Reliability growth of multi-stage single shot systems

Traditionally, reliability growth models are used during development to project growth as the system proceeds through the Test-Analyze-And-Fix (TAAF) stages, and do not continue to be utilized once the system is fielded. In most cases, demonstrated reliability replaces predicted reliability at this point. Single shot systems such as weapons or launch vehicles present a unique challenge in this area. Though these systems remain in storage or standby for long period of time, they are required to be highly reliable and operational use for these systems can remain low. Determination of failure cause and correction during operational use is difficult or impossible for most of these type systems, and so the reliability prediction challenges that exist for development systems remain throughout the life cycle of a single shot system. This requires continued testing of representative articles to measure reliability throughout the life of the system. Highly competitive budget environments limit testing resources for these complex expensive systems. In the best case there are a limited number of data points for classic reliability calculation for systems where high confidence is required. This dissertation uses a Bayesian model to look at continuing the application of reliability growth models throughout the life of a single shot system. Traditional attribute growth models characterizing each event as either a success or failure are an over simplification that does not adequately address the complex nature of today's systems. Models are extended here to account for a multi-stage sequential system and a real data set from the complete test and operational history of a multi-stage single-shot system is examined. Finally this paper examines how the availability of continuously updated reliability predictions can support programmatic decisions throughout a system's life cycle removing decision maker's dependence on estimates formed at the initiation of the program.