【作者】 杨述明；

【导师】 邱静；

【作者基本信息】 国防科学技术大学 ， 机械工程， 2012， 博士

【摘要】 随着装备维修保障模式的逐步转变和故障预测使能技术的不断成熟，装备健康管理（Equipment Health Management，EHM）必然成为未来装备设计、生成和使用的重要组成部分。EHM能力一方面依赖于信息的处理与决策，另一方面更依赖于信息的获取。随着装备复杂性的增加，有必要在装备设计研制一开始就根据EHM的需求考虑可测性问题，选择全面的测试项目、配置合理的传感器、制定科学的测试时机，并采用相应技术手段嵌入设计到装备中。目前的可测性理论与技术主要面向状态监控与故障诊断，没有考虑故障大小、故障演化和健康评估对可测性的需求。如何根据EHM需求，在装备早期设计阶段并行开展可测性设计是提高EHM能力进而提高维修决策能力的根本途径，也是目前我军装备发展中亟待解决的重要问题之一。本文针对目前可测性中融入EHM功能需求后内涵体系尚不明确、关键技术有待理清和突破等实际情况，从可测性指标、可测性模型、可测性优化设计等方面进行系统深入研究。论文的主要研究内容与成果包括：1.在当前可测性理论和框架下，根据EHM的功能需求明确了面向EHM的可测性内涵，提出了面向EHM的可测性技术体系。2.针对目前可测性指标主要用于故障可检测和故障可隔离能力评价，不能有效描述故障可预测和健康可评估能力的问题，在深入分析EHM对可测性的本质需求基础上，从“准确性”和“时效性”两个方面构建了面向EHM的可测性指标体系，并分析了指标间的关联关系。3.系统地分析了可测性中考虑EHM功能后的建模需求，建立了面向EHM的定量不确定分层模型。在系统层，通过定量有向图和功能故障分析建立故障—测试相关性，并以概率、模糊和不确定的形式定量描述故障属性、测试属性和传播属性；在组件层，通过失效物理模型或扩展故障模式、机理和影响分析构建故障演化—测试相关性。装备定量不确定分层模型可以表示成一个多元相关性矩阵，基于该矩阵可实现面向EHM的可测性分析与评估。4.分析了可测性中融入EHM后对测试项目的需求，提出了面向EHM的测试优化选择与传感器优化配置框架与方法。首先提出了测点初步布置的一般原则和方法，提出了面向EHM的测试优化选择模型和方法。在此基础上，从故障特性、传感器特性、故障与传感器之间的匹配特性系统地分析了传感器对故障检测的不确定性，然后基于故障可预测对传感器的需求以及传感器对故障检测的不确定性，建立了以传感器总代价最小为优化目标，以传感器不确定检测下的故障可检测率、故障可隔离率和故障可预测率为约束条件的传感器优化配置模型，并设计了遗传算法进行求解。5.基于可测性和EHM融合后的内涵和技术体系，分析了测试时机优化的必要性和新需求，提出了基于Markov理论的面向EHM的测试时机优化制定方法。首先，以贮存模式装备为背景研究并提出了基于Markov更新过程的周期测试时机优化技术。进一步地，以使用模式装备为背景分析了EHM对动态序贯测试的需求；根据动态序贯测试的特点，给出了部分可观半Markov决策过程的形式化描述；通过引入信念状态把部分可观半Markov决策过程转化为完全可观信念半Markov决策过程；在此基础上建立了以装备在长期运行条件下的平均费用率（主要包括维修费用、测试费用和停机损失费用）最低为优化目标的动态序贯测试时机优化模型。该模型以装备的健康状态为基础动态决策下次测试时机，并考虑了装备健康状态评估的不确定性，更符合EHM的功能需求。所研究的动态序贯测试优化模型同样适用于贮存模式装备。论文以典型机电伺服系统为案例贯穿各章节，验证所提模型与方法，构成了一个完整的工程案例，表明本文所提理论、模型与方法的正确性、可行性与有效性，具有很好的工程实践指导价值。更多还原

