基于小波分析与神经网络滚动轴承故障诊断方法的研究
发布时间：2018-09-12 06:48
【摘要】：滚动轴承是旋转机械中重要的零部件之一,但由于加工工艺、工作环境等原因造成损坏率高、寿命的随机性较大。旋转机械故障种类繁多,但由滚动轴承的故障引起的大约占三分之一,所以掌握滚动轴承的工作状态以及故障的形成和发展,是目前机械故障诊断领域中所研究的重要课题之一 本论文通过分析滚动轴承振动机理、失效原因和信号特征,对轴承振动信号的采集方法进行了改进,采用无线传感器网络技术降低故障诊断系统的复杂性、提升诊断系统的效率。利用滚动轴承振动信号实现其故障检测与诊断,目前主要有机理分析和智能诊断两条途径。机理分析常用方法有随机共振和小波分析等；智能诊断常用方法有神经网络和支持向量机等。但以上各方法在实际应用中均存在其不足之处,从而影响到轴承故障检测与诊断的效果。为此,本文认为非常有必要立足于不断发展的新理论和新方法,紧紧围绕滚动轴承故障机理分析与智能诊断现有方法存在的问题与不足展开研究与探讨。 (1)针对传统有线传感器网络信息采集灵活性差、故障率高的问题,本文在分析滚动轴承振动机理、失效原因和振动信号特征的基础上,设计了滚动轴承振动信号无线采集网络,以802.15.4和ZigBee协议为标准,采用250kbps(?)勺传输速率和无线部署的方式,降低系统复杂性和故障率,为后续轴承故障诊断方法提供基础原理性的支持。 (2)针对噪声较强有用信号较弱环境下的轴承故障问题,研究了一种基于遗传免疫优化粒子群算法的随机共振方法。该方法不仅实现了强噪声背景下的微弱信号提取,而且解决了基本随机共振理论只能处理微弱的小参数信号、不能处理轴承振动这类大参数信号问题。通过展开深入的研究,提出了一种基于遗传免疫的粒子群优化算法,并将其应用于随机共振的关键参数寻优过程,由此进一步提出了基于遗传免疫粒子群优化的自适应随机共振算法,并采用轴承故障实验数据进行了分析与验证。 (3)针对小波理论实际应用过程中存在难以构造理想小波基函数的问题,研究了基于第二代小波变换的滚动轴承故障诊断方法。该方法利用第二代小波变换将滚动轴承故障振动信号分解到不同尺度,提取出共振频带,然后利用Hilbert变换进行解调,再对解调后的信号进行频谱分析得到小波包络谱,从包络谱上获取轴承故障特征信息。通过轴承实验数据应用与分析表明,该方法准确地提取了滚动轴承不同损伤程度故障的特征频率,实现了轴承故障的定量诊断。 (4)针对神经网络本身性能难以继续提高的问题,研究了基于第二代小波变换与神经网络的滚动轴承智能诊断方法。本文从提高神经网络输入端的信号质量入手,利用第二代小波变换与特征评估方法,提出了一种基于第二代小波与神经网络相结合的滚动轴承智能诊断模型,并将该模型应用于实验分析与工程实践中。结果表明从第二代小波分解后信号中提取的联合特征能够揭示更多的故障信息；特征评估方法能够针对诊断对象的健康状态分类选择其相应的敏感特征,大大提高了BP神经网络分类的准确率,验证了本文所建立的智能诊断模型的有效性。 (5)针对滚动轴承故障属于典型小样本的特征,研究了基于参数优化支持向量机的滚动轴承智能诊断方法。基本支持向量机方法存在模型参数不易合理选取而影响到算法性能的问题,本文在详细分析各参数对分类模型的影响的基础上,建立了参数优化模型,并采用遗传免疫粒子群算法作为优化方法,建立了基于遗传免疫粒子群和支持向量机的智能诊断模型,最后将该模型用于轴承故障诊断中。结果表明,该算法不但实现了对SVM分类模型参数的自动优化,提高了SVM分类模型的故障诊断精度,而且对分散程度较大、聚类性较差的故障样本分类有较强的适用性。 通过论文上述内容研究,优化了目前的应用于滚动轴承不同故障条件下的诊断算法,并进行了实验验证,为旋转机械的故障诊断提供了新方向。
[Abstract]:Rolling bearing is one of the most important parts in rotating machinery, but because of the processing technology, working environment and other reasons, the damage rate is high, and the life of the randomness is large. Exhibition is one of the most important topics in the field of mechanical fault diagnosis.
