Moisture evaluation of OIP bushings using artificial neural networks and dielectric frequency response.

Transformers are among the most expensive items among power system apparatus, and they plays a crucial role in the resulting power systems. Oil-impregnated paper (OIP) bushings form a key part of many transformers, and the status of these bushings may thus affect transformer performance. Monitoring OIP bushings is thus important for the safe operation of transformers, and one of the factors that can negatively affect OIP bushings is moisture. Effective evaluation of moisture in OIP bushings is thus required to ensure the safe operation of power systems. The dielectric frequency response (DFR) method has been commonly used in the last decade to study moisture impacts on OIP bushings. In this study, however, the moisture level in OIP bushings was evaluated using a neural network trained on a database formed from previous DFR results for various OIP bushings. Using the finite element method (FEM), a 96 kV OIP bushing was simulated under six different moisture content levels. Statistical indices were then used to extract features from the obtained tanδ-f curves of the simulated bushing under these different moisture contents, with, fitting analysis used to generate additional data. A database was then established using the generated data, and a neural network proposed to categorise the database into three groups based on moisture levels. The results revealed that the neural network effectively categorised 99.5% of data correctly by moisture level, suggesting that an adequately tuned neural network can be used to evaluate the moisture level of OIP bushings. [ABSTRACT FROM AUTHOR]