Abstract:

A common practice is to use random sampling statistics to estimate the demonstrated reliability of a new system from the number of successes and failures. This can grossly overestimate or underestimate the true reliability because of the na?ve assumptions involved. A more detailed approach is presented which accounts for the characteristics of the test, the faults identified and the corrections made. The Bayesian framework accommodates a prior reliability estimate and comprehensively handles uncertainties. The examples show the wide areas of application from determining test strategies to tracking reliability growth. The applications described are believed to be new. The approach needs to be more widely implemented because of its features.