In this paper, by analyzing the vibration mechanism, failure reasons and signal characteristics of rolling bearings, the acquisition method of bearing vibration signals is improved. The wireless sensor network technology is used to reduce the complexity of fault diagnosis system and improve the efficiency of diagnosis system. There are two ways of mechanism analysis and intelligent diagnosis.The common methods of mechanism analysis are stochastic resonance and wavelet analysis,and the common methods of intelligent diagnosis are neural network and support vector machine.But these methods have their shortcomings in practical application,which affect the effect of bearing fault detection and diagnosis. It is often necessary to study and discuss the problems and shortcomings of the existing methods of fault mechanism analysis and intelligent diagnosis of rolling bearings based on the new theories and methods which are constantly developing.
(1) Aiming at the problem of poor flexibility and high failure rate in traditional wired sensor networks, this paper designs a wireless vibration signal acquisition network for rolling bearings based on the analysis of vibration mechanism, failure reasons and vibration signal characteristics of rolling bearings. The network adopts 250kbps (?) spoon transmission rate and wireless transmission rate with 802.15.4 and ZigBee protocol as the standard. The deployment method can reduce the complexity and failure rate of the system and provide the basic principle support for subsequent bearing fault diagnosis methods.
(2) A stochastic resonance (SR) method based on genetic immune optimization particle swarm optimization (GAIMO-PSO) is proposed to solve the bearing faults in the environment of strong noise and weak useful signals. A genetic immune particle swarm optimization algorithm based on genetic immune is proposed and applied to the optimization process of the key parameters of stochastic resonance. Data are analyzed and verified.
(3) Aiming at the problem that it is difficult to construct an ideal wavelet basis function in the practical application of wavelet theory, a fault diagnosis method of rolling bearing based on the second generation wavelet transform is studied. After demodulation, wavelet envelope spectrum is obtained by spectrum analysis of demodulated signal, and the characteristic information of bearing fault is obtained from envelope spectrum. The application and analysis of bearing experimental data show that this method can accurately extract the characteristic frequency of rolling bearing fault with different degree of damage and realize the quantitative diagnosis of bearing fault.
(4) Aiming at the problem that it is difficult to improve the performance of neural network, an intelligent diagnosis method of rolling bearing based on second generation wavelet transform and neural network is studied. The results show that the combined features extracted from the second generation wavelet decomposition signal can reveal more fault information, and the feature evaluation method can select the corresponding sensitive features according to the health status classification of the diagnosis object. It greatly improves the accuracy of BP neural network classification and verifies the validity of the intelligent diagnosis model.
(5) Aiming at the characteristics that rolling bearing faults belong to typical small samples, the intelligent diagnosis method of rolling bearing based on parameter optimization support vector machine is studied. An intelligent diagnosis model based on genetic immune particle swarm optimization and support vector machine is established. Finally, the model is applied to bearing fault diagnosis. The results show that the algorithm not only optimizes the parameters of SVM classification model automatically, but also improves the SVM score. The fault diagnosis accuracy of the classification model is high, and it has a strong applicability to the classification of fault samples with large dispersion and poor clustering.
Based on the above research, the current diagnosis algorithm for different fault conditions of rolling bearings is optimized and verified by experiments, which provides a new direction for fault diagnosis of rotating machinery.
【学位授予单位】：东北林业大学
【学位级别】：博士
【学位授予年份】：2013
【分类号】：TH133.33;TH165.3
[Abstract]:Rolling bearing is one of the most important parts in rotating machinery, but because of the processing technology, working environment and other reasons, the damage rate is high, and the life of the randomness is large. Exhibition is one of the most important topics in the field of mechanical fault diagnosis.
