Towards cost-effective maintenance of power transformer by accurately predicting its insulation condition

Insulation resistance (IR) or Megger test has been commonly performed in both preventive and corrective maintenance activities to verify power transformers´ insulation condition. Other insulation diagnosis tests such as oil breakdown voltage (BDV), water content and dissolved-gas-in-oil analysis have been conducted along with the IR test. In this paper, a prediction model is developed to correlate IR measurements of the power transformer with its oil quality parameters, the concentration of its total dissolved combustible gases (TDCG), and its carbon dioxide to carbon monoxide concentration (CO₂/CO) ratio. Four models, based on feed-forward artificial neural networks with back-propagation, are trained on collected data of real measurements. Accuracy levels of 96%, 84%, 88%, and 91% are obtained for BDV, water content, TDCG, and CO₂/CO ratio respectively. Utilizing the proposed model can reduce maintenance costs by preventing and shortening transformers´ outage times using inexpensive test, i.e. using IR test only.