Data mining applied to abnormality prediction in electrical substation transformersMatheus José da Silva, Starch Melo de Souza, Israel Cavalcante de Lucena, Hemir da Cunha Santiago, Eldrey Seolin Galindo, Antônio Janael Pinheiro, Luciana de Albuquerque Ro

Ensuring that assets in an electrical substation are managed only when needed is essential to minimizing maintenance costs. In this work, we present a system based on Artificial Intelligence (AI) to contribute to decision making about the performance of predictive maintenance. This system will provide substation operators with additional indicators of the operational condition of the transformers, determining which equipment is most likely to fail in the short term, based on analog and digital data from the substation supervision system. Since AI techniques require that the data obtained have the quality to provide satisfactory results, data mining techniques were applied to the records of equipment in a substation. During the analysis of the data provided, several signs of unwanted events were observed, such as abrupt fluctuations in oil temperature and variations in the frequency of alarms. In the analog data, we have records of fluctuations in voltage, current and oil temperature that are relevant to indicate the operating status of the equipment, in addition to the digital data provide information on alarms related to the equipment, which are records when the equipment exceeds some predefined safety threshold. This information was used to provide a broader view of the behavior of equipment during the construction of the predictive model, improving the detection of abnormalities. The system presented in this work models the typical behavior of the equipment, through the information mentioned in the previous paragraph. When predictions differ significantly from expected values, the system signals the operator of the presence of potential anomalies. The results obtained reveal the potential of the system, helping decision making in predictive maintenance of transformers. The development of an innovative methodology to extract knowledge from the supervisory system records proved to be a favorable field for future investigations.