【作者】 宋其江；

【导师】 徐敏强；

【作者基本信息】 哈尔滨工业大学 ， 一般力学与力学基础， 2010， 博士

【摘要】 故障诊断技术是航天器安全、可靠运行的保障。本文研究了基于有向图模型的航天器故障诊断方法。该方法能够描述复杂系统的故障传播机理以及系统元素间的因果关系,是一种基于深层知识的诊断方法,而且缓解了诊断知识获取瓶颈问题,能够对某些未预知的故障进行识别。因此该方法为航天器故障诊断技术的研究提供了一种新的途径。本论文的主要目的是通过对基于有向图模型的诊断方法的研究,找到适用于航天器在轨自主诊断的方法。为了这样的目的,本论文主要进行如下几方面的研究工作。针对航天器故障诊断领域的特点,本文将故障传播有向图和符号有向图(SDG)两种系统描述方法的优点结合起来,提出了一个基于分层有向图的定性诊断模型。它具有SDG模型固有的良好完备性,又能够清楚的反映部件间的连接作用关系,同时对该模型采用了有向图分层策略,并利用测试节点间的定性关系来回溯搜索不相容支路找出故障源候选集合,而且可以根据故障可能性对候选故障源进行的排序。最后将该方法应用到卫星一次电源系统,验证了该方法的有效性。航天器的故障样本和诊断经验的不足,给建模带来了困难,因此,本文采用了根据系统正常的工作模式进行建模的思想,建立系统诊断模型。另外,航天器的许多系统具有多种工作模式,系统关键变量间的定性关系随着工作模式的改变而变化。由于传统SDG模型对这种变化的定性关系的表达还不够完备,采用这种模型进行故障诊断,将会导致诊断分辨率低的问题出现。因此,本文将模糊理论引入到SDG模型中来,提出了一种基于SDG模型和模糊理论相结合的模糊SDG模型。节点变量变为模糊变量,它的每一个模糊子集代表了一种工作模式下的状态;节点间的定性关系通过模糊关系矩阵来表达。通过回溯搜索模糊不相容支路来找出故障源候选集合。实例分析表明了该方法的有效性。对于前面提出的基于模糊SDG的方法,如果模型中存在不可测量节点,该诊断系统将无法进行诊断推理。为了解决该问题,提出了基于模糊SDG模型和贝叶斯推理相结合的模糊-概率SDG模型。该模型节点间的定性关系用条件概率表(CPT)来表达,通过贝叶斯的不确定性推理和不相容支路的判断,可以在不可测节点存在的情况下,完成不确定性推理,找出故障源候选集合。最后建立了某卫星一次电源系统的诊断模型,故障诊断的仿真结果验证了该方法的有效性。故障诊断的准确性依赖于系统模型的精确性,为了提高诊断模型的精确性,在以往建模方法的基础上,本文提出了利用定性趋势分析(QTA)技术来辅助建模的方法。趋势分析能够为SDG模型提供充足的定性趋势信息,通过分析这些信息,可以找出节点变量的工作阈值区间,还有节点变量间的多值逻辑关系。最后,本文提出了结合了其它建模方法的综合建模策略,并针对前面提出的模糊-概率SDG模型给出了建模的步骤。将上面提出的方法应用到某卫星一次电源系统中,建立了更为精确的诊断模型,诊断的仿真结果表明该辅助建模方法提高了诊断的准确性。故障检测的核心问题是传感器的分布问题。针对有向图模型,分别给出了基于可观测性和可靠性的传感器设计方案,并进行了实例分析。最后,通过综合分析前面的研究,提出了新的可靠性设计思想,并对可靠性问题进行了形式化描述,接着给出了基于可观测性和可靠性的传感器分布设计方案。该方案综合考虑了可观测性、可靠性等要求,采用贪婪启发式算法实现了该方案。仿真结果表明该方案满足了可观测性和可靠性的要求,能够快速提高系统的可靠性,更适合系统设计的需要。更多还原

