A Novel Approach Using Feature Engineering for Fault Prediction in Transformers

Condition monitoring of transformers is crucial to maintain and enhance their life longevity. Many traditional approaches have been used in the past to predict the health and identify the faults occurring in transformers. In this paper, some traditional and newer machine learning algorithms are used to predict the type of fault in a transformer. Feature engineering is applied to study its effect on the accuracy of fault classification. The results prove the adaptability of the algorithms to small and medium size datasets. But it also leads to an important observation that the accuracy of these algorithms decreases with increase in size of the dataset.