基于支持向量机的转子系统故障诊断方法研究
发布时间：2018-10-13 09:32
【摘要】：近年来,关于设备状态监测与故障诊断方面的研究工作得到越来越高的重视,相关的理论研究也得到迅速发展。支持向量机在解决基于小样本情况的分类问题方面表现出良好的性能。它根据结构风险最小化原则,具有全局最优解,根据有限的样本信息在模型的复杂性和学习能之间寻求最佳折衷,以获得最好的推广能力并能有效地解决“过学习”问题。 本文结合转子实验台上模拟的常见故障,采用熵带法对故障振动信号进行特征提取。为了使支持向量机具有更高的分类准确率,运用粒子蚁群算法对支持向量机的参数进行优化。针对故障多分类问题围绕以上实验分析和理论算法,本文的主要工作内容和研究结论如下: 1)在转子实验台上模拟了四种典型故障,分析了四种故障的机理并对故障信号进行了滤波消澡、频谱分析、轴心轨迹分析。在此基础上分析了信号在时域的奇异值谱熵、频域的功率谱熵、时频域的小波能谱熵和小波空间谱熵。并计算了四种故障信号的熵带范围,讨论了常规的基于信息熵的故障诊断方法。 2)因直接把熵带作为SVM的训练样本和测试样本存在数据冗余问题,故以熵带数据为基础,对其作为SVM的训练样本进行了数据预处理研究。包括样本归一化和主元特征提取。后续的实验表明,经过处理后的熵带数据不仅能够反映振动信号的特征,而且适合SVM进行模型训练和故障分类。 3)以构造最优分类器为目标,系统地研究了PSO算法和GA算法优化SVM参数后对分类准确率的影响。通过把已经处理好的数据输入到SVM中,分别应用GA和PSO对SVM的核参数与惩罚因子优化并对未知故障类别的样本测试发现,GA优化后的SVM分类性能较差,且模型训练时间较长,而PSO优化得到的SVM具有良好的分类准确率和较快的训练时间。 4)由于本研究是多故障分类问题,而SVM是二分类器,故基于一对多的方法设计了可以分离四种故障的SVM多故障分类器。对各个分类器分别应用PSO算法进行参数寻优。并基于以上算法流程开发了一套基于MATLAB GUI的转子故障诊断系统,子系统一可以实现对振动信号的消澡分析,频谱分析,轴心轨迹分析等;子系统二可以根据样本特点对分类器进行参数寻优,实现对未知故障的判别,实验结果验证了该系统的有效性。
[Abstract]:In recent years, more and more attention has been paid to the research of equipment condition monitoring and fault diagnosis, and the related theoretical research has been developed rapidly. Support vector machines (SVM) show good performance in solving classification problems based on small samples. According to the principle of structural risk minimization, it has the global optimal solution. According to the limited sample information, it seeks the best tradeoff between the complexity of the model and the learning ability in order to obtain the best generalization ability and solve the "overlearning" problem effectively. In this paper, the entropy band method is used to extract the characteristic of the fault vibration signal in combination with the common faults simulated on the rotor test bench. In order to make SVM have higher classification accuracy, particle ant colony algorithm is used to optimize the parameters of SVM. The main contents and conclusions of this paper are as follows: 1) four kinds of typical faults are simulated on the rotor test bench. The mechanism of four kinds of faults is analyzed, and the fault signals are analyzed by filtering bath, spectrum analysis and axis locus analysis. The singular value spectral entropy in time domain, power spectrum entropy in frequency domain, wavelet spectrum entropy and wavelet space spectral entropy in time-frequency domain are analyzed. The entropy band range of four kinds of fault signals is calculated, and the conventional fault diagnosis method based on information entropy is discussed. 2) there is data redundancy in using entropy band directly as SVM training sample and test sample. Therefore, based on the entropy band data, the data preprocessing for SVM training samples is studied. It includes sample normalization and principal component feature extraction. The subsequent experiments show that the processed entropy band data can not only reflect the characteristics of vibration signals, but also be suitable for SVM model training and fault classification. 3) the goal of constructing an optimal classifier is to construct an optimal classifier. The influence of PSO algorithm and GA algorithm on the classification accuracy is studied systematically after optimizing the SVM parameters. By inputting the processed data into SVM, applying GA and PSO to optimize the kernel parameters and penalty factors of SVM, and testing the samples of unknown fault categories, it is found that the SVM classification performance after GA optimization is poor, and the model training time is longer. However, the SVM optimized by PSO has good classification accuracy and fast training time. 4) because this study is a multi-fault classification problem, SVM is a two-classifier. Therefore, based on one-to-many method, SVM multi-fault classifier which can separate four kinds of faults is designed. The PSO algorithm is used to optimize the parameters of each classifier. Based on the above algorithm flow, a rotor fault diagnosis system based on MATLAB GUI is developed. Subsystem one can realize vibration signal analysis, spectrum analysis, axis trajectory analysis and so on. The second subsystem can optimize the classifier parameters according to the characteristics of the samples and realize the identification of unknown faults. The experimental results show that the system is effective.
