Bayesian formulation of the best of liquid and solid reliability methodology

An approach to reliability prediction for liquid or solid rockets is presented. The approach is based on Bayesian statistical methods and incorporates existing reliability methodology. The Bayesian model provides a framework for incorporating various types of information and for regular revision and updating of predictions. It facilitates the calculation of confidence intervals for reliability and of required test numbers to achieve reliability goals. INCE its inception, the rocket propulsion industry has concentrated primarily on designs that maximize payload capability in the context of specified, usually well-defined, mission and performance considerations, with costs and reli- ability of secondary concern. This is not to say that reliability was not considered to be important; it is often a critical factor. It is, however, much more difficult to predict than perfor- mance and weight, particularly at the design stage. As a result, x reliability evaluation was based mainly on full system testing and operational outcomes. The reliability levels that the industry has succeeded in achieving have come about through the use of large margins (safety factors) in part design and of a build-test-fail-fix ap- proach. The success of the design as far as reliability was concerned thus depended on factors that were determined by engineering judgment derived from the company and engi- neer's background, past experience, historical data, trial and error, handbook statistics, and similar information sources. Reliability analyses that were performed, if any, were usu- ally based on oversimplified models and assumptions. In addi- tion, process and operations-induced failures were seldom adequately addressed, if even included in the reliability analy- sis. As a result of these and many other difficulties, past reliability predictions have almost invariably overestimated the reliability of an operational system. Over the past few years, this situation has been changing considerably, primarily as a result of the reliability and cost requirements specified for the Advanced Launch System (ALS), later called the National Launch System (NLS). For example, the ALS Phase A specification required that, for the space transportation engine (STE), a 0.99 reliability be achieved with 90% confidence. This greatly stimulated interest in analytical efforts toward early quantification of reliability, leading in turn to recognition of the need for careful and thorough identification of failure modes as well as the need for development of more realistic engineering and statistical models of failure mechanisms and their likelihood of occur- rence. Moreover, funding is generally not available to achieve high reliability through an extensive development program. Further impetus to early quantification and prediction is the importance of reliability in evaluating competing systems (liq- uids and solids) at the very earliest stages of design. An initial