【作者】 阳同光；

【导师】 蒋新华；

【作者基本信息】 中南大学 ， 控制科学与工程， 2013， 博士

【摘要】 随着我国列车向高速化、信息化发展,列车的运行安全显得尤为重要。HXD1型交流传动电力机车是中国南车集团株洲电力机车有限公司和西门子公司合作开发的重载货运机车,是我国现代化铁路装备技术的重要标志性产品之一。异步牵引电机作为交流传动电力机车的核心设备,其正常运行对整个列车的运行安全起着至关重要的作用,因此,对异步牵引电机进行故障诊断研究具有十分重要的现实意义。本文首先对牵引电机故障诊断研究现状、信号分析方法和牵引电机故障机理进行分析和研究,为牵引电机故障诊断技术提供理论基础,然后以HXD1型电力机车牵引电机为研究对象,针对牵引电机特殊结构和运行环境,提出了几种有效的牵引电机故障诊断方法。考虑到牵引电机为非线性、强耦合的多变量系统,建立精确故障状态数学模型比较困难,提出一种基于混合蛙跳算法脊波神经网络观测器牵引电机故障诊断方法。首先利用脊波神经网络逼近牵引电机非线性部分,建立神经网络观测器,然后采用混合蛙跳算法对脊波神经网络参数进行优化,采用最优设计方法选取观测器反馈增益矩阵,最后根据观测器残差进行牵引电机故障诊断。该方法不仅充分融合了混合蛙跳算法和脊波神经网络的优点,具有较好的学习能力、较快的收敛速度和较高的故障诊断精度,而且能与无速度传感器牵引电机矢量控制进行很好融合,同时进行牵引电机状态识别和故障诊断。在HXD1型电力机车变频调速系统中,异步牵引电机转子断条和定子匝间短路等故障在定子电流中产生的故障特征量将穿越变流器对变流器一次侧(网侧)电流产生影响,因此,可以利用容易测取的变流器网侧电流信号进行牵引电机故障诊断。基于此,提出一种基于整流器一次侧(网侧)瞬时功率牵引电机转子故障诊断方法,在分析牵引电机定子电流和网侧电流的基础上,采用网侧电流、电压信号构建瞬时功率,并对其进行频谱分析,选择故障特征频率2Sfo作为转子断条故障诊断判据。该方法解决了电机电流分析方法中故障特征频率容易被淹没的问题,选择容易测取的网侧信号进行分析,简化了故障诊断系统硬件结构。针对牵引电机故障诊断中难以获取大量故障数据样本,故障特征提取困难等问题,提出一种基于核主成分分析(Kernel primaryComponent Analysis,KPCA)和相关向量机(Relevance Vector Machine, RVM)的牵引电机故障诊断方法。该方法将获取的采样电流信号进行小波包分解,选取频带能量值为样本向量,然后通过KPCA进行降维获得新的特征向量,并输入RVM进行故障模式识别。该方法结合KPCA的非线性特征提取能力和RVM的良好分类能力,具有故障诊断正确率高和分类时间较少的特点。HXD1电力机车牵引电机采用矢量控制技术,转子磁场定向不准是影响电力机车性能的关键问题之一。对故障诊断、转子磁场定向以及速度辨识进行综合分析和研究,发现利用物理量瞬时无功功率能同时进行牵引电机转子故障诊断、磁场定向校正和速度辨识。在此基础上,提出一种基于瞬时无功功率的牵引电机在线故障诊断方法。首先利用转子反电动势与电流矢量构建瞬时无功功率,并对其进行频谱分析,选用特征频率2fo作为转子断条的故障诊断判据,然后,利用该无功功率进行速度辨识,消除了模型参考自适应速度辨识中定子电阻和积分项的影响,最后利用两个模型的无功功率差,通过PI控制器校正转子磁场定向。该方法采用瞬时无功功率进行牵引电机故障诊断、速度辨识和转子磁场定向校正,不仅能有效进行牵引电机转子断条在线故障诊断,而且较好地融合无速度传感器矢量控制技术,提高了系统控制性能。研制了牵引电机交流传动系统半实物仿真平台,基于该平台,完成了无速度传感器牵引电机矢量控制系统以及在线故障诊断系统的软件编制。更多还原

