【作者】 李俭川；

【导师】 温熙森；

【作者基本信息】 中国人民解放军国防科学技术大学 ， 机械电子工程， 2002， 博士

【摘要】 故障诊断与维修决策是实现装备快速故障诊断与维修的关键技术之一，对于提高装备的战备完好率、二次出动率和战斗力再生，保证任务的成功，降低装备的维护和保障费用具有重要作用。但是，由于复杂设备中存在很多错综复杂、关联耦合的相互关系，并存在大量的不确定因素及不确定信息，使得故障诊断与维修决策的实施较为困难。因此，寻求能对复杂设备故障诊断与维修相关的各种信息进行快速融合，并有效处理不确定性知识的决策模型及方法，一直是研究者们不懈努力的方向。 贝叶斯网络是目前不确定知识表达和推理领域最有效的理论模型之一，适用于不确定性和概率性的知识表达和推理，特别适用于有条件地依赖多种控制因素的决策。贝叶斯网络是一种基于网络结构的有向图解描述，具有多源信息一致表达与信息融合能力，能进行双向并行推理，并能综合先验信息和样本信息，使推理结果更为准确可信。因此，贝叶斯网络在故障诊断领域中的应用具有重要意义。 本文以“十五”国防预研项目“基于信息融合的装备快速故障诊断技术”为背景，以某型直升机机载设备为应用研究对象，将贝叶斯网络作为设备故障诊断与维修决策模型，研究复杂情况下以低代价、快速度为诊断决策目标的故障诊断与维修决策方法。论文首先阐述了贝叶斯网络的理论基础，然后针对贝叶斯网络故障诊断与维修决策方法存在的主要问题，提出了基于故障假设—观测—维修操作节点结构的诊断贝叶斯网络模型，并对模型的知识表达方法、模型建造方法和基于诊断贝叶斯网络模型的故障诊断与维修决策算法开展了深入研究，给出了基于诊断贝叶斯网络模型的故障诊断与维修决策系统设计和实现方法，主要内容包括： 1．在分析复杂设备故障诊断与维修决策面临的主要问题、总结现有故障诊断与维修决策模型与方法存在的主要局限的基础上，深入探讨了贝叶斯网络故障诊断与维修决策方法的优势及其存在的主要问题。 2．阐述了贝叶斯网络的概率理论基础，分析了贝叶斯网络推理和贝叶斯网络学习等一般问题。以贝叶斯网络为基础，结合故障诊断与维修决策的需求和存在的问题，提出了基于故障假设—观测—维修操作节点结构的诊断贝叶斯网络模型，并给出了该模型的数学描述与知识构成要素。 3．针对复杂设备诊断知识表达存在的困难，引入面向对象的知识表达方法，建立了面向对象的诊断贝叶斯网络的知识表达体系，提出了基于关系数据库的知识存储和 国防科学技术大学研究生院学位论文提取方法，为复杂设备诊断贝叶斯网络模型建造建立了一种有效的知识表达方法。 4.针对复杂设备诊断贝叶斯网络模型建造与推理中存在的主要问题与困难，提出了基于自上而下思想的诊断贝叶斯网络分级建造方法，给出了一种分级层次诊断贝叶斯网络模型，为复杂设备诊断贝叶斯网络模型的建造提供了系统的指导原则;提出了由设备功能模型建造诊断贝叶斯网络模型的方法和由故障树模型建造诊断贝叶斯网络模型的方法，可以便利地将已有的决策模型直接转化为诊断贝叶斯网络模型。 5.针对目前诊断决策方法还不能解决实际诊断决策过程中的操作依赖关系和大规模模型的推理效率问题，提出了一种基于诊断贝叶斯网络的故障诊断与维修决策算法。通过引入附加操作节点，实现了诊断操作之间存在代价依赖关系时的诊断决策算法;通过引入虚拟维修操作节点，给出了一种分级层次诊断贝叶斯网络模型的诊断决策算法，实现了大规模诊断贝叶斯网络模型的诊断决策。分析表明，该算法能有效降低诊断代价，对于代价值和条件概率值的偏差有较强的适应能力，能较好解决过程祸合、代价关联、操作依赖的复杂故障诊断与维修决策问题。 6.提出了一种基于诊断贝叶斯网络的故障诊断与维修决策系统的实现结构，以某型直升机机载设备故障诊断与维修为背景，给出了基于诊断贝叶斯网络的故障诊断与维修决策系统的设计与实现方法，并以某型变流器(机载设备之一)为对象，阐述其诊断贝叶斯网络模型的建造过程和维修应用实践。结果表明，该系统能有效提高设备的故障诊断与维修决策能力，缩短维修周期。 总之，本文深入研究了贝叶斯网络故障诊断与维修决策方法，提出了基于故障假设一观测一维修操作节点结构的诊断贝叶斯网络模型，提出了诊断贝叶斯网络的面向对象知识表达方法、诊断贝叶斯网络分级建造方法、基于设备功能模型和故障树模型建造诊断贝叶斯网络的方法、基于诊断贝叶斯网络的故障诊断与维修决策算法，形成了较为完备的基于诊断贝叶斯网络的故障诊断与维修决策系统，并在某型直升机机载设备故障诊断与维修中获得成功应用，为装备快速故障诊断与维修决策提供了一条切实可行和卓有成效的途径。更多还原

