A novel method to estimate the lifetime of mineral oil-type power transformers based on the analysis of chemical and physical indicators using artificial intelligence

Oil transformers are one of the most widely used and vital equipment in power distribution networks. Therefore, in order to maintain the service of the electricity distribution system, special attention should be paid to the health and repairs of these equipment's in order to avoid possible costs due to their failure. Since the mineral oil used in these transformers is one of the strongest insulators, it needs a series of maintenance and periodic tests to increase the lifespan of the transformer. This paper proposes a new approach to provide an appropriate management decision and predict the remaining life of the mineral oil transformer using artificial intelligence. Health index formulation is a quick and efficient approach to combining various data and creating an indicator for asset management planning. While the health method using DGA is well presented in the literature, more attention should be paid to the combination with other parameters. Thus, in the proposed plan, evaluating the transformer's life with the actual health index formulation method includes additional factors such as load history, bush condition, ageing, and physical observations, along with common factors such as DGA, oil quality and dissolved gases in oil. Also, AI-based techniques have been used to increase accuracy. Furthermore, an integrated inspection and repair model is provided for maintenance management. Therefore, an optimal strategy is presented to determine the type of repair and inspection interval so that the transformer's life is managed by minimizing the total costs. Moreover, transformer faults are selected to manage the life of transformers and reduce blackouts. A model using transformer failure data has been developed to classify seven types of transformer faults. In order to validate the proposed procedure, the data collected for different transformers have been used. The results show that the proposed model is reliable and can provide a timely asset management decision with minimal cost.