Resumo:
This study makes use of artificial neural networks, a machine learning algorithm, for the classification of signals of partial discharges in high voltage insulators. In addition to that, these signals are further analyzed by means of implementations in Python programming language, seeking to define a severity degree for the partial discharge activity using a fuzzy inference system. The case study of this work is the city of São Luís, capital of the state of Maranhão, whose electric system is under concession of Equatorial Energia. The utility’s service area comprises a 69 kV transmission system, whose structures employ composite line post or suspension insulators. Partial discharges not only cause the degradation of the polymeric surface of the insulators, but can also evolve to extensive arcing, which can further cause flashovers that result in system faults and energy supply interruptions. This work has proposed the development of data acquisition and analysis algorithms for partial discharge signals, to compose an inspection instrument that can be used in the field by means of a compact and portable software-hardware system and an antenna for the acquisition of radiofrequency interference signals from discharge activity. Thus, the utility’s maintenance team will have a decision support tool to assess the structures condition, thus enabling the scheduling of preventive maintenance routines, if necessary, seeking to prevent power supply interruption events. From laboratory tests and field collected data, some results are presented, showing the consistency of the parameters selected to calculate the severity index. The increase of the severity degree is observed with the aggravation of surface pollution and rise of relative humidity.
