Reliability Growth Testing Assessment And Reliability Qualification For Complex System With Small Sample Size During Multi-Stage Development

During the development of complex weapon system, with multi-stage from the demonstrate prototype, principle prototype to formal prototype, the reliability growth tests were adopted to eliminate the weaknesses in system design or manufacture. As a result, the complex system reliability was improved gradually. The reliability growth test (RGT) data of complex system was a dynamical statistic population, and it was difficult to get scientific and reasonable analysis results. Therefore, the RGT analysis was an important theory problem. At the same time, for the cost of complex system was very high, the reliability qualification with small small local test by making full use of multi-source prior information was also an important theory problem. In this dissertation, the complex system RGT theory and reliability qualification methods were researched with Bayesian method as main line, aiming at the theoretical and practical problems of the complex system RGT and reliability qualification. The central works of this paper were followed:1. The traditional parameter model was modified by Bayesian method for multi-stage RGT assessment. The reliability growth process was modeled by the non-homogeneous Poisson process, and a Gamma-Beta prior was proposed for the parameter of NHPP corresponding to the reliability growth process. The model was suitable for multi-stage RGT assessment with small sample size by the transfer rule for reliability growth's information.2. The Bayesian model of RGT was improved. The Bayesian model for reliability growth with New Dirichlet prior distribution method was improved from two aspects for exponential case and binominal case. Firstly, the method for determining prior distribution parameters was given by the method of optimization, it was easy to confirm the parameters of prior distribution, and it solved the problem of how to verify the hyper parameters of the new Dirichlet prior distribution because of these parameters having no specific physical meaning. Secondly, the Gibbs sampling algorithm was used to solve the problem that interference on parameters of Bayesian posterior higher dimensions can't calculate indirectly. These improvings paved the way for the application of the Bayesian model for multi-stage RGT assessment with small sample-size.3. A Bayesian plan of qualification test of complex system was research for making use of prior imformation during multi-stage development. By considering not only the testing information from related products but also differences between such products, a new mixed prior distribution was used by introducing the factors of inheritance, then the Bayesian decision model of the qualification plan was established. Based on the method aboved, the Bayesian plans of qualification test were researched for binomial case and exponential case. Moerever, for the exponential case, it was proved that the classical fixed time testing scheme was a special case of the Bayesian's.The results of the research above were successfully applied to the RGT assessment and the plan of qualification test of a weapon system, and provided a new technology way to the RGT assessment and the plan of qualification test of complex system with small sample size.The research in this dissertation provides important guidances to the complex system RGT assessment and the plan of qualification test, and has high practical value to improve the complex system RGT assessment and the plan of qualification test.