基于小波分析的故障模式提取研究

发布时间：2018-03-28 18:11

本文选题：小波分析　切入点：模式提取　出处：《北方工业大学》2012年硕士论文

【摘要】：先进的故障诊断技术是设备发展的前提,没有先进的故障诊断技术设备发展就无从谈起。故障诊断技术不仅为设备发展提供动力,也是设备平稳运行的保障,避免设备故障带来的损失。在故障诊断技术中,故障特征提取至关重要,小波分析是一种新型的故障提取技术。小波分析技术在高维数据特征模式提取中有着广泛的应用。它不仅能有效降低数据维度,还能在时域和频域对信号数据进行分解,提取出信号中的激励特征。但小波分析在故障模式提取应用中有四方面配置选择问题：小波基函数选择、小波分解层数、小波系数选取和特征向量生成算法选用。这四个方面问题影响着小波分析提取出的故障模式的优劣,并制约了电路板故障智能诊断的正确率。本文通过对小波的研究,提出了小波分析在故障模式提取中的评价标准,以便于小波分析史好发挥其特性。小波分析作为一种新型的信号分解工具,能根据时域和频域的需求,拉远或拉近“显微镜”镜头,达到提取信号故障特征的目的,因此小波分析更适合提取信号低频中的轮廓信息和高频中的奇异信号特征。论文介绍了能量、极大值、小波熵三种特征提取算法和BP神经网络、支持向量机两种模式识别算法,以及信息融合技术。并在此基础上提出了波动性函数、信噪比、时间复杂度、诊断正确率等基于小波分析的故障模式提取评价标准。最后选用与或逻辑输出控制电路作为测试电路,采集信号数据。测试中,根据电路的工作原理和常见故障类型,选取25个采样节点对38种电路状态分三次采集信号波形数据。然后,以小波分析作特征提取算法,以神经网络和和支持向量机作分类器,验证了提出的小波分析配置评价标准是有效的。
[Abstract]:Advanced fault diagnosis technology is the premise of equipment development. Without advanced fault diagnosis technology, there can be no development of equipment. Fault diagnosis technology not only provides power for equipment development, but also guarantees the smooth operation of equipment. In fault diagnosis, fault feature extraction is very important. Wavelet analysis is a new fault extraction technology. Wavelet analysis is widely used in feature pattern extraction of high-dimensional data. It can not only reduce the dimension of data, but also decompose the signal data in time and frequency domain. But in the application of wavelet analysis in fault mode extraction, there are four problems in configuration selection: wavelet basis function selection, wavelet decomposition layer number, The selection of wavelet coefficients and the selection of feature vector generation algorithm affect the advantages and disadvantages of the fault mode extracted by wavelet analysis, and restrict the correct rate of fault intelligent diagnosis of circuit board. The evaluation criteria of wavelet analysis in fault mode extraction are put forward in order to give full play to the characteristics of wavelet analysis history. As a new signal decomposing tool, wavelet analysis can draw far or close the lens of "microscope" according to the demand of time domain and frequency domain, so as to extract the fault feature of signal. Therefore, wavelet analysis is more suitable for extracting contour information from low frequency signal and singular signal feature from high frequency. This paper introduces three feature extraction algorithms: energy, maximum, wavelet entropy, BP neural network and support vector machine. On the basis of this, the evaluation criteria of fault pattern extraction based on wavelet analysis, such as volatility function, signal-to-noise ratio, time complexity and diagnostic accuracy, are proposed. Finally, the control circuit of and or logic output is selected as the test circuit to collect the signal data. In the test, according to the working principle of the circuit and the common fault type, Twenty-five sampling nodes are selected to collect the signal waveform data three times for 38 circuit states. Then, wavelet analysis is used as feature extraction algorithm, neural network and support vector machine are used as classifiers. It is verified that the proposed wavelet analysis configuration evaluation criteria are effective.
【学位授予单位】：北方工业大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TH165.3

【参考文献】