新型大气数据传感系统故障自诊断关键技术研究

发布时间：2018-07-13 18:48

【摘要】：新型大气数据传感系统是一种不但可同时测量飞行器的飞行高度、速度、攻角和侧滑角等多种飞行参数,而且可进行自身状态在线自确认的大气数据系统。该系统充分继承了嵌入式大气数据传感技术和自确认传感技术的优势,适用于现代飞行器的高隐身、高机动性和高可靠性需求。本课题旨在研究故障检测、故障定位及故障诊断等状态自确认方法,解决新型大气数据传感系统的若干关键技术问题。论文的主要研究内容如下:(1)针对新型大气数据传感系统的故障传播问题,研究一种基于模糊概率Petri网的故障传播分析方法。利用模糊概率Petri网的强大建模和逻辑推理性能,分析系统的最大概率故障传播路径,分别建立系统组件级和系统级的故障传播规律模型,获取可充分覆盖测试样本集的主要故障模式。试验结果表明,压力传感器异常、信号采集及处理电路异常以及测压孔堵塞是大气数据系统的主要故障,与专家知识及工程经验得出的结论一致。(2)针对新型大气数据传感系统的故障检测及故障源定位问题,研究一种基于小波核主元分析和故障指示向量的多故障检测及识别方法。利用核主元分析方法分析多路测压通道间的内在关系,研究待测样本在高维特征残差空间内投影量的变化与故障检测的关系,验证小波核的多分辨率分析能力在瞬时性故障检测中的优势;根据测压点布局的冗余特性,研究攻角和侧滑角参数分别与垂向和纵向测压点的内在关系,建立故障指示向量知识库表征测压通道状态,验证系统在低马赫数小攻角和高马赫数大攻角情形下,通过故障指示向量匹配实现故障源定位的有效性。实验结果表明,该方法可实现多故障无遗漏检测,总故障数小于3的典型故障检测率大于90%,故障源定位率为100%。(3)针对新型大气数据传感系统的非线性故障特征提取和多故障分类问题,研究一种基于集合经验模态分解和多分类相关向量机的故障诊断方法。利用集合经验模态分解的信号自适应分解特性,分析不同类型故障输出信号在不同本征模分量上的能量特征差异性,建立不同类型故障特征向量集,验证集合经验模态分解的抗模态混叠和故障特征提取性能;利用多分类相关向量机的小样本学习、分类结果概率形式输出、单模型多分类等特性,分析故障诊断与分类结果不确定性的关系,研究基于交叉验证的最优核参数选取方法,建立不同故障模式的多分类器模型,验证多分类相关向量机的多故障类型同时识别优势。与传统经验模态分析方法相比,该方法具有明显的抗模态混叠优势,对系统正常工作、压力波动大、压力跳变、压力偏置和压力恒值输出等样本识别为对应故障类型的平均分类概率分别大于86%和80%,故障分类正确率为100%。(4)为验证研究的新型大气数据系统故障检测、故障识别及故障诊断方法的有效性,设计一种新型大气数据系统仿真试验平台,模拟产生各种真实故障,对系统分布式压力传感测量进行标定和测试,获取正常测试样本和故障仿真数据样本集。
[Abstract]:The new atmospheric data sensing system is a kind of flight parameters which can not only measure flight height, speed, angle of attack and sideslip angle simultaneously, but also can carry out the self confirmed air data system on line self state. This system fully inherits the advantages of embedded atmospheric data sensing technology and self recognition sensing technology. The high stealth, high mobility and high reliability of the generation of aircraft is required. This topic aims to study the state self validation methods such as fault detection, fault location and fault diagnosis to solve some key technical problems of the new atmospheric data sensing system. The main contents of this paper are as follows: (1) fault propagation for the new atmospheric data sensing system A fault propagation analysis method based on fuzzy probability Petri net is studied. Using the powerful modeling and logic reasoning performance of fuzzy probability Petri net, the maximum probability fault propagation path of the system is analyzed. The model of system component level and system level fault propagation law is established respectively, and the main reasons that can fully cover the test sample set are obtained. The test results show that the abnormal pressure sensor, the abnormal signal acquisition and processing circuit and the blockage of the pressure measurement hole are the main faults of the atmospheric data system, which are consistent with the conclusions obtained by the expert knowledge and engineering experience. (2) a new kind of fault detection and fault location based on the new atmospheric data sensing system is studied. Nuclear principal component analysis and fault indicator vector multiple fault detection and recognition method. Using kernel principal component analysis method to analyze the internal relationship between multi-channel pressure measurement channels and study the relationship between the change of the projection quantity of the sample in the high dimensional feature residual space and the fault detection, and verify the multi-resolution analysis ability of the wavelet kernel in the instantaneous fault detection. In accordance with the redundancy characteristics of the pressure point layout, the internal relationship between the angle of attack and the sideslip angle and the vertical and longitudinal pressure measurement points are studied. The fault indicator vector knowledge base is established to represent the state of the pressure measurement channel. The fault indicator vector matching is used to verify the failure of the system in the case of low Maher number and high Maher number of attack angle. The experimental results show that the method can achieve multiple failure detection, the total fault number is less than 3, the typical fault detection rate is more than 90%, the location rate of the fault source is 100%. (3) for the nonlinear fault feature extraction and multi fault classification problem of the new atmospheric data sensing system, and a research based on the set of empirical mode decomposition is studied. The fault diagnosis method of the multi classification correlation vector machine. Using the adaptive signal decomposition characteristic of the set empirical mode decomposition, the difference of energy characteristics of different types of fault output signals on different eigenmode components is analyzed, and different types of fault feature vectors are set up to verify the anti modal aliasing and fault characteristics of the integrated empirical mode decomposition. By using the small sample learning of the multi classification correlation vector machine, the probability form output of the classification result and the single model and multi classification, the relationship between the fault diagnosis and the uncertainty of the classification results is analyzed. The optimal kernel parameter selection method based on the cross validation is studied, and the multi classifier model is set up to verify the multi classification phase. Compared with the traditional empirical mode analysis, the method has obvious advantages of anti modal aliasing. The average classification probability of the sample identification for the corresponding fault types, such as the normal working of the system, the pressure fluctuation, the pressure jump, the pressure bias and the pressure constant output, is greater than 86% and 80%, respectively. The correct rate of fault classification is 100%. (4) to verify the effectiveness of the new atmospheric data system fault detection, fault identification and fault diagnosis method. A new atmospheric data system simulation test platform is designed to simulate various real faults and to calibrate and test the distributed pressure sensing measurement of the system, and to obtain the normal test sample. This and fault simulation data sample set.
【学位授予单位】：北京理工大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TP79

【参考文献】