【作者】 金洋；

【导师】 徐敏强； 王日新；

【作者基本信息】 哈尔滨工业大学 ， 一般力学与力学基础， 2013， 博士

【摘要】 故障诊断技术是在轨卫星自主控制与管理的重要组成部分，它为卫星自主控制提供必要的支持，并为其在轨安全、可靠地运行提供保障。本论文以在轨卫星故障诊断技术为背景，根据卫星自主控制对故障诊断系统的要求以及当前故障诊断技术存在的问题，选择传递系统模型为主要研究对象，以第一原理诊断理论为诊断算法的基础，对基于该模型的诊断理论与应用技术进行了深入的研究。主要的研究内容及成果如下：针对在轨卫星自主诊断系统对算法实时性的要求，提出基于传递系统模型的诊断框架。通过在诊断框架中引入分离策略，减小诊断算法的在线计算量，提高算法的实时性；同时，对传统诊断理论中诊断与冲突的概念进行扩展，并提出状态记忆机制，通过利用历史的诊断结果，使得诊断系统具有记忆性，避免冗余的计算量。将诊断框架应用到某卫星一次电源系统上，通过与传统方法进行对比，验证提出的方法能够有效地提高诊断效率。为了提高诊断结果的准确性，对传递系统模型的结构和能够描述的知识类型进行研究。通过在原有的系统模型中增加表示因果关系的诊断知识，并采用分层结构对诊断知识进行组织，提出分层传递系统模型及相应的故障诊断方法。在诊断过程中，利用表示因果关系的诊断知识对推理过程进行指导，缩小状态匹配空间，在提高诊断模型中知识表示完备性的同时，提高诊断效率以及诊断结果的准确性和分辨率。将该诊断方法应用到某卫星测控系统上，验证该方法在诊断结果的准确性和效率上优于传统的诊断方法。针对传统的传递系统模型诊断方法无法处理由于状态不完全匹配而导致的诊断结果存在不确定性的问题，定义能够衡量两个状态之间接近程度的距离函数，对状态之间的相似度进行评价。同时，基于此距离函数提出模糊传递系统模型及相应的诊断方法，通过计算观测与目标状态之间的距离，对当前状态转移的趋势进行评估。以某卫星一次电源系统的典型故障模式为例，验证该诊断方法能够解决诊断过程中的不完全匹配问题，提高诊断结果的准确性。通过研究已有的可诊断性分析方法，针对其只根据系统的结构进行静态分析，导致对传递系统模型的分析结果不准确的问题，提出适用于传递系统模型的可诊断性分析方法。该方法保留系统的静态分析过程，并考虑系统运行的动态过程，从结构和功能两方面对传递系统模型的可诊断性进行讨论。利用提出的分析方法确认满足可诊断性的部件，在可检测性方面既保证其故障效果是可测量的，又保证故障效果不会被其它部件的功能结果所淹没。同时，在可分离性方面也可以保证各个故障模式之间的故障效果不会发生混淆。将某卫星一次电源系统作为实例，对其典型故障模式的可诊断性进行讨论，说明该分析方法的有效性。更多还原

【Abstract】 Fault diagnosis technology is an important part of the spacecraft autonomouscontrol and management system, it may provide the necessary support for thespacecraft autonomous control, and the security for its safe and reliable on-boardoperation. In this thesis, based on the background of spacecraft fault diagnosistechnique, the diagnosis theory and its applications which focus on the transitionsystem model and the first principle diagnosis method were conducted deeply inorder to satisfy the demand for the fault diagnosis system from the spacecraftautonomous control and the existing problems. The main research contents andresults are as follows:According to the real-time requirements for the algorithm from the spacecraftautonomous diagnosis technique, the separation strategy of conflict recognition iscarried out and the calculation method of analytical redundancy is put forward, thatmakes the main calculation of the conflict recognition to be completed in theoff-line phase. Meanwhile, the state memory mechanism which introduces thehistorical diagnosis result is researched in the candidate generation process in orderto avoid unnecessary computation. Applying the diagnosis framework to a primaryelectrical power system, and compared with traditional methods, it is verified thatthe proposed method can effectively improve the efficiency of diagnosis.In order to improve the accuracy of the diagnosis system, the structure of thetransition system model and types of the diagnostic knowledge are studied. Byintroducing the causality diagnosis knowledge into the model and using thehierarchical structure to organize the diagnosis knowledge, the hierarchicaltransition system model and the corresponding diagnosis method are proposed.When diagnosing through this method，the causality diagnosis knowledge is used toimprove the the completeness of the model and guide the reason process in order toreduce the state matching space, and it will cause the improvement of the efficiencyof diagnosis, the accuracy and the resolution of the diagnosis result. Applying thismethod to the telemetry and command subsystem of a satellite, the accuracy of thediagnosis and efficiency of diagnosis are verified better than those of the traditionaldiagnosis method.Aiming at the fuzzy problem of the diagnosis result due to the fact that thetraditional diagnosis method based on transition system model cannot match theobservation and the target state completely, the distance function based on theviewpoint of energy is defined to evaluate the similarity between the two states.Based on the distance function， the fuzzy transtion system model and the corresponding diagnosis method are proposed, and the distance function is used tocalculate the distances between the observation and each target state in order toevaluate the current trend of state transition. Using the typical failure modes of theprimary electrical power system as example, the effectiveness for solving theproblem of incomplete matching and the improvement of the accuracy of thediagnosis result are verified.Through the research of the existing analysis method for the diagnosability ofmodel, aiming at the problem that these methods are only limited to the staticanalysis based on the sensor layout and structure of the model which causes that theanalysis for the transition system model is accurate, a diagnosability analysismethod for the transition system model is proposed. This method carries on thediscussion about diagnosability from the structure and the function of model,preserves the static analysis process and the dynamic process of system operation isalse considered. Using this method, if a component is confirmed that satisfies thediagnosability, that means, the failure effect can be detectable, and withoutannihilation producted by the functional outcome of other components. In addition,it can also ensure that the failure effect among the various failure modes cannot beconfused for the function. This method is applied on the primary electrical powersystem, and its effectiveness can be verified.更多还原