基于EMD与神经网络的柱塞泵故障诊断方法

发布时间：2017-12-30 18:10

本文关键词：基于EMD与神经网络的柱塞泵故障诊断方法　出处：《华中科技大学》2011年硕士论文　论文类型：学位论文

【摘要】：混凝土泵车是一种负载变化复杂、工作环境恶劣的工程机械,其内部元件常由于疲劳或油液污染引起各种故障的发生,混凝土泵车一旦出现事故往往对生产安全造成很大危害。根据统计,柱塞泵是混凝土泵车中最主要的故障源之一,但是柱塞泵作为泵车泵送动力的来源,既是旋转机械,又是往复运动机械和机液转换元件,工作过程中既有机械零件间的振动,又有工作介质引起的冲击,因而其给诊断带来了很大的难度。本研究分析了柱塞泵的结构特点与运动规律,指出常见故障的发生位置与振动频率,通过传统频谱分析方法做出功率谱与包络谱,但由于介质冲击影响严重无法找出故障特征。进而提出利用Hilbert-Huang变换的核心理论EMD,将实验台中测取的五种常见故障与正常状态的振动信号进行分解,得到与自身频率组成相符的各频段信号波形,通过分析将其中蕴含故障信息丰富的高频段分量建立时间序列AR模型,并将模型参数作为故障诊断的特征参数,作为后续神经网络的输入。神经网络具有强大的非线性映射和并行处理的功能,网络中的权值向量模拟人大脑神经元的记忆方式,可以较为稳定的存储来自输入样本的特征信息,并利用已有信息对新输入做出分类辨识,因此本研究利用较新型的Fuzzy ARTMAP神经网络对柱塞泵六种状态振动信号的特征参数进行学习与分类,结果表明该方法可以有效地完成诊断。本文还就特征参数个数对神经网络的分类效率影响进行了讨论,并实现了在不影响诊断准确率的条件下对特征参数数量的减少。
[Abstract]:Concrete pump car is a kind of construction machinery with complex load change and bad working environment. The internal components of concrete pump car are often caused by various failures due to fatigue or oil pollution. According to statistics, piston pump is one of the main fault sources of concrete pump vehicle, but piston pump is the source of pump power. It is not only a rotating machine but also a reciprocating moving machine and a mechanical and hydraulic conversion element. In the working process there is both the vibration between the mechanical parts and the impact caused by the working medium, so it is very difficult to diagnose. In this paper, the structure characteristics and motion law of piston pump are analyzed, and the location and vibration frequency of common faults are pointed out. The power spectrum and envelope spectrum are obtained by traditional spectrum analysis method. However, due to the serious impact of the medium, the fault characteristics can not be found, and then the core theory of EMD based on Hilbert-Huang transform is proposed. By decomposing the five kinds of common faults and the vibration signals in normal state, the signal waveforms of each frequency band which are consistent with its own frequency composition are obtained. Time series AR model is established by analyzing the components of high frequency band which contain abundant fault information, and the model parameters are regarded as the characteristic parameters of fault diagnosis. As the input of the subsequent neural network, the neural network has powerful nonlinear mapping and parallel processing functions. The weight vector in the network simulates the memory mode of human brain neurons. It can store the feature information from the input sample stably and use the existing information to classify and identify the new input. Therefore, a new type of Fuzzy ARTMAP neural network is used to study and classify the characteristic parameters of six state vibration signals of piston pump. The results show that the method can effectively complete the diagnosis. The effect of the number of characteristic parameters on the classification efficiency of neural networks is also discussed in this paper. The reduction of the number of feature parameters without affecting the diagnostic accuracy is realized.
【学位授予单位】：华中科技大学
【学位级别】：硕士
【学位授予年份】：2011
【分类号】：TH322

【参考文献】