In this paper, by analyzing the vibration mechanism, failure reasons and signal characteristics of rolling bearings, the acquisition method of bearing vibration signals is improved. The wireless sensor network technology is used to reduce the complexity of fault diagnosis system and improve the efficiency of diagnosis system. There are two ways of mechanism analysis and intelligent diagnosis.The common methods of mechanism analysis are stochastic resonance and wavelet analysis,and the common methods of intelligent diagnosis are neural network and support vector machine.But these methods have their shortcomings in practical application,which affect the effect of bearing fault detection and diagnosis. It is often necessary to study and discuss the problems and shortcomings of the existing methods of fault mechanism analysis and intelligent diagnosis of rolling bearings based on the new theories and methods which are constantly developing.
(1) Aiming at the problem of poor flexibility and high failure rate in traditional wired sensor networks, this paper designs a wireless vibration signal acquisition network for rolling bearings based on the analysis of vibration mechanism, failure reasons and vibration signal characteristics of rolling bearings. The network adopts 250kbps (?) spoon transmission rate and wireless transmission rate with 802.15.4 and ZigBee protocol as the standard. The deployment method can reduce the complexity and failure rate of the system and provide the basic principle support for subsequent bearing fault diagnosis methods.
(2) A stochastic resonance (SR) method based on genetic immune optimization particle swarm optimization (GAIMO-PSO) is proposed to solve the bearing faults in the environment of strong noise and weak useful signals. A genetic immune particle swarm optimization algorithm based on genetic immune is proposed and applied to the optimization process of the key parameters of stochastic resonance. Data are analyzed and verified.
(3) Aiming at the problem that it is difficult to construct an ideal wavelet basis function in the practical application of wavelet theory, a fault diagnosis method of rolling bearing based on the second generation wavelet transform is studied. After demodulation, wavelet envelope spectrum is obtained by spectrum analysis of demodulated signal, and the characteristic information of bearing fault is obtained from envelope spectrum. The application and analysis of bearing experimental data show that this method can accurately extract the characteristic frequency of rolling bearing fault with different degree of damage and realize the quantitative diagnosis of bearing fault.
(4) Aiming at the problem that it is difficult to improve the performance of neural network, an intelligent diagnosis method of rolling bearing based on second generation wavelet transform and neural network is studied. The results show that the combined features extracted from the second generation wavelet decomposition signal can reveal more fault information, and the feature evaluation method can select the corresponding sensitive features according to the health status classification of the diagnosis object. It greatly improves the accuracy of BP neural network classification and verifies the validity of the intelligent diagnosis model.
(5) Aiming at the characteristics that rolling bearing faults belong to typical small samples, the intelligent diagnosis method of rolling bearing based on parameter optimization support vector machine is studied. An intelligent diagnosis model based on genetic immune particle swarm optimization and support vector machine is established. Finally, the model is applied to bearing fault diagnosis. The results show that the algorithm not only optimizes the parameters of SVM classification model automatically, but also improves the SVM score. The fault diagnosis accuracy of the classification model is high, and it has a strong applicability to the classification of fault samples with large dispersion and poor clustering.
Based on the above research, the current diagnosis algorithm for different fault conditions of rolling bearings is optimized and verified by experiments, which provides a new direction for fault diagnosis of rotating machinery.
【学位授予单位】：东北林业大学
【学位级别】：博士
【学位授予年份】：2013
【分类号】：TH133.33;TH165.3
【相似文献】
相关期刊论文 前10条
1 庄哲民,林志强;基于神经网络的滚动轴承检测[J];仪器仪表学报;2000年01期
2 刘良顺;魏立东;宋希庚;薛冬新;;基于RBF神经网络的滚动轴承故障诊断方法[J];农业机械学报;2006年03期
3 刘乐平;林凤涛;;基于小波包特征向量与神经网络的滚动轴承故障诊断[J];轴承;2008年04期
4 周瑞峰;张志新;王亮;;滚动轴承故障智能诊断技术的研究[J];化工装备技术;2009年02期
5 王利英;;小波包特征熵-神经网络在轴承故障诊断中的应用[J];河北工程大学学报(自然科学版);2008年01期
6 唐茗;李力;;滚动轴承故障程度识别与诊断研究[J];轴承;2008年08期
7 骆志明,冯庚斌;遗传BP网络在滚动轴承故障诊断中的应用[J];铁道学报;1999年05期
8 刘建文;傅攀;任s，
本文编号：2238223
本文链接：https://www.wllwen.com/kejilunwen/jixiegongcheng/2238223.html
