学科：载运工具运用工程,出版时间：2018,导师：林建辉指导,学位授予单位：西南交通大学,论文作者：庄哲著,
中文
随着我国经济发展以及城市化步伐的加快，人们对日常出行的便捷性、舒适性要求越来越高。以高速铁路及地铁为代表的轨道交通在我国经济社会发展中起着重要的地位和作用。加快轨道交通建设是提升人民生活水平、缓解城市交通压力、维持国家经济增长、调整相关产业结构的重要举措。我国高速列车及地铁城轨车辆绝大部分采用电力牵引形式，一般通过车辆顶部受电弓与接触网滑动接触作用，获取电能。作为承担能量传递任务的受电弓系统，一直是车辆大系统的关键部件之一，随着车辆运行速度的提升，受电弓系统在自身结构复杂、运行环境恶劣、受多因素耦合作用等不利因素影响下，机械、电气及风管部件故障时有发生，特别是机械部件，裂纹失效模式占比较大，而传统的基于图像识别、地面检测等方法的受电弓监测技术无法满足我国日益发展的车辆运营需求。因此，研究保障受电弓可靠运行的方法及技术，有着重要意义。 本文针对受电弓可靠、稳定运行条件下遇到的主要问题，从基于振动信号的数据驱动故障诊断技术入手，着重对受电弓机械部件裂纹失效展开研究，主要进行了如下工作： (1)针对传统集成经验模态分解(EEMD)算法中涉及的白噪声幅值系数及总体平均次数无法根据信号特征有效选取的问题，提出了一种EEMD自适应参数选取算法，该算法通过分析不同白噪声幅值系数对待分析信号的极值点分布均匀性的影响规律，自适应地选择最有效的可以优化极值点分布特性的白噪声幅值，通过对仿真信号的分析，证明该方法在不降低分解精度的前提下，可有效降低模态混叠效应，减少计算时间，提高分解效率。 (2)在改进EEMD分解算法的基础上，将分解后的本征模态函数(IMFs)分量进行相空间重构，根据各分量的信号特点，利用改进嵌入维度及延迟时间参数计算方法C-C算法选取相空间重构过程中的参数，利用自适应极大似然估计方法评估IMFs相空间重构信号的本征维数，最后使用局部切空间流形学习算法，将含噪信号在各频带空间的高维重构相空间内进行低维映射，提取低维空间内对应信号的有效成分，实现了信号与噪声的成功分离。与其他降噪方法对比结果显示，所提方法具有一定优势，可应用于受电弓振动信号的降噪处理。 (3)构建了基于改进EEMD分解的受电弓振动信号信息熵测度模型，定义了改进EEMD能量熵、改进EEMD奇异熵、改进EEMD排列熵、改进EEMD近似熵、改进EEMD样本熵、改进EEMD模糊熵等六种时频分解信息熵特征，根据受电弓振动信号特点，对信息熵计算过程中涉及的多种参数展开优化选取，提出了受电弓振动信号信息熵特征提取模型，通过对将获取的故障特征输入基于粒子群参数优化的支持向量机(PSO-SVM)进行受电弓故障识别分析，探究针对受电弓故障最为敏感的测点及故障特征提取类型，验证了现代时频分析算法与信息熵联合的诊断方法在受电弓故障振动信号特征提取中的可行性与有效性。 (4)通过对基于改进EEMD熵特征算法的提取识别，受电弓顶管振动信号对受电弓支架裂纹等故障形式较为敏感，但受电弓碳滑板振动信号的相关熵特征对于受电弓故障的分离度较差，不能满足故障诊断的实际需要。因此针对碳滑板振动信号特点，开展二代小波分解及信息熵特征提取方法研究，大幅提升碳滑板垂向振动数据故障特征识别率，证明了该方法关于碳滑板振动数据故障特征提取的有效性。 (5)针对基于个别测点故障信号时频信息熵在受电弓诊断过程中存在精度不高、诊断特征过度冗余、分类识别耗时较久等问题，提出了基于特征选择与特征降维的受电弓故障诊断模型。该模型首先提取受电弓振动信号的多种时频域信息熵特征，而后集合至高维特征空间内，利用ReliefF算法、距离评价指数算法、联合互信息算法分别获得各自的特征排序结果，后将此特征排序结果进行信息融合，形成多重特征排序准则条件下的特征集合，选取识别精度最高的特征维度后，再采用流形学习进行再次降维分析，从而极大程度上降低特征维数，并且提升诊断正确率与计算速度，受电弓实测数据分析结果验证该分析模型的有效性。 关键词：受电弓；故障诊断；聚合经验模态分解；二代小波分解；信息熵；特征选取；降维
英文
With the rapid development of China’s economy and urbanization, people’s requirement for convenience and comfort traveling is in great demand. The high-speed railway and the urban rail transit play significant roles in people’s daily life. The construction of rail transit is meant to enhance people’s living standards, ease the pressure on urban traffic, maintain national economic growth, as well as adjust the relevant industrial structure. The majority of Chinese high-speed and subway vehicles apply electric traction; the electric power is generally acquired through the sliding contact between pantograph on the roof of vehicle and catenary. The pantograph, as a major part of power transmission, is undoubtedly one of the critical components of vehicle system. With the raising of vehicle’s speed, the pantograph faces more and more reliability chanllenge. The mechanical and electrical components fail occasionally, which seriously affects the stability and safety of vehicle. Therefore, it is significant to study the methods and techniques to ensure the reliable of the pantograph. However, the fault diagnosis for pantograph is still at the initial stage, which is far from reaching the demand of fast-growing vehicle in China. Therefore, aiming at the main problems encountered in the reliable and stable operation of pantograph, this paper starts from the research of pantograph fault diagnosis method. (1)Aiming at the feature values extraction problem that refers in the traditional ensemble empirical mode decomposition (EEMD). the white noise amplitude coefficient and its total average times cannot be effectively selected according to the signal characteristics. Therefore, an EEMD adaptive parameter selection algorithm is proposed. This method analyzes the different white noise amplitude parameters to work out the distribution influences on extremum points of the signal. This method can adaptively select the most effective white noise amplitude, which allows to optimize the distribution of extremum points. By analyzing the simulation signals, it is proved that the method not only can effectively reduce the modal aliasing under the premise, not reducing decomposition accuracy, but also do save the calculation time and improve the decomposition efficiency. (2)Based on improved EEMD