【作者】 张筱磊；

【导师】 沈毅；

【作者基本信息】 哈尔滨工业大学 ， 控制科学与工程， 2013， 博士

【摘要】 执行未来深空探测任务的航天器，需要具备在无人工干预条件下自主进行状态监测与故障诊断的能力。本文从信息处理的角度，研究了航天器自主故障诊断中的信号估计、故障检测及诊断推理问题。此外，结合航天器典型组成结构，探讨了故障诊断系统的实现方式。针对单维测量信号，在只有传感器输出数据且无冗余条件下，研究了基于经验模态分解的信号估计方法。通过分析已有基于经验模态分解信号估计方法的不足，提出一种评价信号成分的能量指标，进而基于此指标提出了一种估计信号的模态分量区间能量方法。针对典型信号，通过不同参数设定条件下的大量仿真实验，验证了模态分量区间能量信号估计方法的有效性。针对单传感器测量数据的故障检测问题，研究了基于概率分布模型的故障检测方法。将本文提出的模态分量区间能量信号估计方法与贝叶斯推理相结合，基于高斯混合模型建立无故障测量信号的概率分布模型，并将其作为基准通过置信界检测突变与缓变故障。通过针对航天器电源系统的仿真实验，验证了故障检测方法的可行性。针对包含未知噪声的非线性动态系统的状态与输出估计问题，研究了基于Dirichlet过程混合模型的信号估计方法。在高斯分布假设条件下，针对包含未知噪声的非线性系统，提出一种在线估计系统状态与输出的DPM-EKF滤波器。针对包含未知噪声的非线性系统的输出估计仿真实验结果表明，DPM-EKF滤波器的估计结果明显优于EKF滤波器的估计结果。针对包含未知过程噪声的非线性系统的故障诊断问题，研究了基于Shiryayev序列概率比的故障检测与隔离方法。在故障量大小未知条件下，提出一种结合Shiryayev序列概率比与预测分布的在线故障检测与隔离方法。在本文提出的故障检测与隔离方法中，首先应用DPM-EKF滤波器，在线计算噪声未知条件下非线性系统的输出残差，进而对于未知的故障参数，基于预测分布迭代计算Shiryayev序列概率比。通过航天器姿态确定系统的传感器故障仿真实验，验证了故障检测与隔离方法的有效性。针对具有切换运行模态的混杂系统故障隔离问题，研究了基于有向因子图的故障隔离方法。对于带有切换行为的混杂系统，通过分析贝叶斯网络在故障因果性建模能力上的不足，提出了一种基于混合键合图及因果路径分析建立有向因子图诊断模型，并基于最大后验概率隔离故障的方法。在航天器电源系统的仿真实验中，根据混合键合图构建的有向因子图诊断模型，以及依据最大后验概率准则得到的诊断结果验证了故障隔离方法的有效性。针对以总线为中心的航天器电气连接结构，提出一种递阶化的故障诊断系统结构，其在电气连接上作为总线上的独立功能设备，在诊断方式上则对航天器不同功能层次应用不同的方法。依据航天器系统所包含的典型设备与组成方式，基于递阶化的故障诊断系统结构，设计并实现了包含故障诊断模块，部件模拟器以及外部终端的航天器仿真验证系统，测试结果验证了递阶化故障诊断系统结构的可行性。更多还原

【Abstract】 The scientific tasks performed in deep space require the spacecraft with thecapacity of self health monitoring and self fault diagnosis. From the informationprocessing point of view, the signal estimation, fault detection, fault isolation andinference problem are studied in this thesis. Besides, this thesis makes furtherdiscussion about the implementation of fault diagnosis system based on typicalspacecraft system configuration.For univariate measurement signal, based on empirical mode decompositionmethod, the signal estimation problem is studied in condition with non-redundancyoutput sensor data. Through analyzing the defects of existing signal estimation methodsbased on emprirical mode decomposition, an evaluation criteria based on signal energyand a signal estimation method based on zero crossovers of intrisic mode fuction areproposed. Two different synthetic signals are simulated under different settings, thesimulation results illustrate that the presented energy based signal estimation techniquecan improve signal-to-noise ratio of the synthetic signals obviously. Based on univariatemeasurement signal and probabilistic distribution model, the fault detection problem isstudied. The univariate measurement signal without fault is used to constructprobabilistic distribution model based on Gaussian mixture model, and the probabilisticdistribution model is applied to detection abrupt and incipient fault based on confidencebound. Bus voltage signal in spacecraft power system is simulated to validate thefeasibility of the proposed fault detection method.Based on Dirichlet process mixture model, the estimation problem of state andoutput with unknown noise interference in dynamic system is studied. To reduce theunknown noise interference and get accurate state and output estimation, the Dirichletprocess mixture model is adopted to model the noise. The DPM-EKF filter is proposedto estimate the state of nonlinear system with unknown noise. The simulation resultsshow that the signal-to-noise ratio of state estimation can be improved obviously.Based on Shiryayev sequential probability ratio, the fault diagnosis problem fornonlinear dynamic systems is studied. The online fault detection and isolation algorithmwith unknown fault magnitude is presented based on Shiryayev sequential probabilityratio and posterior predictive distribution. The fault diagnosis task for spacecraft attitudedetermine system can be accomplished through combining the output estimation and theproposed fault detection and isolation algorithm. Through simulation of abrupt fault inangular rate sensor and angular position sensor, the diagnositic results demonstrate thevalidity of the proposed approach. By using probabilistic inference in fault diagnosis, the probability of fault causecan be deduced from probabilistic graph model, the impact of uncertainty existed indiagnosis model can also be mitigated. The insufficiency of Bayesian network inmodeling the independence is analyzed through instances. The directed factor graph isproposed to serve as diagnosis model for representing the fault propagation relationswithin the spacecraft components. For spacecraft with switching operational mode, theapproach for setting up diagnosis model based on directed factor graph is proposedthrough using causal paths between basic elements of hybrid bond graph. In simulationexperiment, a diagnosis model based on directed factor graph is built up base on hybridbond graph of spacecraft power system, and the fault injection is performed in one of itsoperational modes. The effectiveness of the proposed method is supported by the resultof fault diagnosis based on maximum a posteriori criterion.The electrical connection patterns of on-board computer centric structure and buscentric structure are studied based on typical spacecraft configuration. For spacecraftwith bus centric connection pattern, fault diagnosis system with hierarchical structure isproposed. The fault diagnosis system acts as a separated device to be connected into thesystem bus, and carry out different fault diagnosis algorithm within different hierarchy.Based on math model and hierarchical structure, a spacecraft fault diagnosis simulationand verification system, which contains fault diagnosis emulator, actuator emulator,sensor emulator, etc., is designed and constructed through software and hardwaresimulation. The feasibility of the hierarchical structure of fault diagnosis system iscorroborated through hardware-in-the-loop simulation.更多还原