【作者】 张媛；

【导师】 秦勇；

【作者基本信息】 北京交通大学 ， 安全技术及工程， 2014， 博士

【摘要】 安全是轨道交通永恒的主题,尤其是在轨道交通事业集中建设和高速发展的全盛时期,安全问题更是全社会关注的焦点。轨道交通列车的正常服役是保障轨道交通系统安全高效运营的必要条件,而轨道交通列车能否正常运行直接取决于其中的运行安全关键设备的服役状态。但是,我国现有的轨道交通列车及其运行安全关键装备在线服役状态的辨识、预测、诊断和控制技术和手段远远不能满足轨道交通系统动态化系统化的主动安全保障需求。针对这一重大问题,急需研究并提出系统化的列车运行安全关键设备服役状态辨识和预测的方法。鉴于此,本文在进行了如下研究工作：1.在了解分析列车关键设备状态监测、安全域估计理论和方法等相关的国内外研究现状的基础上,参考借鉴相关领域的已有成果,系统地提出了基于安全域理论的列车运行安全关键设备服役状态辨识方法。阐述了安全域的基本概念和内涵,说明了基于安全域估计进行状态辨识的基本原理；提出了基于安全域估计理论的状态辨识方法框架,给出了方法实施的通用步骤,并针对方法实现中的边界估计这一关键技术问题,根据研究对象是否便于建立数据模型提出了两条并行的技术路线：基于模型的边界估计技术和数据驱动的边界估计技术；对于本文使用的数据驱动的边界估计技术实现,提出了采用支持向量机的方法,并根据状态辨识需求给出了二分类的和多分类的支持向量机算法。2.在基于安全域估计理论的状态辨识方法框架的基础上,提出了一种面向实时状态特征的安全域状态辨识方法。该方法在状态特征提取方面,首先采用了较新颖的局部均值分解方法将数字信号分解为多个分量,然后计算了信号分量的直接时域特征以及基于能量和熵的两类实时状态特征指标；以列车滚动轴承作为实例,分别利用不同工况环境下的数据对算法的辨识精度、鲁棒性和实时性进行了全面测试,实验结果表明基于实时状态特征的状态辨识方法的计算效率很高,而辨识精度和抗干扰性能方面表现一般。3.从基于数据的统计分布特性提取状态特征方面考虑,提出了面向统计状态特征的安全域状态辨识方法。首先清晰地给出了基于统计状态特征提取和支持向量机的状态辨识方法,详细叙述了其实现步骤；然后,针对方法步骤中的统计状态特征提取问题,细致地阐述了基于主成分分析的统计状态特征提取方法；仍以列车滚动轴承作为实例,通过不同工况环境下的多组实验仿真,测试了方法的辨识精度、鲁棒性和实时性,实验结果表明基于统计状态特征的状态辨识方法具有极高的辨识精度和优越的鲁棒性,但该方法计算负担大执行效率低,实时性方面表现一般；最后简要介绍了作者参与的国家863重点项目中关于安全域状态辨识方法的现场应用系统的设计工作。4.基于列车运行安全关键设备服役状态辨识方法的研究结果,进一步对基于状态监测的剩余寿命预测问题进行了探讨。叙述了几类常用的基于状态监测的剩余寿命预测方法,给出了各类方法的基本原理、优缺点和适应环境；详细讨论了能够同时融合可靠性信息和状态监测信息的比例风险模型,对基于比例风险模型的剩余寿命预测方法进行了详细阐述；基于滚动轴承的全寿命振动数据,进行了试验仿真,结果表明,相对于仅依靠状态监测信息的剩余寿命预测方法来说,本文提出的基于统计状态特征的比例风险模型能够精确地预测设备的剩余寿命。更多还原

【Abstract】 :With the rapid development of rail traffic construction, safety issues have always been a focus of public attention. Rail vehicles, whose quantity has substantial increased recently, are not only primary tools and direct carriers for the rail transportation, but also the key to ensuring its safety and efficiency. Far more important is that the safety status of the whole rail vehicle depends crucially on its key equipments. However, for the online status of the rail vehicles’equipments, the existing technologies and methods of the identification, prediction, diagnosis and control are far from enough to meet requirements of the dynamic, systematic and active safeguard. To address this important issue, the service state identification and prediction methods have been studied as follows for the key equipments in rail vehicles operation safety.1. Based on understanding of research status about equipment condition monitoring and safety region estimation, and existing achievements in related fields were referenced. First, for the key experiments in rail vehicles, the identification method of service status based on Safety Region Estimation (SRE) was proposed. The basic concept and content of the safety region were elaborated, and the basic principle of the above identification method was explained. Second, the methodological framework of the SRE state identification method was presented, and the universal process was given. And in order to solve the key problem of boundary estimation, two parallel technical routes were described:for the objects of easy-to-modeling, model-based approach was put forward; and for the objects of hard-to-modeling, data-driven approach was developed. Third, for the latter approach, it can be achieved using Support Vector Machine (SVM). So two-class SVM and multi-class SVM were both described in detail to meet the demand of the two-state identification and multi-states identification, respectively.2. On the basis of methodological framework of SRE state identification, a specific state identification method with real time state feature was presented. First, to extract the real-time state feature, signals were decomposed into multiple components by Local Mean Decomposition (LMD) algorithm, and then several features of each component were computed which include time domain features, energy feature and entropy feature. Second, in order to verify the validity of the method, rolling bearings of rail vehicles were used as an example. Identification accuracy, robustness and real-time performance of this method were thoroughly tested using vibration data under two different conditions. The experimental results show that the state identification method with real time state feature has high computational efficiency, but general accuracy and robustness.3. Considering the statistical distribution feature extractions form a lot of state data, a SRE state identification method with statistical state feature was proposed. First, to compute statistical state feature, a new algorithm based on Principal Component Analysis (PCA) was given. And by the algorithm, the two statistics control limits of T2and Squared Prediction Error (SPE) were extracted as state characteristics. And then, rolling bearings were still used as a practical example, method performance testing was finished with two types of data under laboratory environment and simulate actual working condition, as the previous method based on real time state feature. The testing results indicated that the identification accuracy and the robustness are both satisfying, but the computational efficiency is relatively low because of large computational burden. Finally, some system design work and achievements for practical applications, which relying on the author participated National High-tech R&D Program of China (863Program), were briefly introduced.4. With the above research results of service status identification method, Residual Useful Life (RUL) prediction based on condition monitoring was discussed. Firstly, several types of commonly RUL prediction methods were introduced, including the basic principles, advantages and disadvantages of each type of method. Secondly, Proportional Hazard Model (PHM) by the fusion of reliability information and online monitoring information was discussed in detail. Thirdly, the RUL prediction method based on PHM and statistical state feature was put forward, and the method testing was completed by using of rolling bearings’ life cycle vibration data. The testing results and the comparison with other RUL prediction approach indicate the validity and superiority of the proposed method.更多还原