decomposition algorithm, the decomposition of the intrinsic mode function (IMFs) of reconstructed phase space components, according to the signal characteristics of each component, using C-C algorithm to select the embedding dimension of phase space reconstruction process and time delay parameters, using the adaptive maximum likelihood estimation of intrinsic dimension evaluation method of IMFs phase space signal reconstruction, finally using the local tangent space manifold learning algorithm, high dimensional reconstruction of the signal with noise in the frequency space of phase space in low dimensional mapping, extraction of effective components which contains low dimensional space of the corresponding signal, to achieve a successful separation of signal and noise. Compared with other noise reduction methods, the results indicate that the proposed method has some advantages and can be applied to the noise reduction processing of the pantograph’s vibration signal. (3)Construction of improved EEMD decomposition and second generation wavelet decomposition of the pantograph vibration signal model based on information entropy, definition of the improved EEMD energy entropy, improved EEMD singular entropy, improved EEMD permutation entropy, improved EEMD approximate entropy, improved EEMD sample entropy, improved EEMD fuzzy entropy, the pantograph vibration signal characteristics, the information entropy of various parameters in the process of calculating optimization selection, puts forward on the pantograph vibration signal information entropy feature extraction model, through the fault characteristics and particle swarm optimization support vector machines (PSO-SVM) on the expansion analysis of pantograph fault identification, inquiry for the extraction type measuring points and fault sensitive of pantograph, and verify the modern time. The frequency analysis algorithm and the information entropy combined diagnosis method is feasible and effective in the feature extraction of the pantograph fault vibration signal. (4)According to the entropy feature based on the improved EEMD extraction algorithm, a pantograph pipe jacking is sensitive to the vibration signal of the pantograph bracket crack fault form, at present, the pantograph slipper vibration signal correlation entropy features for the pantograph fault isolation is poor, cannot satisfy the actual demands of fault diagnosis. Therefore, aiming at the characteristics of carbonglider vibration signal, second generational wavelet decomposition and information entropy feature extraction method are studied, which improves the fault recognition rate of carbon slide vertical vibration data greatly, also proves the effectiveness of the method on fault feature extraction of carbon slide plate vibration data. (5)Aiming at the problems of low accuracy, redundant diagnosis feature, time-consuming classification and recognition in the pantograph diagnosis process, a new pantograph fault diagnosis model based on feature selection and feature dimension reduction is proposed. This model firstly extracting the characteristics information entropy of multi-pantograph vibration signal in time and frequency domain, then set to high dimensional feature space, using the ReliefF algorithm, the distance evaluation index and joint mutual information (JMI) algorithm respectively ranking results of their respective features, will be the result for this sort of information fusion, form a set of characteristics of multiple feature ranking criteria; carry on the selection of the highest recognition accuracy of the feature dimension, then the manifold learning dimensionality analysis will analyze again, Therefore, greatly reduce the feature dimension, improve the diagnosis accuracy and calculation speed; the pantograph measured data analysis results validate the analysis model. Key words: pantograph; fault diagnosis; ensemble empirical mode decomposition; the second generation wavelet decomposition; information entropy; feature reduction; dimensionality reduction
