A STATISTICAL-METHOD FOR EVALUATING ELECTRICAL FAILURES

Statistical tools have been developed for performing Bayesian analysis of electrical failures in the case of two-parameter distributions. So ftware implementations graphically present confidence regions for the two parameters and confidence intervals for the failure probability as a function of applied stresses. For the Normal and Lognormal distribu tions these are the same as classical confidence regions, while for Gu mbel and Weibull distributions they are nearly the same. Examples are provided for the energy capacity of metal-oxide varistors and for insu lator flashover probability.