Chinese Journal of Computers   Full Text
Title	Bayesian Networks Based Reliability Analysis of Phased-Mission Systems
Authors	LIU Dong1),2) ZHANG Chun-Yuan1) XING Wei-Yan3) LI Rui1)
Address	1)(School of Computer, National University of Defense Technology, Changsha 410073) 2)(Key Laboratory of National Defense Technology, Academy of Equipment Command & Technology, Beijing 101416) 3)(China Huayin Ordnance Test Center, Huayin, Shaanxi 714200)
Year	2008
Issue	No.10(1814—1825)
Abstract &   Background	Abstract The paper presents a Bayesian networks (BN) framework for the reliability analysis of phased-mission systems (PMS), named PMS-BN model. A PMS consists of consecutive and non-overlapping time periods, with system configuration, success criteria, and component behavior varying from phase to phase. Firstly, each phase is represented by a BN framework, named phase-BN. Then, in order to figure the dependences across the phases, all the phase-BN are combined by connecting the root nodes that represent the same component but belong to different phases, and connecting the leaf nodes of phase-BN with a new node representing the whole PMS mission. The new constructed BN is called PMS-BN. In PMS-BN model, each phase time is divided into m segment, and the reliability analysis of PMS is performed by a discrete-time BN model acting on PMS-BN. Two examples are used to expatiate on the proposed approach. The PMS-BN based method provides a new efficient way to analyze the reliability of PMS, especially for those with dynamic phases. Moreover, it is also applicable to system diagnosis and sensitivity analysis. If all the non-root nodes in constructed PMS-BN own not more than 2 father nodes, the computational complexity of evaluating the PMS reliability is O(Nm3), where N is the number of non-root nodes. Keywords phased-mission systems； Bayesian networks； reliability analysis； computational complexity； sensitivity analysis Background The reliability analysis of Phased-Mission Systems(PMS) is different from that of normal systems, since the existence of more than one phase in PMS leads to some complexities which do not occur in a single phased-system. In PMS, dependences exist not only within a phase but also across the phases. The PMS reliability was investigated greatly in these ten years, and the research center lies in University of Virginia and Duke University, USA. There have been a variety of methodologies for the analysis of PMS, such as cut sets, binary decision diagrams (BDD) and Markov chain model. BDD based approaches provide an efficient way to analyze PMS. However, it is limited to the static systems (just like cut sets based models) and can not handle dynamic systems that are characterized by dynamic behaviors, such as function dependence, redundancy and sequential failure. Although Markov chain model is a powerful tool to handle dynamic systems, it has to confront the state explosion problem. Accordingly, it is a challenge to get an applicable and efficient method to analyze the general PMS that are characterized by both static and dynamic behaviors. The paper introduces Bayesian networks (BN) modeling into the reliability analysis of PMS, and presents a new model, named PMS-BN. PMS-BN provides a compact and intuitive way to analyze PMS with dynamic phases. In PMS-BN, the dependences within a phase and across the phases can be expressed easily by BN framework, and the reliability of PMS can be computed with less complexity compared by Markov chain model. This research is supported by the National Natural Science Foundation of China (60673148, 60703073) and the National High Technology Research and Development Program (863 Program) of China (2006AA704302). The aim of the research is to establish a systematic method to analyze PMS, such as space information systems and aircraft systems. Until now, the research team has published more than 20 papers about the fault tolerant design and the reliability analysis of space information systems.