A Regression Algorithm for Transformer Fault Detection

A transformer's failure can lead to disruption in power, decrease in system reliability and monetary loss to the utility and distribution companies. Fault detection of transformers is a critical step for improving the reliability of distribution systems. Regular maintenance checks can detect most of faulty conditions, but due to high cost and difficulty, the maintenance checked can only be performed annually. This paper proposes a simple on-line monitoring algorithm that uses a minimum set of sensor information, including ambient temperature, hot spot temperature, and load, to estimate several system parameters such as oil and thermal properties of the transformer and detect abnormal behavior. Fault can be detected when these parameter estimations experience sudden changes or the estimated values have sufficient deviation from their nominal values.