A Bayes procedure for estimation of current system reliability

During development of a new system, a test-analyze-and-fix program is often applied to eliminate the flaws in the first design of the system. If the corrective actions are effective, the system reliability increase. This process can be carried out for a planned time period or until a fixed reliability goal is achieved. In both cases, it is important to evaluate the system reliability during the test phase. Reliability growth can be reasonably modeled by a nonhomogeneous Poisson process with a decreasing power intensity law. For such a process, a Bayes procedure for estimating the failure intensity at the time of the last failure and of the current system reliability is analyzed for failure-truncated sampling. A large simulation study shows that the procedure, even for vague prior information, is more efficient than maximum likelihood and at least as efficient as the quasi-Bayes method of J.J. Higgins and C.P. Tsokos (1981).<
>