Transformer failure prediction using Bayesian analysis

A procedure is described for predicting transformer failures in order to quantify the expected impacts on service reliability. The procedure is designed to get the most out of sparse data by formal incorporation of engineering experience. The method is particularly well adapted to failure frequency forecasts and outage predictions for large, expensive apparatus for which accelerated life testing is not appropriate and for which historical failure data is limited because of an inherent low failure rate. The procedure makes use of a systematic method for combining only the most credible features of engineering models with real-world experience and has been referred to as both calibrated engineering analysis and combined analysis (CA). Bayesian methods are utilized to formalize the statistical aspects of CA. An application example is presented which demonstrates how the procedure was used to predict the economics of adding redundant transformer capacity at 20 single-transformer substations for the purpose of improving service availability