Reliability demonstration of a multi-component Weibull system under zero-failure assumption.

摘要

This dissertation is focused on finding lower confidence limits for the reliability of systems consisting of Weibull components when the reliability demonstration testing (RDT) is conducted with zero failures. The usual methods for the parameter estimation of the underlying reliability functions like maximum likelihood estimator (MLE) or mean squares estimator (MSE) cannot be applied if the test data contains no failures. For single items there exists a methodology to calculate the lower confidence limit (LCL ) of reliability for a certain confidence level. But there is no comparable method for systems. This dissertation provides a literature review on specific topics within the wide area of reliability engineering. Based on this and additional research work, a first theorem for the LCL of system reliability of systems with Weibull components is formulated. It can be applied if testing is conducted with zero observed failures. This theorem is unique in that it allows for different Weibull shape parameters for components in the system. The model can also be applied if each component has been exposed to different test durations. This can result from accelerated life testing (ALT) with test procedures that have different acceleration factors for the various failure modes or components respectively. A second theorem for BX – lifetime, derived from the first theorem, has been formulated as well.;The proven theorem is integrated into a model to analyze the sensitivity of the estimation of the Weibull shape parameter β. This model is also applicable if the Weibull parameter is subject to either total uncertainty or of uncertainty within a defined range.;The proven theorems can be utilized as the core of various models to optimize RDT plans in a way that the targets for the validation can be achieved most efficiently. The optimization can be conducted with respect to reliability, BX – lifetime or validation cost. The respective optimization models are mixed-integer and highly non-linear and therefore very difficult to solve. Within this research work the software package LINGO™ was utilized to solve the models. There is a proposal included of how to implement the optimization models for RDT testing into the reliability process in order to iteratively optimize the RDT program based on failures occurred or changing boundary conditions and premises.;The dissertation closes with the presentation of a methodology for the consideration of information about the customer usage for certain segments such as market share, annual mileage or component specific stress level for each segment. This methodology can be combined with the optimization models for RDT plans.;The first theorem on LCL of system reliability is firstly proven for systems with two components only. In the following the proof is extended towards the general case of n components. There is no limitation on the number of components n. The proof of the second theorem on BX – lifetime is based on the first proof and utilizes the relation between BX and reliability.;Key words: System Reliability - Series System - Reliability Demonstration Testing Weibull Distribution - Accelerated Lifetime Testing - Lower Confidence Limit Confidence Level - Lower Confidence Bound - Zero Failure - Success Run.