【学位授予单位】：兰州理工大学
【学位级别】：硕士
【学位授予年份】：2011
【分类号】：TH165.3
本文编号：2268118
[Abstract]:In recent years, more and more attention has been paid to the research of equipment condition monitoring and fault diagnosis, and the related theoretical research has been developed rapidly. Support vector machines (SVM) show good performance in solving classification problems based on small samples. According to the principle of structural risk minimization, it has the global optimal solution. According to the limited sample information, it seeks the best tradeoff between the complexity of the model and the learning ability in order to obtain the best generalization ability and solve the "overlearning" problem effectively. In this paper, the entropy band method is used to extract the characteristic of the fault vibration signal in combination with the common faults simulated on the rotor test bench. In order to make SVM have higher classification accuracy, particle ant colony algorithm is used to optimize the parameters of SVM. The main contents and conclusions of this paper are as follows: 1) four kinds of typical faults are simulated on the rotor test bench. The mechanism of four kinds of faults is analyzed, and the fault signals are analyzed by filtering bath, spectrum analysis and axis locus analysis. The singular value spectral entropy in time domain, power spectrum entropy in frequency domain, wavelet spectrum entropy and wavelet space spectral entropy in time-frequency domain are analyzed. The entropy band range of four kinds of fault signals is calculated, and the conventional fault diagnosis method based on information entropy is discussed. 2) there is data redundancy in using entropy band directly as SVM training sample and test sample. Therefore, based on the entropy band data, the data preprocessing for SVM training samples is studied. It includes sample normalization and principal component feature extraction. The subsequent experiments show that the processed entropy band data can not only reflect the characteristics of vibration signals, but also be suitable for SVM model training and fault classification. 3) the goal of constructing an optimal classifier is to construct an optimal classifier. The influence of PSO algorithm and GA algorithm on the classification accuracy is studied systematically after optimizing the SVM parameters. By inputting the processed data into SVM, applying GA and PSO to optimize the kernel parameters and penalty factors of SVM, and testing the samples of unknown fault categories, it is found that the SVM classification performance after GA optimization is poor, and the model training time is longer. However, the SVM optimized by PSO has good classification accuracy and fast training time. 4) because this study is a multi-fault classification problem, SVM is a two-classifier. Therefore, based on one-to-many method, SVM multi-fault classifier which can separate four kinds of faults is designed. The PSO algorithm is used to optimize the parameters of each classifier. Based on the above algorithm flow, a rotor fault diagnosis system based on MATLAB GUI is developed. Subsystem one can realize vibration signal analysis, spectrum analysis, axis trajectory analysis and so on. The second subsystem can optimize the classifier parameters according to the characteristics of the samples and realize the identification of unknown faults. The experimental results show that the system is effective.
【学位授予单位】：兰州理工大学
【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2条
1 许红波;基于环境参数的过渡环境下人体热感觉预测[D];大连理工大学;2011年
2 胡常安;基于混合杂草算法—神经网络的转子故障数据分类方法研究[D];兰州理工大学;2012年
，本文编号：2268118
本文链接：https://www.wllwen.com/kejilunwen/jixiegongcheng/2268118.html
