Steady-state availability estimation using field failure data

This paper introduces a novel technique for computing confidence limits associated with steady-state availability estimation using field failure data. The proposed cumulative downtime distribution (CDD) method implements a simple, though powerful, availability inference procedure based on the statistical properties of the distribution of sample means of the cumulative system outage time. Another advantage of this new approach over more traditional estimation methods is that it makes no assumptions regarding the lifetime or time to repair distributions of the system under observation. A simulation model was developed to compare the coverage probability of the confidence limits computed using the CDD method and the more traditional two-state equivalent (TSE) method. Simulation runs are used to support that confidence intervals determined with the CDD method seem to be exact. On the other hand, confidence intervals determined using the TSE method seem to be only approximated. Additionally, the CDD method was shown to provide an excellent framework for the application of other statistical inference procedures such as hypothesis testing. Our future research intends to verify the quality of the CDD method using more complex system models and more exhaustive simulation experiments. We also want to verify the algorithm behavior applied to deployed systems with different maturity levels.