In this paper, we analyze the probability of failure of space launch systems in their first five launches. The results are based first on frequentist statistics, then on a Bayesian analysis of the early performance of 41 types of rockets including NASA's shuttle. The Bayesian method proves particularly useful when the experience base is small because it allows estimation of the reliability of a new rocket before the realization of a large number of trials. We compute the probability distribution of the failure rate and its mean (failure probability) based on binary failure/no failure events. Our main conclusions are that the failure probabilities are higher for the first and second launches (~1/3 and ~ 1/4 respectively) than for subsequent ones -- a result that is perhaps lower than sometimes assumed -- and remains roughly constant thereafter. Within a family of vehicles (e.g., the Delta's), the first generation has a higher mean failure rate than the second and third.
展开▼