基于小波包—神经网络的机车轴承故障诊断的研究

【作者】 张国瑞；

【导师】 鲁五一；

【作者基本信息】 中南大学 ， 控制科学与控制工程， 2010， 硕士

【摘要】 机车轮对滚动轴承是铁路机车中重要而又极易发生故障的部件,由于轴承严重故障而引起的重大事故时而发生,对人民生命财产安全带来安全隐患,所以对机车轮对滚动轴承故障诊断的研究是一个亟待解决的课题。本文在总结其他有关课题的工作的基础上,对铁路机车轮对轴承在途故障监测进行研究。通过分析轴承发出的声学信号来实现对各种不同类型故障的诊断与分类。首先,本文从机车轴承声音信号产生的机理出发,研究了滚动轴承的振动原理,负载时由于内外圈、滚动体刚度不均匀和腐蚀、磨损等原因引起的振动以及故障情况下的振动特点。分析了轴承声音信号的时域、频域的各种特征参数和频谱结构,在进行性能比较后选取了进行故障诊断的最佳特征参数,为滚动轴承振动故障诊断提供了理论依据；然后,介绍了小波分析的基本原理与应用,具体介绍了连续小波变换、离散小波变换、小波包分析和各种常用小波基,以及它们各自的使用场合。并对各种典型故障轴承的信号进行小波包分解,重构了正常、外圈、内圈、滚动体和磨损故障有单个损伤点时的仿真分解信号,对其进行特征提取,作为神经网络的输入信号,包络解调后的功率谱分析可作为模式识别准确性的参考；最后将粒子群算法优化神经网络引入故障模式识别中。在对BP神经网络的缺陷进行研究之后,提出了使用粒子群优化算法对BP神经网络进行改进的优化网络,在对轴承信号特征参数进行模式识别的应用中将其与普通BP神经网络进行性能对比,并使用通过小波包分析处理后的样本数据对网络进行训练和测试。本文对基于小波包-PSO优化神经网络的铁路机车轴承故障诊断系统进行实例分析,通过试验证明这种方法是有效和可行的,对铁路在途故障预警系统的建立具有一定的推进作用。更多还原

【Abstract】 Wheel rolling bearing is a very important part in railway locomotive but easily broken. In railway Major incidents caused by serious rolling bearing fault happens every now and then which brings on people’s lives and property security risks, so research on locomotive rolling bearing fault diagnosis is an urgent subject. This paper summarized the work of other relevant issues, achieving rolling bearing fault diagnosis and classify of locomotive in-transit.This paper firstly researched the principle of rolling bearing noises and the abnormal sounds produced by uneven bearing cup, cone or roller or corrosion, etc. and their characteristics, analyzing the various feature parameters and spectral structure of the bearing acoustic signal in time domain and frequency domain. The best feature parameters was selected after the performance comparison for fault diagnosis, providing a theoretical basis for rolling bearing fault diagnosis by acoustic signal. Then, the paper introduced the basic principle of wavelet analysis and application, and the continuous wavelet transform, discrete wavelet transform, wavelet packet analysis and a variety of commonly used wavelet bases, and their respective applied occasions, decomposing and reconstructing the acoustic signal of normal bearings and single cup/cone/roller spall bearings, extracting the feature parameters as the input parameters of neural network. The power spectrum after envelope demodulation could be used as reference for the accuracy of pattern recognition. Finally, the paper used particle swarm optimized neural network into fault pattern recognition. Optimizing BP neural network by particle swarm optimization algorithm was proposed after researching the defects of BP neural network, comparing with the common BP neural network in pattern recognition with bearing acoustic signal feature parameters, training and testing using the sample data after wavelet packet analysis and processing.The paper analyzed the railway locomotive bearing fault diagnosis system based on wavelet packet and PSO optimized neural network. The experiment show that this method is effective and feasible and good to the in-transit railway truck fault early warning system.更多还原