Abstract: The condition of the oil-impregnated paper is an essential point of life diagnostics for a power transformer. The degree of polymerization (DP) of paper insulation is considered as a good indicator for the determination of the deterioration level of the insulation paper that indicates the remnant life of the transformer. In these years, researchers have been able to implement classification analysis methods on a power transformer through a database of measurement data. Further studies related to the development of machine learning for the assessment of power transformers must be accomplished in order to formulate a comprehensive model. The objective of this paper is to develop a reliable algorithm to determine the current state of the oil-paper insulation based on the monitoring characteristics. With nominal classification base and numerically base using Fuzzy Inference System (FIS) and Back Propagation Neural Network (BPNN), and study its behavior. Both methods evaluate dielectric characteristic parameters, i.e., acidity and interfacial tension (IFT), of the insulating oil, and dissolved gas analysis (DGA) measurement results; the concentration of carbon monoxide (CO) and carbon dioxide (CO2), and four possible combinations variable input. This paper describes the structure of FIS and BPNN and gives a comparison of both methods to the performance to data sets of a real transformer fleet. The result shows that both models can be used to predict the value of DP accurately and to improve the reliability of the result.

Key words: Transformer, degree of polymerization, dissolved gas, dielectric characteristic, FIS, back-propagation neural network.

Cite: Martua Mario Gultom, Tobias Kinkeldey, Peter Werle, Suwarno , "Machine Learning Approaches for Estimating the Degree of Polymerization of Paper Insulation Impregnated with Uninhibited Insulation Oils," International Journal of Materials Science and Engineering, Vol. 9, No. 4, pp. 59-70, November 2021. doi: 10.17706/ijmse.2021.9.4.59-70
Copyright © 2021 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).