Condition Monitoring of Oil-immersed Transformers Using AI Edge Inference for Incipient Fault Prediction: A case study

The question of transformer dependability and reliability remains a salient feature in determining the stability of the power supply system. Transformers are amongst the most costly and crucial components of the transmission and distribution network. Every year in Kenya, about 3866 distribution transformer failures are witnessed, representing a failure rate of about 6%. A post-mortem examination of the failed transformers revealed that 99.1% of these failures are caused by faults that progressively deteriorate. These failures are costly since they cumulatively result in more than 25,000 hours of power outages annually. Considering Kenya's development agenda and industrialization priorities outlined in Vision 2030, this state is exceedingly undesirable. In this paper, a machine learning multinomial classification model is deployed to the edge of the network for condition monitoring of power transformer units. Specifically, oil-immersed transformers are considered. The model was built using the edge impulse service. From the results, an accuracy of 99.9% was attained, which shows that the edgedeployed machine learning model DGA can perform well even when deployed to the edge of the network. This study therefore recommends that the use of machine learning models for condition monitoring of power transformers should be explored further as this will aid in the efficient assessment, addressing, and acting on developing issues detected on transformer units.