| New Bayesian models for analyzing the reliability growth of a complex system are introduced, which allow for the incorporation of delayed fixes. These models, unlike models in the literature, allow for variability between prototypes and the incorporation of the decision maker's expert judgment. Different choices of informative prior are considered. Bayes inferences for models, which include posterior pdf and mean for parameters and credibility interval estimation for system failure rate, are provided. Numerical studies and Monte Carlo comparisons are performed. For the purpose of comparison, three different types of simulated test data are considered. When only a few data and the modest amount of prior information on the parameters are available, simulations show our Bayes models largely outperform the popular Crow model (1975) in the estimations of parameters and system failure rate. In application, a delayed fixes test plan DF(n) is introduced to decide the type of test and timing for fixes in a reliability development test. We set up a cost model to solve the DF(n) policy problem. DF(n) analysis provide a way to see how critical the constrains of a program are. A Bayesian decision algorithm is provided, which can reflect the actual changes of prior status. |
