2-FAL as the straight function of moisture and the temperature (ambient). The design of the proposed structure is depicted in the block diagram of Fig. 2.

Fig. 2
A schematic has input parameters on the left and output on the right. Input parameters have the function of temperature and moisture. Output has the function of 2-F A L.

Schematic of proposed ANFIS-based intelligent system

Full size image

The proposed ANFIS system takes temperature and moisture as the inputs and estimates 2-FAL as the output. The values of temperature were lying in the range from 90 to 130 °C and moisture (%) varies from 1 to 3%. A dataset which comprises 100 input–output points has been given to the system for the training. To obtain an optimal ANFIS system, different types and number of membership functions have been attempted. Trapezoidal membership function is found to be the best membership function with least error of 0.0024. The model is tested and validated for ten samples comprised different temperature and moisture values.

4 Result and Discussion

The ANFIS structure is designed and implemented using the MATLAB environment to estimate the concentrations of 2-FAL against easily measurable parameters. The optimal ANFIS model has been found using hit-and-trial method, i.e., by varying the type and number of the membership functions. The external experimental data available in literature are then tested using the optimal ANFIS model [11]. Table 1 displays the experimental data points used to evaluate the ANFIS model’s performance. Observing Table 1, one can conclude that the rise in temperature directly affects the evolution of 2-Furfuraldehyde in the transformer’s insulating oil. However, the increase in the moisture accelerated the 2-FAL generation more rapidly. As a result, the proposed ANFIS model was found to be more accurate for the detection of transformers’ cellulose degradation product more appropriately.

Table 1 Estimated output of the developed ANFIS system as the function of temperature and moisture
Full size table

5 Conclusion

The paper successfully implemented the ANFIS model for 2-Furfuraldehyde (2-FAL) estimation. The model employs the parameters like temperature and moisture to measure the 2-FAL. The parameters used to develop the system are easily measurable and thereby make the system more economical for the utilities. The ANFIS structure easily correlates the input and the output by appropriately tuning the membership function. The system has been tested and validated with the available data in the literature and found to estimate more precise and accurate outcomes. As a result, the ANFIS model offers us a productive substitute for measuring the values of DP with the help of diagnostic parameters while restricting the intrusive measurements approach.