Comparison of three methods for the quantitative analysis of common cause failures

A comparison is presented of three methods to predict the reliability characteristics of redundant systems subject to independent and common cause failures. Markov models are used to show that the ''beta factor'' method, developed and used in the HTGR risk assessment study conducted at General Atomic, is theoretically consistent with alternative approaches that are based on the multivariate exponential distribution developed by Marshall and Olkin. Comparative assessments of simple redundant systems based on actual reliability data are used to investigate differences between the ''beta factor'' method and the ''geometric mean'' approach used in the Reactor Safety Study.