【作者】 张红林；

【导师】 张春元；

【作者基本信息】 国防科学技术大学 ， 计算机科学与技术， 2011， 博士

【摘要】 随着科学技术特别是计算机技术的发展,各种控制和容错技术被广泛应用,各元件之间相互作用和影响,许多系统的结构变得越来越复杂,系统可靠性表现出复杂性、依赖性、非单调性、多态性等特征,使得可靠性分析面临越来越高的挑战。复杂性、顺序性、依赖性是动态系统的三个典型特征,本文针对这三个典型特征,从模块化分析、动态系统失效模式、不完全覆盖、可靠性仿真等4个方面深入研究适用于动态系统的可靠性分析模型。本文工作的主要贡献和创新总结如下:1.基于模块化的可靠性分析技术研究提出基于亲戚依赖关系的独立模块识别方法。该方法将动态故障树转化为依赖树,依赖树只包含事件之间的依赖信息,去除了故障树中各逻辑门的逻辑含义;通过改进深度优先遍历算法,使其适用于含有重复事件的情况;通过为每个节点加入祖先集属性和依赖集属性,使其适用于事件间含有相互依赖关系的情况。该方法以各节点的祖先集属性和依赖集属性进行识别,与相互依赖基本事件、重复事件、动态逻辑门无关,适用于具有相互依赖基本事件和重复事件的动态故障树。经过实际案例验证和仿真验证,该方法适用于具有相互依赖基本事件和重复事件的动态故障树。提出基于同构对象的故障树分析方法。该方法通过识别故障树中的同构对象,将故障树分成几组不同的模块,同类模块仅求解一次,对于具有较强结构化特性的系统,能够明显减少其求解开销,提高分析效率。该研究使得现有的模块化分析方法更易于应用于动态系统的可靠性分析中。2.基于扩展割序的系统失效模式研究提出定性和定量分析动态系统失效行为的扩展割序模型。首次提出扩展割序的概念,将“导致系统失效的带有某种时序限制的最小基本事件集合”称为扩展割序,其与以往类似概念相比扩大了外延、增强了表达能力,并不要求该集合中所有的基本事件都处于某个特定顺序中,甚至没有时序限制。提出动态故障树最小扩展割序集的生成算法和不交化算法,并将不交化扩展割序集转化为标准扩展割序集。然后对标准扩展割序各割项进行冲突检测、时限集精简、基本事件集排序和量化计算,求得各标准扩展割序的量化结果,再综合求解系统可靠性参数。实验验证表明,将该模型应用到OBC系统上时,相比同类模型,求解开销从14.42s减少到5.82s。该模型解决了如何定位系统失效模式,以及如何利用这些失效模式获得系统可靠性参数的问题。3.基于不完全覆盖模型的可靠性分析技术研究提出基于模块化组合分析的不完全覆盖模型求解方法。该方法对基于表格的算法进行拓展改进,利用ELC模型和FLC模型,对各逻辑门进行了分析,并给出了求解的算法或公式,同时舍弃了原算法中利用表格进行循环调用的实现手段,改以现代程序设计技术中普遍支持的递归函数实现。提出基于扩展割序的不完全覆盖模型求解方法。该方法对SEA算法进行拓展改进,利用不交扩展割序的概念,对不完全覆盖模型的可靠性进行求解时,将其分解成两项进行。仅需对各元件的可靠性参数进行轻微调整,便可以使用扩展割序模型计算第二项。该研究在一定程度上解决了动态系统不完全覆盖模型的可靠性度量问题。4.基于通用更新过程的可修系统可靠性仿真模型研究提出一般分布的自适应近似抽样方法AAAS。该方法根据分布函数的特性自适应调节“拟合参数”,以达到即快且准的抽样结果。实验验证表明,在划分同样的时间段数下,该方法比同类方法具有更高的抽样精度。提出基于通用更新过程的可修系统可靠性仿真模型。该模型以元件修复效力参数综合考虑元件维修后可能处于的各种状态,并利用AAAS方法抽样出下一模拟时刻;基于扩展割序理论,给出系统修复策略;然后在以上研究的基础上,从系统成功完成任务能力的角度出发,利用事件表建立了动态可修系统仿真的逻辑关系,根据对系统完成任务和元件运行特征的识别得到反应系统运行的整个历史纪录和可靠性参数的统计值。该研究使得仿真模型能够更加真实地反映系统的运行过程。综上所述,本文主要针对动态系统所具有的“复杂性、顺序性、依赖性”等特征,展开相关可靠性分析技术的研究。这些研究或者首次给出了相关问题的解决方法;或者在未成熟方法的基础上进行基础理论拓展,提高模型的表达能力和求解效率;或者对现有技术进行改进,扩大其适用范围;或者从较少被关注的角度展开研究。更多还原

