Selection of Beta Prior Distribution Parameters from Component Failure Data

A description of classical and Bayesian techniques to estimate component failure probabilities is presented. Of particular concern is the estimation, from typically sparse component failure data, of values for the parameters of the assumed beta prior distribution (used in the Bayesian analysis) and of the failure probability distribution for a particular component with an observed performance history. Three methods for the parameter estimation are described and compared, viz. (i) matching data moments to the prior distribution moments, (ii) matching data moments to marginal distribution moments, and (iii) the maximum likelihood method. Results are presented for data from standby diesel generators used in several nuclear power plants.