A Beta approximation of unimodal density in finite interval

• Main Authors: HU, WEI-YUAN, 胡偉元
• Other Authors: LI, MENG-FENG
• Format: Others
• Language: zh-TW
• Published: 2009
• Online Access: http://ndltd.ncl.edu.tw/handle/63624695121315318147

碩士 === 國立臺北大學 === 統計學系 === 97 === Reliability is the probability that a component or a system can function at designed level during some specified period of time. When the distribution of component lifetime is unknown, nonparametric methods are used to estimate the system reliability, Rs(t)=P{T≥t}. Due to the randomization of sampling, the reliability of a component varies from lot to lot. In order to describe further the variation and uncertainty of the reliability, it can be viewed as a random variable, also called believed-reliability, which is often assumed to be distributed as a Beta distribution. The purpose of this research is to present a numerical method to evaluate system believed-reliability by component believed-reliabilities under the circumstance that components distributed as Beta function independently and through the system structure function. Let the believed-reliabilities of two components at time t be independent random variables X~Beta(α1,β1) and Y~Beta(α2,β2), α1, β1, α2, β2 > 0. This research first discusses the case of a series composed of two components to find the Beta approximation for the believed-reliability, Z=XY by minimizing the L1-norm between the density functions, and then extends it to the case of the series and parallel composed of n components. The Beta approximation of the believed-reliability for the system composed of series, parallel, series-parallel and parallel-series configuration can be derived by the previous methods.