基于小波包和模糊神经网络电机故障诊断研究
发布时间：2018-09-19 13:00
【摘要】：异步电动机有着简单的结构,在实际应用中价格也并不昂贵,较其他产品运行相对稳定,因此在日常生产生活中有着举足轻重的作用。由于机械负载、工作环境等各方面的影响,电机出现故障使其运行不稳定形成经济等方面损失,乃至危及人身安全。因此,人们在研究异步电机的同时投入了大量精力去研究其故障诊断。随着计算机被广泛的应用到信号处理中,电机的故障诊断方法也在人们的不断努力下取得了很大的突破。本文在大量的浏览中英文文献的基础上,将小波包与模糊RBF神经网络相结合,并运用遗传算法进行优化的方法,研究了异步电机故障的诊断方法。振动信号能很全面的反映出电机的运行状态,小波包分析具有很高的分辨率等特点,本文在小波包能量故障特征提取的基础上主要研究了以下内容:研究了信号的能量特征提取,选取经验上常用的几种小波包基,分别用来对振动信号进行分析,将幅值与能量做除法,选取商值最大的db3,用其对信号进行三层小波包分解,即可得到能量特征向量,不同的振动信号究其根本是其振动能量的不同,振动信号通过小波包分解,各层信号系数功率谱也有所不同,根据系数功率谱判断不同的故障。研究了小波包与模糊RBF神经网络相结合的异步电机故障诊断方法。该方法先利用小波包分析处理振动信号并提取能量特征向量分别作为训练样本和检测样本,将故障的程度看作一个模糊的概念并对输入信号进行模糊化。隐含层节点数通过一些经验公式可以得到,并逐个进行训练,选择输出误差较小训练时间短的节点数。由于网络的参数不容易确定,将三种常用的基函数中心选取方法相比较最终选择有监督法并结合梯度下降法。用训练样本对该网络训练,完成后,将检测样本输入对其做检验。研究了小波包与遗传优化模糊RBF神经网络相结合的异步电机故障诊断方法。为了解决梯度下降法选取基函数中心时容易陷入极小值的问题,将上面的算法与遗传算法结合到一起,网络结构相同,隐含层节点数经过选择后取值比较少,在此基础上运用了遗传算法与梯度下降法优化基函数中心等网络参数,避免了出现极小值,减少了隐含层节点数,缩短了收敛时间,准确度也有所提升。
[Abstract]:Asynchronous motor has a simple structure and is not expensive in practical application. It is relatively stable compared with other products, so it plays an important role in daily production and life. Because of the influence of mechanical load, working environment and so on, the failure of motor makes its operation unstable and result in economic loss, even endangering personal safety. Therefore, people devote a lot of energy to the fault diagnosis of asynchronous motor at the same time. With the wide application of computer in signal processing, the fault diagnosis method of motor has made a great breakthrough with the continuous efforts of people. Based on a large amount of literature in Chinese and English, this paper combines wavelet packet with fuzzy RBF neural network, and studies the fault diagnosis method of asynchronous motor using genetic algorithm. The vibration signal can fully reflect the running state of the motor, and wavelet packet analysis has the characteristics of high resolution, etc. Based on the energy fault feature extraction of wavelet packet, the following contents are studied in this paper: the energy feature extraction of signal is studied, and several kinds of wavelet packet bases, which are commonly used in experience, are selected to analyze the vibration signal respectively. The amplitude and energy are divided and the db3, with the largest quotient is selected to decompose the signal into three layers of wavelet packet. The energy eigenvector can be obtained. The different vibration signal is based on the difference of its vibration energy, and the vibration signal is decomposed by wavelet packet. The power spectrum of signal coefficients varies from layer to layer, and different faults are judged according to the power spectrum of coefficients. The fault diagnosis method of asynchronous motor based on wavelet packet and fuzzy RBF neural network is studied. In this method, the vibration signal is processed by wavelet packet analysis and the energy eigenvector is extracted as the training sample and the detection sample respectively. The degree of fault is regarded as a fuzzy concept and the input signal is blurred. The number of hidden layer nodes can be obtained by some empirical formulas and trained one by one to select the number of nodes with smaller output error and shorter training time. Because the parameters of the network are not easy to determine, the three commonly used methods of selecting the basis function center are compared with the supervised method and the gradient descent method. The network is trained with training samples. The fault diagnosis method of asynchronous motor based on wavelet packet and genetic optimization fuzzy RBF neural network is studied. In order to solve the problem that the gradient descent method is easy to fall into the minimum value when selecting the center of the basis function, the above algorithm is combined with the genetic algorithm. The network structure is the same, and the number of hidden layer nodes is less after selection. On this basis, genetic algorithm and gradient descent method are used to optimize the network parameters such as the center of the basis function. The minimum value is avoided, the number of hidden layer nodes is reduced, the convergence time is shortened, and the accuracy is improved.