【Abstract】 As trains become speedier and more information-based, the safety of the railway is also more prominent. HXD1AC drive electric locomotive is a heavy freight one developed by China’s CSR Zhuzhou Electric Locomotive Co., Ltd. and Siemens. As the core equipment of the AC electric locomotives, AC traction motors’safe operation is essential to the operation of the entire train, so fault diagnosis of AC traction motor is of great practical significance.In this paper, the state of motor fault diagnosis research, the signal analysis methods and traction motor fault diagnosis mechanism are researched and analyzed to provide the theory basis. For the special structure and operating environment of traction motor, several new-effective traction motor fault diagnosis methods are proposed.Considering that it is difficult to establish an accurate fault status mathematical model as the traction motor is nonlinear multivariable system with strong coupling, a shuffled frog leaping algorithm ridgelet neural network fault diagnosis method for traction motor is proposed. Firstly, the ridgelet neural network is used to approximate the nonlinear part of traction motor, and establish the neural network observer, and then the shuffled frog leaping algorithm is used to optimize the ridgelet neural network parameters, and the optimal design method is used to select the feedback gain matrix of observer, finally the residual of observer is used for traction motor fault diagnosis. With good learning ability, high fault diagnosis precision and fast convergence speed, this method not only combines the advantages of shuffled frog leaping algorithm and ridgelet neural network, but can fuse well with speed sensorless traction motor vector control, and detect the traction motor faults and identify the motor states at the same time.In variable frequency speed regulation system of HXD1electric locomotive, the fault characteristic quantity generated in the stator current because of traction motor broken bar fault and stator inter-turn short circuit of asynchronous will pass through the converter to converter primary side (network side) and influence the current of primary side, therefore the easily measured current signal of the network side in traction converter can be used for motor fault diagnosis. A novel fault diagnosis based on the rectifier side instantaneous power of traction motor is proposed in this paper. At first, we analyze the stator current and the primary-side current, and then build the instantaneous power. By analyzing instantaneous power spectrum, we select the characteristic frequency2Sf0as broken rotor bars fault diagnostic criterion. The method overcomes the problems of the fault characteristic frequency covered easily by fundamental frequency in the traditional method of current analysis, and simplifies the hardware structure of fault diagnosis system by selecting the network side signal.A variety fault classification of traction motor fault diagnosis is proposed based on Kernel primary component analysis (KPCA) and Relevance vector machine (RVM).This method uses the sampling current signal with wavelet analysis to construct the learning sample vectors, and then uses the KPCA to reduce dimension, and the new fault characteristic vectors are inputed RVM to carry out training and fault classification, finally,traction motor faults diagnosis achieve more satisfactory results by the use of RVM.In HXD1electric locomotive traction motor vector control system, rotor field-oriented inaccuracy is one of the key problems affecting electric locomotive performance. We comprehensively analyze the fault diagnosis, rotor field orientation and speed identification, and find that instantaneous reactive power can be used not only for traction motor rotor fault diagnosis, but also for field oriented control and speed identification. Firstly, we constructe the instantaneous reactive power through vector cross product with the rotor EMF and current, and select the characteristic frequency of2Sf1as the rotor broken bar fault diagnosis criterion by analyzing the instantaneous reactive power spectrum. And then we use the instantaneous reactive power to identify speed, eliminating the effect of traditional MRAS speed identification of stator flux linkage equations of the stator resistance and the integral effect. Finally, we adjust rotor magnetic field oriention using PI with the difference of two reactive power models. The method, as it selects the instantaneous reactive power to detect traction motor fault, identifies the speed and corrects the rotor flux orientation, not only can effectively diagnose traction motor broken rotor bar fault, but also can integrate the speed sensorless vector control technology to improve the control performance of the system.Hardware in loop(HIL) simulation platform of AC drive of traction motor is developed.The software of speed sensorless traction motor vector control system and online fault diagonosis system are established based on the platform.更多还原