【Abstract】 Fault diagnosis technique is very important to ensure the safety and reliability of spacecraft. This dissertation studies the spacecraft fault diagnosis based on the digraph model. The approach is a qualitative model-based approach, the model qualitatively describes the characteristic of system structure, and can express cause-effect relation among process variables and propagation relation among the faults, and the model is a deep knowledge-based model, and can solve the bottleneck problem in KA and identify some unknown fault, so the approach provide a new solutions for fault diagnosis of spacecraft. The main purposes of this dissertation are to make a systematic study on digraph-based fault diagnosis technique to realize the on-board autonomous fault diagnosis. For the purpose, the main work is as follows:According to the characteristic in fault diagnosis domain of spacecraft, this dissertation proposes a novel qualitative diagnosis approach based on hierarchical digraph integrating the goodness of the fault propagation graph and SDG(signed directed graph), the model has the better completeness and explanation facility of SDG, and can express the connection relation among all components. The hierarchical strategy was adopted for digraph so that it reduces search space for failure sources; the fault propagation consistent branches was backward searched by qualitative relations between measured nodes ,and set of failure source candidates were found, the furthermore, the failure source candidates are ranked according to the fault possibility rate. The diagnosis system for primary power system of some satellite was modeled with the approach, and diagnosis simulation result show the approach is valid.In the aerospace field,because the fault experience and fault data is insufficient, this has brought a lot of difficulties for the fault diagnosis modeling, so the idea that modeling by the data in all normal work modes of system is adapted. In addition, due to the work threshold and relation between processes variables vary with the work state, the tradition SDG model is insufficient expression of the qualitative relation, and has lower resolution of diagnosis. so the fuzzy theory introduced into SDG model, the node variable was expressed as fuzzy variable , each fuzzy subset of which describes the state in each working mode , the cause-effect relationship between the nodes was described by fuzzy relation matrix. The algorithm is presented to identify fuzzy consistent branch by fuzzy reasoning and to search the set of failure source candidates. The validity of approach was identified by the application example.Under the condition of the unmeasured node, the approach based on the fuzzy SDG model will be not available, In order to solve the problem, the fuzzy probabilistic SDG is proposed based on model of fuzzy SDG and Bayesian inference, the cause-effect relationship between the nodes was described by conditional probabilities table (CPT), the set of failure source candidates is found out by Bayesian inference and backtracking algorithm under the condition of the unmeasured node. The primary electrical power supply system in certain a satellite was modeled with the proposed approach, the diagnosis simulation result show the approach is valid.The accuracy of diagnosis depends on precision of the system model. To improve the precision of system model, on the basis of the previous method for modeling, the novel modeling method was proposed that the technology of QTA (qualitative trend analysis) can aid in modeling. The qualitative trend information by QTA can transform the valid information for SDG model, the thresholds and dynamic qualitative relation of node variable are determined by analyzing the information, furthermore, the synthetical modeling method combined with other modeling methods is summarized. At last the primary electrical power supply system in certain a satellite was modeled, the diagnosis simulation result show the modeling method improve the precision of the model and the accuracy of the diagnosis .The core problem of fault detection is the sensor location. According to the digraph model, the design scheme of sensor location based on observability, reliability was respectively given. The applications and analysises based on the schemes were presented. Through the above research, the new design idea of sensor location based on reliability was proposed, and the reliability problem was described again based on the idea, furthermore, the optimal design scheme for location of detection sensors was presented. this scheme takes into account both the fault observability, the reliability of fault detection, and sensor placement costs. The greedy heuristics algorithm was designed for the scheme. Furthermore, the proposed approach was applied to primary electrical power supply system in certain a satellite, and the diagnosis simulation results show that the design scheme has met the design requirements of the observability and the reliability, and can rapidly enhancement reliability and is more suitable for system needs.更多还原