Improving Reliability Analysis with System Data from Operating Environments

点击翻译按钮获取中文摘要

9812868 Kvam In industries that cannot afford extensive testing programs to ensure product reliability and effectiveness, operational data is necessary for product reliability assessment. This grant supports the development of statistical methods for deducing component reliability while the component is working in an operational system. Typical systems considered include k-out-of-n systems for which k or more working components are needed to maintain the n-component system. Research centers on analyzing data from operating systems in nonstandard (i.e., non-laboratory) conditions, including (1) censored or truncated data, (2) systems that are repaired imperfectly, (3) component groups with known stochastic differences, such as one component having higher average reliability than another, and (4) components that have known reliability properties, such as a failure rate that is known to increase with time or use. For example, with systems that are repaired imperfectly, a maintenance policy that dictates when old systems or components must be replaced with newer ones is critical. Naturally, this maintenance policy will be greatly affected if the failure rate of the components is known to increase with system use. A primary application for the statistical methods developed from this grant are from the nuclear industry, where components are so reliable that component testing leads to few or no failures. In many nuclear systems, component redundancy is used to ensure high overall reliability, and system operation can sometimes mask individual component data. Censoring, imperfect repair, and order restrictions often combine in this environment to make inference problems difficult. As another application, producers in the software industry risk having their products become obsolete during prolonged testing programs, thus data from day to day operation may be necessary to augment accelerated life testing and other planned analyses of software reliability.