【作者】 贾天丽；

【导师】 王卓；

【作者基本信息】 北京交通大学 ， 安全技术及工程， 2011， 硕士

【摘要】 我国城市轨道交通在迅速发展的同时,轨道交通列车的运营安全问题日益突出。城市轨道交通列车是机电一体化的复杂运行系统,其走行系的系统状态和性能在不断演化的过程中,会形成安全隐患,甚至引发事故,严重影响市民的日常出行,同时诱发严重的社会问题。基于神经网络的故障诊断是智能故障诊断理论与技术的一个重要研究方向。本文对走行系的滚动轴承故障诊断问题展开以下研究：1、针对目前轨道交通列车走行系的滚动轴承故障诊断中存在的问题,论文结合小波包分析、BP神经网络和证据理论,提出了一种综合故障诊断方法。在对加速度信号进行小波包分解的前提下,利用神经网络进行训练,达到故障识别的目标。并通过证据理论对不同神经网络的训练成功概率进行融合,选择相对最优的故障诊断方法。2、论文利用轴承试验台采集内环故障、外环故障、滚动体故障和正常轴承四种滚动轴承的加速度信号,对其进行小波包分析,通过对采集的原始数据分别进行三层,四层小波包分析,并把分析结果作为BP神经网络的输入样本进行网络训练。选择合适的小波包分解层数。3、利用不同神经网络对小波包分解产生的样本进行训练,对轨道列车走行部的滚动轴承进行了故障诊断和识别,找出轴承的故障位置。论文还将小波包理论和算法有机地结合在一起,吸取了二者的优点,提高了滚动轴承故障诊断的效率4、基于证据理论方法进一步对各故障位置的诊断概率进行融合,选出最优诊断方法。将不同故障类型的诊断精度看作证据,实现其在时间域、空间域上进行的融合,对融合结果进行比较,选出最优诊断方法,从而提高了诊断的准确度。更多还原

【Abstract】 ABSTRACT:As urban rail transit of China develops rapidly, the operation safety problem of railway highlights increasingly. Urban rail transit train is complex operation system integration of electromechanical integration. The evolving process of the running gear system state and performance will form the security hidden danger, and even cause accidents, which influences seriously people daily travel and causes serious social problem. Fault diagnosis based on neural network is an important research direction to intelligent fault diagnosis of the theory and technology. This paper is to fault diagnosis of running gear rolling bearing on the following research1. Aiming at the problems of rolling bearing fault diagnosis of running gear in railway line, based on wavelet packet analysis, the BP neural network and evidence theory, this paper puts forward a comprehensive fault diagnosis method. In for acceleration signal wavelet packet decomposition premise, trained by use of neural network, the goal of fault recognition achieves. And through the fusion of training success probability of the evidence theory to different neural network, the best fault diagnosis method is chosed relatively.2. Through acceleration signal of four kind of rolling bearing with fault as the inner loop, the outer loop, the rolling element and the normal rolling bearing, earring on the wavelet packet analysis, using analysis of the original data of three, four wavelet packet, taking the results as the input sample of the BP neural net for network training, the paper choose the right wavelet packet to decompos layers.3. Training by using samples of different neural network to wavelet packet decomposition production, with the help of fault diagnosis and recognition for railway train running gear of the rolling bearing, the bearing fault position is found out. This paper links judiciously wavelet packet theory with algorithm, draws on their advantages, improves the efficiency of the bearing fault diagnosis.4. Based on evidence theory method, each fault position diagnosis probability further fuse to select the optimal diagnostic method. Considered evidence for different fault type the diagnostic accuracy, realizing fusion in the time domain and the space domain, comparing the results of fusion, to select the optimal diagnostic methods, the accuracy of the diagnosis improves.更多还原