MULTIPLE-FAILURE PROBABILITY PREDICTION OF EMERGENCY DIESEL GENERATOR GROUPS AND GENERAL k-OUT-OF-n SYSTEMS

Abstract

Since multiple-failure event can destroy the safety policy of redundancy, accurate prediction of multiple-failure probability is of great importance. However, the statistical dependence between component failures might lead to unrealistic estimation of the conventional system failure probability model since it is valid only in the situation of independent component failures. On the other hand, the lack of multiple-failure event data makes the statistical estimation of multiple-failure probability suffer serious uncertainty. General failure event data experienced by other systems might be the only data available to estimate the system under study. In order to evaluate the target system in such a situation, an appropriate approach is highly required by which the right information can be mined from the operating experience of reference systems. Based on the multiple-failure information contained in load-strength interference relationship, this paper presents an approach to estimate multiple-failure probability of dependent k-out-of-n system according to failure event data available. The data may come from the operating experience of the system to be evaluated or reference systems with similar load-strength interference relationships. As examples, the failure probabilities of emergency diesel generator groups are estimated according to the multiple-failure event data of reference groups of different sizes. The estimation results are consistent well with the operating records.