基于复杂网络社团聚类的机械故障诊断方法及其应用研究

发布时间：2018-06-14 15:15

本文选题：复杂网络 + 社团聚类　；参考：《湖南科技大学》2014年硕士论文

【摘要】：随着工业技术的进步，大型复杂机械正朝着大型化、复杂化、集成化发展，设备一旦发生重大故障将严重影响工业生产，造成重大经济损失。因此，对大型复杂机械进行准确故障诊断，确保机械设备安全准确运行是当前机械故障诊断领域的研究热点之一。复杂网络是近年兴起的新的研究方法，是一种用来描述复杂系统的重要模型和工具。它将系统中的元素视为网络节点，节点之间的连接边表示元素之间的关系，通过对节点与边的分析挖掘网络中的自组织、自相似、小世界等特性，其中小世界性表现为节点相互连接组成的小集合，这些集合内部连接紧密，与集合外部连接较少，，这种特性又称为社团特性，这正好与故障诊断领域同类故障样本之间联系紧密、不同故障样本之间联系稀疏的特性相对应。论文将故障样本视为复杂网络中的节点，建立故障样本复杂网络模型，并开展社团聚类诊断分析。主要工作如下：（1）开展了故障样本复杂网络社团特性分析，确定相似度函数、网络边权、边阈值等因数，建立故障样本网络模型；研究了基于互信息评价网络模型中重要节点的计算方法，并验证了该方法较其他方法的优越性。（2）研究了基于复杂网络社团聚类的故障诊断方法。应用复杂网络社团结构特性，将网络划分为若干个社团，利用社团模块性合并指标变化开展社团聚类，最后合并的社团对应不同故障类型，实现诊断；以滚动轴承故障诊断实例验证了方法的有效性。（3）研究了基于复杂网络社团聚类的改进K-means聚类诊断方法。针对K-means聚类算法依赖于初始聚类数K值和初始聚类中心的不足，利用复杂网络社团聚类为K-means聚类算法确定K值，通过计算网络节点关联度选取重要节点作为初始聚类中心，开展聚类诊断；有效克服了K-means聚类算法中K值和初始聚类中心选取困难的问题；并以滚动轴承故障诊断实例，验证了该方法的有效性。（4）研究了基于复杂网络社团聚类的复合故障特征分离方法。应用经验模态分解将复合故障信号分解为若干个不同频段的IMF分量，将每个IMF分量视为网络中社团，进行同类社团合并，最后得到对应不同单一故障的各个社团，实现复合故障的有效分离。以转子不平衡和轴承内圈复合故障分离实例、轴承内圈和滚珠复合故障分离实例，验证了方法的有效性。
[Abstract]:With the progress of industrial technology, large and complex machinery is facing large, complicated and integrated development. Once the equipment has serious failure, it will seriously affect industrial production and cause significant economic loss. Therefore, accurate fault diagnosis of large and complex machinery and ensuring the safe and accurate operation of mechanical equipment are the current field of mechanical fault diagnosis. Complex network is a new research method in recent years. It is an important model and tool used to describe complex systems. It treats the elements in the system as network nodes, the connections between nodes represent the relationship between elements and the self organization, self similarity and small world in the network through the analysis of the nodes and edges. The small world is characterized by small world representation of small sets of nodes connected to each other. These sets are closely connected inside and are less connected to the set. This feature is also called community characteristics, which is closely related to the similar fault samples in the fault diagnosis field. Fault samples are considered as nodes in complex networks. A complex network model of fault samples is established, and community clustering diagnosis analysis is carried out.
(1) carry out the analysis of the community characteristics of the complex network of fault samples, determine the similarity function, the network edge weight, edge threshold and other factors, establish the fault sample network model, study the calculation method of the important nodes in the mutual information evaluation network model, and verify the superiority of the method compared with the other methods.
(2) the fault diagnosis method based on complex network community clustering is studied. Using the complex network community structure characteristics, the network is divided into several societies, and the community clustering is carried out by the changes of the association modular merging index. The final merged community corresponds to the different fault types, and the diagnosis is verified by the rolling bearing fault diagnosis example. The validity of the law.
(3) the improved K-means clustering diagnosis method based on complex network community clustering is studied. The K-means clustering algorithm relies on the initial clustering number K value and the shortage of the initial cluster center. The complex network community clustering is used to determine the K value by the K-means clustering algorithm, and the important node is selected as the initial cluster center through the calculation of the node association degree of the network. The clustering diagnosis is carried out, and the problem of selecting the K value and the initial cluster center in the K-means clustering algorithm is effectively overcome, and the validity of the method is verified by a fault diagnosis example of rolling bearing.
(4) the complex fault feature separation method based on complex network community clustering is studied. The complex fault signal is decomposed into IMF components of several different frequency bands by using empirical mode decomposition. Each IMF component is regarded as a community in the network, and the similar associations are merged. Finally, a complex fault is obtained to achieve a complex fault. The effective separation of the rotor imbalance and bearing inner ring compound fault is illustrated. An example of bearing inner ring and ball compound fault separation is used to verify the effectiveness of the method.
【学位授予单位】：湖南科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TH165.3

【参考文献】