Power transformer health prediction using machine learning

Ensuring good conditions and functionalities of these power transformers, these units are constantly monitored and maintained through the implementation of various condition-based maintenance activities. However, despite all of these preventive maintenance practices in place, some transformer defects are still left undetected, especially at an early stage. There is a lack of a holistic risk evaluation system in the power utility company to support and guide the scheduling and prioritization of condition-based maintenance activities. It is reported that there was a total of 20 power transformer failure cases during the years 2005-2019. These failures led to higher operating expenses, arising from the cost of repair and loss of revenues due to outages and downtime. As such, the outcome of this research aims to fill in this gap in the preventive maintenance system currently in practice in the power utility company by developing a transformer failure prediction system to complement the existing maintenance testing activities that are performed routinely as a part of condition-based maintenance in Malaysia. A Tier 1 to Tier 2 prediction algorithm is developed in this project with the help of artificial intelligence to accelerate the availability of Tier 2 electrical test results. This allows early assessment of the transformer's electrical parameters. Thereafter, the predicted Tier 2 test results can beused in conjunction with transformer age, loading, visual inspection as well as Tier 1 oil test results to predict failure probability and fault type through the development of a lookup table. Overall, this algorithm aims to speed up and improve the transformer health assessment to act as an early warning system for future tripping and failure events. This allows condition-based maintenance activities that are currently in practice to prioritize transformers that are undergoing more severe deterioration before permanent irreversible damage occurs.