【学位授予单位】：东北石油大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TM343;TP183
本文编号：2250198
[Abstract]:Asynchronous motor has a simple structure and is not expensive in practical application. It is relatively stable compared with other products, so it plays an important role in daily production and life. Because of the influence of mechanical load, working environment and so on, the failure of motor makes its operation unstable and result in economic loss, even endangering personal safety. Therefore, people devote a lot of energy to the fault diagnosis of asynchronous motor at the same time. With the wide application of computer in signal processing, the fault diagnosis method of motor has made a great breakthrough with the continuous efforts of people. Based on a large amount of literature in Chinese and English, this paper combines wavelet packet with fuzzy RBF neural network, and studies the fault diagnosis method of asynchronous motor using genetic algorithm. The vibration signal can fully reflect the running state of the motor, and wavelet packet analysis has the characteristics of high resolution, etc. Based on the energy fault feature extraction of wavelet packet, the following contents are studied in this paper: the energy feature extraction of signal is studied, and several kinds of wavelet packet bases, which are commonly used in experience, are selected to analyze the vibration signal respectively. The amplitude and energy are divided and the db3, with the largest quotient is selected to decompose the signal into three layers of wavelet packet. The energy eigenvector can be obtained. The different vibration signal is based on the difference of its vibration energy, and the vibration signal is decomposed by wavelet packet. The power spectrum of signal coefficients varies from layer to layer, and different faults are judged according to the power spectrum of coefficients. The fault diagnosis method of asynchronous motor based on wavelet packet and fuzzy RBF neural network is studied. In this method, the vibration signal is processed by wavelet packet analysis and the energy eigenvector is extracted as the training sample and the detection sample respectively. The degree of fault is regarded as a fuzzy concept and the input signal is blurred. The number of hidden layer nodes can be obtained by some empirical formulas and trained one by one to select the number of nodes with smaller output error and shorter training time. Because the parameters of the network are not easy to determine, the three commonly used methods of selecting the basis function center are compared with the supervised method and the gradient descent method. The network is trained with training samples. The fault diagnosis method of asynchronous motor based on wavelet packet and genetic optimization fuzzy RBF neural network is studied. In order to solve the problem that the gradient descent method is easy to fall into the minimum value when selecting the center of the basis function, the above algorithm is combined with the genetic algorithm. The network structure is the same, and the number of hidden layer nodes is less after selection. On this basis, genetic algorithm and gradient descent method are used to optimize the network parameters such as the center of the basis function. The minimum value is avoided, the number of hidden layer nodes is reduced, the convergence time is shortened, and the accuracy is improved.
【学位授予单位】：东北石油大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TM343;TP183
【参考文献】
相关期刊论文 前10条
1 曹志彤,陈宏平,E．Ritchie;基于混沌神经网络动态联想记忆的电机故障诊断[J];电工技术学报;2000年03期
2 黄迪山;孙月明;童忠钫;;希尔伯特变换在故障诊断中的应用[J];中国纺织大学学报;1991年06期
3 刘劲，吴小辰，孙扬声，陈德树;电力系统静态失稳和周期振荡的局部分叉分析[J];电力系统自动化;1995年12期
4 陈书旺;张喜英;;模糊理论在RBF神经网络中的应用[J];电脑知识与技术;2008年04期
5 李占锋,韩芳芳,郑德忠;基于BP神经网络的电机转子故障诊断的研究[J];河北科技大学学报;2001年03期
6 赵林桂,詹渝明,陈明涛;基于傅里叶变换的异步电机转子故障检测[J];东北电力技术;2004年07期
7 孙丽颖,屈丹,闫钿;傅里叶变换与小波变换在信号故障诊断中的应用[J];辽宁工学院学报;2005年03期
8 乔俊飞;韩红桂;;RBF神经网络的结构动态优化设计[J];自动化学报;2010年06期
9 李春华;荣明星;;基于小波包和改进BP神经网络算法的电机故障诊断[J];现代电子技术;2013年15期
10 赵飞鹏;沈久珩;;机械设备的状态监测与故障诊断 第二讲 振动信号分析[J];有色设备;1990年02期
，本文编号：2250198
本文链接：https://www.wllwen.com/kejilunwen/dianlilw/2250198.html
