A BAYESIAN APPROACH TO SETTING EQUIPMENT PERFORMANCE CRITERIA

摘要

Reliability performance criteria for equipment important to safety are often set using a hypothesis-testing approach that disregards the analyst's complete state of knowledge about the reliability parameter. This is an especially acute problem when the evaluation period is too short to obtain a significant amount of equipment-specific data. From the analyst's point of view, false positives are undesirable, as they lead to unnecessary expenditure of resources. The regulator is more concerned with false negatives, of course, which are likely, given limited data and a typical significance level of 0.05. The Bayesian approach outlined in this paper uses the posterior predictive distribution to set the performance criterion, expressed typically as an allowed number of failures in a specified number of demands or operating time. This approach uses the entire uncertainty distribution for the parameter of interest, instead of a single value (e.g., mean) and can be applied easily in a sequential fashion from one evaluation period to the next. It is easily implemented using freely available simulation software, and does not depend on the prior being in a particular mathematical form (i.e., conjugate).