基于神经网络的道岔智能故障诊断方法的研究

发布时间：2018-05-01 07:23

本文选题：道岔智能故障诊断 + BP神经网络　；参考：《兰州交通大学》2011年硕士论文

【摘要】：随着铁路运行速度的逐年攀升,较快的列车运行速度对道岔提出了更加严格的要求。本论文是道岔监测系统项目的一个子课题,是在道岔监测系统提供的大量道岔状态数据的基础上,应用神经网络对道岔进行智能故障诊断。本文是对道岔智能故障诊断的一次初步尝试,目的是给道岔监测系统的故障诊断功能提供一种可行性实现方法。论文首先介绍神经网络的定义和原理,并分别的从两个典型神经网络——反向传播神经网络(BP神经网络)和径向基神经网络(RBF神经网络)两方面详细阐述了网络的构造,学习算法及其应用。随后,为了方便构造神经网络系统,在介绍道岔转换系统结构的基础上,分析各种典型故障的机理,对杂乱无章的各种故障进行了统一分类,并系统地介绍提供各种道岔监测数据的道岔监测系统。最后,运用MATLAB分别构造BP神经网络模型和RBF神经网络模型。多次测试后,对网络性能进行对比研究,发现诊断结果基本达到预期的诊断要求,完成了道岔智能故障诊断的理论研究。本文重点在以下几个方面进行探索与研究：按照道岔故障机理和神经网络结构特性,把道岔故障分为三类,每类构造一个子神经网络,总体组建成一个并行神经网络系统框架。选择最优BP算法。对某个子神经网络构造BP神经网络模型,应用多种常见BP算.法分别对网络训练并测试,从测试结果中获得每种BP算法的优势和劣势。设计基于BP算法的并行神经网络故障诊断模型。针对于每一个子神经网络,利用经验公式得出隐含层神经元个数的最小范围,然后在最小范围内对隐含层神经元个数逐个尝试,分析不同隐含层节点数对网络性能的影响,采用Levenberg-Marquart算法构造最优BP神经网络。然后训练网络并进行故障诊断测试。设计基于RBF算法的并行神经网络故障诊断模型。针对于每一个子神经网络,通过多次试验获取隐含层神经元个数和径向基分布密度的最优值并构造性能最佳的RBF神经网络,然后训练网络并进行故障诊断测试。通过一系列理论研究和大量仿真试验证明：神经网络技术运用在道岔智能故障诊断方面是切实可行的。该方法能快速、有效地诊断出故障原因,为维修人员提供技术支持。
[Abstract]:With the increasing of railway running speed, the higher train speed puts forward more strict requirements for turnout. This paper is a sub-topic of the turnout monitoring system project, which is based on a large number of switch state data provided by the turnout monitoring system, and uses the neural network to diagnose the intelligent fault of the switch. This paper is a preliminary attempt for intelligent fault diagnosis of turnout, which aims to provide a feasible method for fault diagnosis of turnout monitoring system. Firstly, the definition and principle of neural network are introduced, and the structure of neural network is described in detail from two aspects: back propagation neural network (BP) and radial basis function neural network (RBF). Learning algorithm and its application. Then, in order to facilitate the construction of neural network system, on the basis of introducing the structure of switch switching system, the mechanism of various typical faults is analyzed, and the disorderly faults are classified uniformly. The turnout monitoring system which provides all kinds of turnout monitoring data is introduced systematically. Finally, BP neural network model and RBF neural network model are constructed by MATLAB. After many tests, the performance of the network is compared, and it is found that the diagnosis results basically meet the expected diagnostic requirements, and the theoretical research on intelligent fault diagnosis of switch is completed. This paper focuses on the following aspects of exploration and research: According to the fault mechanism of switch and the characteristics of neural network structure, the switch faults are divided into three types, each of which is composed of a sub-neural network and a parallel neural network system framework. The optimal BP algorithm is selected. The BP neural network model is constructed by a subneural network, and many common BP calculations are applied. The advantages and disadvantages of each BP algorithm are obtained from the test results. A parallel neural network fault diagnosis model based on BP algorithm is designed. For each sub-neural network, the minimum range of the number of neurons in the hidden layer is obtained by empirical formula, and then the number of neurons in the hidden layer is tried one by one in the minimum range to analyze the effect of the number of hidden layer nodes on the performance of the network. The optimal BP neural network is constructed by Levenberg-Marquart algorithm. The network is then trained and tested for fault diagnosis. A parallel neural network fault diagnosis model based on RBF algorithm is designed. For each sub-neural network, the optimal values of the number of hidden layer neurons and the radial basis distribution density are obtained through several experiments, and the RBF neural network with the best performance is constructed, and then the network is trained and tested for fault diagnosis. Through a series of theoretical research and a large number of simulation experiments, it is proved that the application of neural network technology in intelligent fault diagnosis of turnout is feasible. This method can quickly and effectively diagnose the fault cause and provide technical support for maintenance personnel.
【学位授予单位】：兰州交通大学
【学位级别】：硕士
【学位授予年份】：2011
【分类号】：TH165.3;TP183

【引证文献】