Modeling data uncertainty in system availability analysis

Reliability and availability analyses of modern engineering systems are usually driven by statistical models of part and system lifetime. These models (E.g. Weibull distribution for part life) are based on accelerated test data, field experience or obtained from analytic probabilistic design algorithms. Traditionally, the uncertainty associated with these distribution parameters has not been quantified or factored into the availability model, and as a result only a mean value of system availability is reported. This often leads to serious errors in assessing the "spread" or variability in true system availability, which is an important metric for system engineers and decision makers. In this paper, two new methods for predicting statistical confidence bounds on system availability are presented. The first approach is based on sampling from multivariate parameter distributions and the second is based on sampling from joint likelihood-ratio confidence intervals. Finally, both algorithms are validated with numerical case studies.