【Abstract】 With the development of science and technology, especially computer technology, various control and fault tolerate techniques are widely applied, and the structure is becoming more and more complex for many systems. Since the components among system are interrelated and interactional, the reliability behaves some characteristics, such as completely, sequence, interdependence, nonmonotonicity, multi-state, etc, which make reliability analysis face the challenge. Three typical characteristics of dynamic system are complexity, sequence and interdependence. This thesis will consider these three characteristics and research the reliability analysis models of dynamic system from four aspects, i.e. modular analysis, failure mode of dynamic system, imperfect coverage model, and reliability simulation.The main contents and conclusions of the thesis are outlined as follows:1. Research on the technology of reliability analysis based on modularizationAn identification method of independent module based on kinship dependency relation is proposed. This method converts DFT to dependent tree, which only contains the dependency relations among the events, and removes the logic meaning of each logic gate. Through the amelioration of depth-first search algorithm, this method can be applied to DFT with repeated events. Through attaching ancestor property and dependency property for each node, this method is appropriate to deal with the interdependent relation among the events. This method identifies independent modules through the ancestor property and dependency property, and has no relation with the dependent basic events, repeated events or dynamic logic gates, so that it can be applied to dynamic fault tree with dependent basic events and repeated events. Both the practical case and simulation indicate that this method can be applied to DFT with dependent basic events and repeated events.An analysis method of fault tree is proposed based on isomorphic object. Through identifying the isomorphic object, decomposes the fault tree into various groups of independent modules. Since each group of independent module is only computed once, this method can greatly reduce the computing cost and improve the analysis efficiency for those systems with structural characteristics.This research makes existing modular analysis methods to be more easily applied to dynamic system.2. Research on system failure mode based on extended cut sequenceAn extended cut sequence (ECS) model to analyze the failure behaviors of dynamic system qualitatively and quantitatively is proposed. The concept of ECS is firstly proposed, which is the set of basic events under some temporal constraints. ECS enlarges the extension and the expressive capabilities, and does not require the basic events in a particular order, not even temporal constraints. Generative algorithm and disjoint algorithm of minimal ECS set are given, and then standard ECS set is transformed from disjoint ECS. Then, the test of conflict, the refinement of temporal restriction set and the topology sort of basic event set are carried out to quantitatively analyze each standard ECS. Finally the system reliability parameters are settled through synthetic solving. The experiment on OBC system shows that this method can decrease the time cost from 14.42s to 5.82s compared to aother recent model.The reliability analysis model based on ECS can locate the failure modes of dynamic system and solve the problem how to compute the unreliability of dynamic systems using these failure modes.3. Research on the reliability analysis based on imperfect coverage modelA modular solving method for imperfect coverage model (IPCM) is proposed. This method expands the table-based algorithm. Based on ELC (Elementary Level Coverage) model and FLC (Fault Level Coverage) model, each logic gate is analyzed, and the algorithm and formula is given. This method discards the recursive call in the original algorithm, and uses the recursive function which is supported in almost all the modern programming techniques.A solving method based on ECS for IPCM is proposed. This method expands SEA (Simple & Efficient Algorithm) algorithm. Based on the conception of disjoint ECS set, the IPCM failures is separated into two terms to compute the system reliability. This method can use the model of ECS to compute the second term, by slightly alter the reliability parameters of each component.This research solves the problem of IPCM reliability analysis for dynamic system to a certain degree.4. Research on the reliability simulation model of repairable system under generalized renewal processIn order to accurately acquire the sampling time of residual life, an auto adjust approximate sampling (AAAS) method is proposed for general distribution. Based on the feature of the distribution function, AAAS can automatically adjust the fitting parameter to quickly and accurately sample the simulation time. The experiment indicates that this method has higher accuracy than another recent method under the same number of time segments.With respect to the reliability analysis of repairable system, a reliability simulation model under generalized renewal process is proposed. This model samples the next simulation time by the method of AAAS, and uses the repair effective parameter to represent the state after repair. Then ECS theory is used to give the repair strategy. Finally based on the above considerations, from the aspect of successful accomplishing the task, the logic relationships are acquired by means of event lists. According to the identification of system state, the result of simulation provides the entire history of the mission and statistics of reliability parameters.This research make the simulation reflect the truer picture of the run process of the system.In a conclusion, this thesis carries out the research on the technology of reliability analysis with respect to the characteristics of dynamic system, such as complexity, sequence, interdependency, etc. Some researchs solve the unsolved problem. Some researches expand the fundamental theory based on the immature method, and improve the expression capabilities and solving efficiency. Some researches improve the existing technologies and enlarge the applicability. Some researches are carried out from the aspect which received less attention.更多还原