基于模糊聚类的故障诊断技术研究

发布时间：2018-06-28 09:41

本文选题：模糊聚类 + 故障诊断　；参考：《南京航空航天大学》2012年硕士论文

【摘要】：基于模糊聚类的故障诊断技术是一类十分重要的故障诊断技术，在对复杂庞大的系统进行故障诊断时有着独特的优势，对系统先验知识的需求较少，不需要精确的数学解析模型，可以从大量的系统监控数据中获取系统运行模式信息。在工业生产流程和机器设备日趋复杂的今天，研究基于模糊聚类的故障诊断技术有着十分重要的理论意义和工程应用价值。基于模糊聚类的故障诊断技术一直是相关领域内的热点研究对象，从诞生之日起已经涌现出了很多不同的聚类算法和诊断方法。但是在未知故障的诊断问题上的研究一直比较薄弱，并没有形成成熟的方法和共识。本文便从未知故障的诊断问题出发，研究如何隔离未知故障与已知故障及如何隔离同一类型故障的不同强度，并在模糊聚类算法和在线诊断方案两个方面进行了深入的理论分析和大量的实验验证，分别提出了改进型可能性GK聚类算法（IPGK）和基于故障向量的在线诊断方案，通过这两种新方法的结合，较好的解决了未知故障的诊断问题。本文的主要内容如下： 1、回顾故障诊断技术的发展历程，分类介绍了几种常见的故障诊断方法。着重总结了与基于模糊聚类的故障诊断技术相关的研究成果、研究文献、基础理论和技术方法，阐述了模糊聚类、模式识别和故障诊断之间的紧密联系，对基于模糊聚类的故障诊断技术涉及到的几个关键词进行了解释。 2、简单介绍了模糊数学的发展历程，介绍了模糊聚类技术的发展历程和理论背景，从理论分析和实验验证两方面研究了应用最为广泛的模糊c-均值聚类（FCM）算法，结果表明未知故障的诊断需要聚类算法能够检测超椭球体或超线性分布的数据，，并具有适合检测孤立点的特性。根据这一具体需求，针对FCM算法、可能性c-均值聚类（PCM）算法、改进型可能性c-均值聚类（IPCM）算法三种有代表性的成熟算法各自的特点和不足，进行算法的研究和改进工作，提出了基于马氏距离的改进型可能性GK聚类算法(IPGK)。通过仿真实验说明，此算法能较好地处理超椭球体或超线性分布的数据，并具有适合检测孤立点和诊断未知故障的特性。 3、通过理论分析和仿真实验指出，常见的在线诊断方法无法隔离同一类型故障的不同强度，会导致错误诊断。在IPGK算法的基础上进行简单修改即可得到单一类别数据的聚类计算方法，实现了在线数据聚类中心的计算。引入方向残差的概念得到了故障强度的计算方法和故障数据聚类中心的分布规律，并在模糊聚类领域内将方向残差引申为故障向量，提出了基于故障向量的在线诊断方法。通过实验说明这一在线诊断方法可以对不同强度下的同一故障进行隔离。 4、将IPGK算法和基于故障向量的在线诊断方案结合在一起形成基于模糊聚类的故障诊断流程，并在TE工业过程和近地空间飞行器仿真模型上进行了诊断实验。仿真结果表明，相对于现有的模糊聚类故障诊断技术，本文提出的方法可以较好对未知故障和已知故障进行隔离，也可以较好的对同一类型故障的不同强度进行隔离，并具有一定的抗干扰能力。
[Abstract]:The fault diagnosis technology based on fuzzy clustering is a kind of very important fault diagnosis technology. It has a unique advantage in the fault diagnosis of complex and huge systems. It needs less prior knowledge of the system and does not need accurate mathematical analytic model. It can obtain system operation mode information from a large number of system monitoring data. With the increasing complexity of industrial production process and machine equipment, it is of great theoretical significance and engineering application value to study the fault diagnosis technology based on fuzzy clustering.
The fault diagnosis technology based on fuzzy clustering has always been a hot research object in the related fields, and many different clustering algorithms and diagnostic methods have emerged from the date of birth. However, the research on the diagnosis of unknown faults has been relatively weak, and does not form a mature method and consensus. This paper is from the unknown fault. On the basis of the diagnosis, we study how to isolate the different strengths of the unknown fault and the known fault and how to isolate the same type of fault. In the two aspects, the fuzzy clustering algorithm and the online diagnosis scheme have been analyzed in depth, and a large number of experimental verification are carried out. The improved probability GK clustering algorithm (IPGK) and fault vector based on the fault vector are proposed. Based on the combination of the two new methods, the online diagnosis scheme solves the problem of unknown fault diagnosis.
1, review the development process of fault diagnosis technology, classify several common fault diagnosis methods, summarize the research results related to the fault diagnosis technology based on fuzzy clustering, study literature, basic theory and technical method, and elaborate the close connection between fuzzy clustering, pattern recognition and fault diagnosis. Several key words involved in clustering fault diagnosis technology are explained.
2, the development course of fuzzy mathematics is briefly introduced, the development course and the theoretical background of fuzzy clustering are introduced. The most widely used fuzzy c- mean clustering (FCM) clustering algorithm is studied from two aspects of theoretical analysis and experimental verification. The result shows that the diagnosis of unknown fault needs the clustering algorithm to detect the superellipsoid or superlinear distribution. According to this specific requirement, according to this specific requirement, the characteristics and shortcomings of three representative mature algorithms of the FCM algorithm, the possibility c- mean clustering (PCM) algorithm and the improved probability c- mean clustering (IPCM) algorithm are studied and improved, and the Mahalanobis distance based modification is proposed. The progressive possibility GK clustering algorithm (IPGK). The simulation experiment shows that the algorithm can handle the data of superellipsoid or superlinear distribution better, and it has the characteristic of detecting the outlier and diagnosing the unknown fault.
3, through theoretical analysis and simulation experiments, it is pointed out that the common on-line diagnosis method can not isolate the different intensity of the same type of fault and can lead to the error diagnosis. The method of clustering calculation of single category data can be obtained by simple modification on the basis of the IPGK algorithm, and the calculation of the online data clustering center is realized. The probability of the residual error is introduced. The calculation method of the fault strength and the distribution of the cluster center of the fault data are read out, and the residual error of the direction is extended into the fault vector in the fuzzy clustering field. An on-line diagnosis method based on the fault vector is proposed. The experiment shows that the on-line diagnosis method can isolate the same fault under the same intensity.
4, the IPGK algorithm and the fault vector based online diagnosis scheme are combined to form a fault diagnosis process based on fuzzy clustering, and the diagnosis experiments are carried out on the TE industrial process and the near earth space vehicle simulation model. The simulation results show that the method proposed in this paper can be compared with the existing fuzzy clustering fault diagnosis technology. Good isolation of unknown faults and known faults can also better isolate different strengths of the same type of fault and have a certain anti-interference capability.
【学位授予单位】：南京航空航天大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TH165.3

【参考文献】