基于弱信号特征提取的早期诊断方法及其应用研究

发布时间：2018-06-03 13:17

本文选题：弱特征提取 + 早期故障诊断　；参考：《哈尔滨工业大学》2011年硕士论文

【摘要】：工业自动化水平的提高对设备提出了早期故障诊断的需求。由于复杂的工作环境和现场噪声的干扰,设备的故障信号容易被噪声污染,如何有效的从低信噪比监测信号中提取弱故障特征成为故障诊断领域迫切需要解决的问题。结合故障信号自身的特点,运用弱信号特征提取方法和早期故障诊断理论解决机械故障诊断中信号降噪和弱特征加强等问题以及进行设备的早期故障诊断是当前机械故障诊断领域迫切需要研究的重要课题之一。本文研究了信号的降噪方法、设备故障的弱特征提取、早期故障诊断理论及其在轴承故障诊断中的应用。 (1)采用小波分析的多分辨率算法,信号被分解为不同尺度上的低频分量和高频分量,信号中的微弱特征随着尺度因子的递进变化逐渐被放大。在每个分解尺度上,将变换系数与通过阈值规则设置的门限因子进行比较处理,实现了信号的降噪处理,提高了信噪比和加强了信号中弱特征。 (2)基于信号的局部特征分析,经验模态分解采用包络分析理论将故障信号分解为频率从高到低变化的固有模态分量。针对经验模态分解过程的边界效应问题和传统延拓方法的缺陷,本文采用了基于奇异值分解和支持向量回归机的端点预测延拓方法。奇异谱分析方法能够有效地检测信号中的周期特征,为确定支持向量回归机延拓点的数目提供了一种可行的方法。通过阈值扫描法剔除了分解产生的“伪”分量,分析表明该方法能够有效地抑制边界效应和提取信号中的弱特征分量。 (3)针对设备早期故障特征不明显、可分性差的特点,采用支持向量机建立最优分类决策模型。采用高斯核函数增加样本的可分性,实现了故障特征空间的映射变换。本文以轴承故障为研究对象,系统地研究了故障类别、训练样本数目和故障诊断精度之间的关系,分析表明该方法能够有效地解决小样本情况下故障诊断精度低的问题。 (4)基于Matlab语言和VB环境的混合编程技术开发了信号的特征提取和设备的早期故障诊断可视化操作系统,提供了一个操作方便、可扩展性强的特征提取和早期故障诊断平台。
[Abstract]:The improvement of the level of industrial automation puts forward the requirement of early fault diagnosis for the equipment. Because of the complex working environment and the interference of the field noise, the fault signal of the equipment is liable to be polluted by noise. How to extract the weak fault feature from the low signal-to-noise ratio monitoring signal becomes an urgent problem to be solved in the field of fault diagnosis. Combined with the characteristics of the fault signal itself, The application of weak signal feature extraction method and early fault diagnosis theory to solve the problems of signal noise reduction and weak feature enhancement in mechanical fault diagnosis and the early fault diagnosis of equipment are urgently needed in the field of mechanical fault diagnosis. One of the important subjects of research. In this paper, the signal denoising method, the weak feature extraction of equipment fault, the theory of early fault diagnosis and its application in bearing fault diagnosis are studied. The signal is decomposed into low frequency component and high frequency component in different scales, and the weak feature of the signal is amplified gradually with the change of scale factor. 1) using the multi-resolution algorithm of wavelet analysis, the signal is decomposed into low-frequency and high-frequency components on different scales. In each decomposition scale, the transform coefficient is compared with the threshold factor set by the threshold rule, the signal noise reduction is realized, the signal-to-noise ratio (SNR) is improved and the weak feature in the signal is strengthened. 2) based on the local characteristic analysis of the signal, the empirical mode decomposition uses the envelope analysis theory to decompose the fault signal into the inherent modal component of the frequency change from high to low. Aiming at the boundary effect problem of empirical mode decomposition process and the defects of the traditional continuation method, the endpoint prediction continuation method based on singular value decomposition (SVD) and support vector regression machine (SVM) is used in this paper. The singular spectrum analysis method can effectively detect the periodic characteristics in the signal, which provides a feasible method for determining the number of extension points of the support vector regression machine. The "pseudo-component" component produced by decomposition is eliminated by threshold scanning method. The analysis shows that the method can effectively suppress the boundary effect and extract the weak characteristic component from the signal. 3) aiming at the characteristics that the early fault feature is not obvious and the separability is poor, support vector machine (SVM) is used to establish the optimal classification decision model. The Gao Si kernel function is used to increase the separability of the samples and the mapping transformation of fault feature space is realized. In this paper, the relationship among the fault types, the number of training samples and the fault diagnosis accuracy is systematically studied. The analysis shows that this method can effectively solve the problem of low fault diagnosis accuracy in the case of small samples. Based on the mixed programming technology of Matlab language and VB environment, the visual operating system of signal feature extraction and early fault diagnosis is developed, which provides a convenient and extensible platform for feature extraction and early fault diagnosis.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2011
【分类号】：TH165.3

【参考文献】