【Abstract】 With the gradual transformation of equipment maintenance support modes and theever increasing maturity of fault prognostics technologies, equipment healthmanagement (EHM) is bound to become an important part in the design, production andusage of future equipments. On the one hand, EHM performance relies on informationprocessing and decision making, and is more dependent on information acquisition onthe other hand. With the increase of equipment complexity, it is necessary to taketestability into account according to EHM requirements at equipment design stage, e.g.,how to select comprehensive test items, configure rational sensors, make scientific testtiming and adopt the corresponding technologies to realize the testability design. Thecurrent testability theory and technology, which are mainly for condition monitoringand fault diagnostics, do not consider the requirements of fault size, fault evolution andhealth evaluation for testability. How to develop testability design concurrentlyaccording to EHM requirements at the early design stage is a fundamental way toimprove EHM performance and further improve maintenance decision ability, and isalso a problem urgently to be solved during the development of equipments.In view of the fact that after considering EHM requirements in testability, theconnotations and architecture are not yet clear and the key technologies still await to bebreakthrough, this dissertation conducts in-depth studies on testability index, testabilitymodel and testability optimization design. The main content and outcomes are listed asfollows:1. Based on the current testability theory and framework and according to thefunctional requirements of EHM, the connotation of testability for EHM is defined andthe technology architecture of testability for EHM is proposed.2. To address the problems that the current testability indices are mainly used toevaluate fault detectability level and fault isolability level and are unable to describetestability level for EHM comprehensively, based on the qualitative intrinsicrequirements analysis of EHM for testability, testability indices architecture for EHM isconstructed from the aspects of “accuracy” and “timelines”, and the relationshipsbetween the indices are further analyzed.3. The modeling requirements after considering EHM functions in testability areanalyzed systematically and a quantified uncertainty hierarchical model (QUHM) isconstructed. At the system level, fault-test dependency is modeled through quantifieddirected graph and functional fault analysis; meanwhile fault attributes, test attributesand fault propagation attributes are assigned to the nodes and directed edges in the formof probability, fuzziness and uncertainty. At the component level, fault evolution-testdependency is obtained by physics-of-failure models or extended failure modes, mechanisms and effects analysis. The QUHM can be represented by a multipledependency matrix, based on which testability analysis and evaluation for EHM can berealized.4. The requirements of test item for EHM are analyzed, and a general process fortest optimization selection and sensor optimization configuration is proposed. Based onthe process, test preliminary selection rules and approaches are presented firstly. Then, atest selection and optimization model for EHM is formulated and a Boolean logic-basedoptimization method is also designed. Finally, fault detection uncertainty is analyzedsystematically from fault attributes, sensor attributes and fault-sensor matchingattributes. Based on the framework and fault detection uncertainty analysis, a sensoroptimization configuration model is formulated, which takes total sensor cost asoptimization object and the testability indices under uncertain detection as constraintconditions. Due to NP-hard property of the model, a genetic algorithm is designed toobtain the optimal sensor configuration.5. Based on the connotations and technology architecture of testability for EHM,the necessity and new requirements of test timing optimization are analyzed, and aMarkov-based test timing optimization method is studied. Firstly, for the storageequipments, a periodic test timing optimization approach based on Markov renewalprocess is presented. Further, for the usage equipments, the requirements of EHM fordynamic sequential test (DST) are analyzed. According to the characteristics of DST,the partially observable semi-Markov decision processes (POSMDP) is formulated.Then, POSMDP is converted to be completely observable belief semi-Markov decisionprocesses, based on which a dynamic sequential test timing optimization model isconstructed. The goal is to minimize the long-run expected average cost per unit time.The proposed method, which decides the next test timing based on equipment healthstate and considers the health evaluation uncertainty, is more suitable for the functionalrequirements of EHM. The DST is also applicable to storage equipments.A typical electromechanical servo system is taken as an example to verify andvalidate the proposed models and methods in the corresponding sections. The resultsshow that the testability indexes, testability model and testability design methods arefeasible, reasonable and available, and are of great engineering significance.更多还原