Bayesian Networks applied to Failure Diagnosis in Power Transformer

This work describes the structure, learning and application of Bayesian Network to diagnosis of faults in power transformer through the dissolved gases analysis (DGA) in oil. The Bayesian Network uses the concentration ratios of gases methane/hydrogen (CH₄/H₂), ethane/methane (C₂H₆/CH₄), ethylene/ethane (C₂H₄/C₂H₆) and acetylene/ethylene (C₂H₂/C₂H₄), as elements that activate the network diagnosis: normal deterioration, electrical failure and thermal failure. The learning was performed from historical database, and the Bayesian Network presented a high degree of reliability and consistency. The simulations suggest good results when compared to some existing in the literature.