Prognostic modeling of transformer aging using Bayesian particle filtering

The goal of condition monitoring is to accurately assess the current health of an asset, in order to generate a prognosis, i.e. predict its remaining useful life. In the absence of a fault which causes premature failure, transformer degradation is linked to paper aging. Research and experience have resulted in models of paper aging where hotspot temperature is the key driver. However, these deterministic equations give a false sense of certainty about remaining insulation life. This paper demonstrates the use of Bayesian particle filtering for transformer life prognostics. This technique allows quantification of the uncertainties surrounding aspects such as the initial degree of polymerization of the paper, the relationship between hotspot temperature and measurands, and the accuracy of measurements. A case study from an in-service 180 MVA transformer is used to illustrate its potential.