On Zero-Failure Testing for Bayesian High-Reliability Demonstration

Abstract

Recent results are summarized on the required testing of a system, in order to demonstrate a level of reliability with regard to the system's use in a process after testing. It is explicitly assumed that testing reveals zero failures, which is realistic in situations where high reliability is required and where failures during testing lead to redesign of the system. Several related aspects are discussed, including the choice of prior distribution, what to do if failures do occur during testing, and possibilities to take dependencies between tasks into account. Throughout, it is emphasized that, for reliability demonstration, one should try not to rely too much on mathematical assumptions that cannot be justified by the data, which gives particular restrictions owing to the nature of data from zero-failure tests. All of these topics raise interesting questions for future research.