全矢谱—支持向量数据描述及故障诊断应用研究
发布时间：2019-05-08 08:02
【摘要】：在设备故障诊断中,传统的数据处理方法只能针对单通道数据进行分析,而单通道数据往往不能将设备的空间运动信息完整的表征出来。而作为全信息分析方法的一种,全矢谱分析技术在处理同源多通道故障信号的同时能够体现更全面准确的转子运动空间特征信息。在此基础上本文将全矢谱技术与支持向量数据描述相结合,提出了全矢谱支持向量数据描述(Vector Spectrum Support Vector Data Description, VSSVDD)故障诊断方法。针对支持向量数据描述(Support Vector Data Description, SVDD)分类方法中训练样本数目受限的问题,本文对SVDD分类器作二次改进,引入动态支持向量数据描述(DSVDD)分类方法。该方法在训练样本中不断注入新的样本进而不断更新分类边界,从而更准确的表征了目标样本的区域边界。本文主要研究和解决问题如下： 第一,支持向量数据描述方法是建立在统计学习理论之上的,核函数的引入可以把低维空间的非线性问题转化为高维空间的线性问题。选择不同的核函数对SVDD的分类效果不同。 第二,运用全矢谱分析方法对采样数据进行分析处理,并且提取典型倍频上的幅值作为SVDD分类器的特征向量。实验表明经过全矢谱特征提取后的SVDD的分类效果较未经特征提取SVDD的分类效果更为明显。通过实验研究验证了全矢谱支持向量数据描述故障诊断方法对测试样本进行分类的可行性与有效性。 第三,运用全矢谱支持向量数据描述方法对设备性能退化评估引入隶属度和相对距离的概念避免了超球体边界误差带来的影响,可以将测试样本的状态更加精确的表述出来；同时又体现了状态变化的过程。 第四,提出动态支持向量数据描述分类方法的改进型。该方法的提出改变了原来SVDD分类方法中,分类器经过一次训之后分类边界永不改变的现状。它将测试得到的目标样本与本次测试以前的支持向量集一起形成新的训练样本,然后对SVDD重新训练。这样分类边界将更能体现设备的正常样本特征。
[Abstract]:In the fault diagnosis of equipment, the traditional data processing method can only analyze the single channel data, but the single channel data can not completely represent the spatial motion information of the equipment. As one of the methods of full information analysis, the full vector spectrum analysis technique can not only deal with the same source multi-channel fault signal, but also reflect the more comprehensive and accurate information of the rotor motion space. On this basis, this paper combines full vector spectrum technique with support vector data description, and proposes a fault diagnosis method for (Vector Spectrum Support Vector Data Description, VSSVDD) based on full vector spectrum support vector data description. In order to solve the problem that the number of training samples is limited in the support vector data description (Support Vector Data Description, SVDD) classification method, this paper makes a second improvement on the SVDD classifier, and introduces the dynamic support vector data description (DSVDD) classification method. In this method, new samples are continuously injected into the training samples, and then the classification boundaries are constantly updated, thus representing the region boundary of the target samples more accurately. In this paper, the main research and solutions are as follows: first, support vector data description method is based on statistical learning theory, the introduction of kernel function can transform the nonlinear problem of low-dimensional space into the linear problem of high-dimensional space. Different kernel functions have different effects on SVDD classification. Secondly, the method of full vector spectrum analysis is used to analyze and process the sampled data, and the amplitude on the typical frequency doubling is extracted as the feature vector of the SVDD classifier. The experimental results show that the classification effect of SVDD after full vector feature extraction is more obvious than that of SVDD without feature extraction. The feasibility and effectiveness of the fault diagnosis method based on full vector spectrum support vector data description for classification of test samples is verified by experimental research. Thirdly, the concept of membership degree and relative distance is introduced into the evaluation of equipment performance degradation by using full vector spectral support vector data description method to avoid the influence of hypersphere boundary error, and the state of test samples can be expressed more accurately. At the same time, it reflects the process of state change. Fourth, an improved classification method for dynamic support vector data description is proposed. The proposed method changes the status quo of the original SVDD classification method, in which the classification boundary of the classifier never changes after a training. It forms a new training sample with the support vector set before this test, and then retrains the SVDD. In this way, the classification boundary can better reflect the normal sample characteristics of the equipment.
【学位授予单位】：郑州大学
【学位级别】：硕士
【学位授予年份】：2011
【分类号】：TH165.3
本文编号：2471739
[Abstract]:In the fault diagnosis of equipment, the traditional data processing method can only analyze the single channel data, but the single channel data can not completely represent the spatial motion information of the equipment. As one of the methods of full information analysis, the full vector spectrum analysis technique can not only deal with the same source multi-channel fault signal, but also reflect the more comprehensive and accurate information of the rotor motion space. On this basis, this paper combines full vector spectrum technique with support vector data description, and proposes a fault diagnosis method for (Vector Spectrum Support Vector Data Description, VSSVDD) based on full vector spectrum support vector data description. In order to solve the problem that the number of training samples is limited in the support vector data description (Support Vector Data Description, SVDD) classification method, this paper makes a second improvement on the SVDD classifier, and introduces the dynamic support vector data description (DSVDD) classification method. In this method, new samples are continuously injected into the training samples, and then the classification boundaries are constantly updated, thus representing the region boundary of the target samples more accurately. In this paper, the main research and solutions are as follows: first, support vector data description method is based on statistical learning theory, the introduction of kernel function can transform the nonlinear problem of low-dimensional space into the linear problem of high-dimensional space. Different kernel functions have different effects on SVDD classification. Secondly, the method of full vector spectrum analysis is used to analyze and process the sampled data, and the amplitude on the typical frequency doubling is extracted as the feature vector of the SVDD classifier. The experimental results show that the classification effect of SVDD after full vector feature extraction is more obvious than that of SVDD without feature extraction. The feasibility and effectiveness of the fault diagnosis method based on full vector spectrum support vector data description for classification of test samples is verified by experimental research. Thirdly, the concept of membership degree and relative distance is introduced into the evaluation of equipment performance degradation by using full vector spectral support vector data description method to avoid the influence of hypersphere boundary error, and the state of test samples can be expressed more accurately. At the same time, it reflects the process of state change. Fourth, an improved classification method for dynamic support vector data description is proposed. The proposed method changes the status quo of the original SVDD classification method, in which the classification boundary of the classifier never changes after a training. It forms a new training sample with the support vector set before this test, and then retrains the SVDD. In this way, the classification boundary can better reflect the normal sample characteristics of the equipment.
【学位授予单位】：郑州大学
【学位级别】：硕士
【学位授予年份】：2011
【分类号】：TH165.3
【引证文献】
相关硕士学位论文 前1条
1 付玉荣;全矢—模糊聚类及其在故障诊断中的应用研究[D];郑州大学;2013年
，本文编号：2471739
本文链接：https://www.wllwen.com/kejilunwen/jixiegongcheng/2471739.html