【Abstract】 Fault diagnosis and maintenance decision is one of the key techniques for weapons’ fast fault diagnosis and maintenance. It is very important for improving combat readiness rate and rebirth of battle effectiveness, and guaranteeing task success. It can considerably reduce the cost of maintenance and support. With so much anfractuous and coupling correlation, many uncertainty factors and much uncertainty information in complex devices, the fault diagnosis and maintenance decision is very difficult. The researchers in this field always try their best in investigating decision models and decision algorithms, which can quickly fuse various data related to device’s fault diagnosis, and effectively handle the uncertainty knowledge.Bayesian Network (BN) is one of the most effective theoretical models for uncertainty knowledge expression and reasoning. It can be applied to decision with various dependent factors. BN is a directed acyclic graph with network structure, which is intuitionistic and easy understanding. It can handle multi-information expression, data fusion and bi-directional parallel reasoning. The ability to colligate the prior information and the current information makes the inference much more accurate and believable. So it is a better choice to use BN for complex device’s fault diagnosis.Supported by the National Defense Preparatory Research Projects, Research on Weapon’s Fast Fault Diagnosis Technique Based on Data Fusion, this dissertation takes the BN as a diagnostic decision model, and investigates decision methods with the objective of fast and low-cost diagnosis under complicated situations. After introducing the theoretic basis of the BN, it points out several problems of BN fault diagnosis and maintenance decision methods, and proposes a diagnostic Bayesian Network (DBN) model based on the fault hypothesis-observation-maintenance operation nodes. It deeply investigates several key techniques including knowledge expression of DBN, model construction of DBN, and diagnostic decision algorithms based on DBN. Also it presents the designment and realization method of fault diagnosis and maintenance decision system based on DBN. The main points can be summarized as follows.1. The main difficulties faced by fault diagnosis of complex devices are analyzed, and the major limits of existed fault diagnosis decision models and methods are summarized. Then the fault diagnosis decision method based on DBN is put forward, and several advantages and main problems of this method are pointed out.2. After expatiating the probability theoretic basis, it analyzes some issues including BN inference, and BN learning. A DBN structure consisting of fault hypothesis nodes, observation nodes, and repair nodes is brought forward, which combines the general BN and the requirements of fault diagnosis and maintenance decision. The mathematic description and knowledge inscapes of DBN are set forth at last.3. There exist many difficulties for knowledge expression of complex devices’ fault diagnosis. Through importing object oriented knowledge expression method, it establishes object oriented diagnostic Bayesian Networks (OODBN) frame and its knowledge operating methods including storage and pick-up, which erects an effective knowledge expression method for complex devices’ DBN models.4. In order to handle the construction and reasoning problems, A hierarchy DBN model construction method is established, which is based on TOP-DOWN idea, and which provides a systemic principle for DBN construction of complex devices. Constructing DBN from function models and constructing DBN from fault tree models make it easier for engineers to translate already existed models into DBN models.5. Based on the summary of fault diagnosis decision methods in existence, which use decision theory and BN, fault diagnosis decision method with DBN models is provided. One fault diagnosis decision method under dependent cost is realized through the introduction of appended operation nodes. And one fault diagnosis decision method更多还原