适于系统非稳定工况的故障检测与分类诊断方法研究

The non-steady condition refers to such working conditions as the system is not working steadily during the startup, braking and the mutations of other movement processes. Actually, the failure operations of the system under the non-steady condition is relatively higher than the system under the steady condition, especially when it is difficult to monitor the system under the non-steady condition and when the expertise help is not timely and handy. Consequently, this has already caught much attention of the researchers in this field.. There are mass variables, great changes of the variables in amplitude, higher response speed, more complex correlation relationship among many variables in the system under the non-steady condition. The quantitative model could not be established without the equation of motion or the assistance from expert system under the non-steady condition in particular. Furthermore, the methods for fault detection and diagnosis under the steady condition could barely be detected effectively with the consideration of the system under the non-steady condition. Therefore, the program intends to discovery useful knowledge in database first, and then to find research methods applicable for the fault detection and classification of the diagnosis of the system under the non-steady condition: .1).To induce and deduce the feature attributes and universal laws of the system under the non-steady condition;.2).To build the new fault detection limit suitable for the system under the non-steady condition according to the traditional methods, and then propose some new fault detection methods; .3).To introduce the idea of the Effective Principal Component, and combining with the spatial and temporal clustering analysis methods to develop the failure diagnosis methods according to the characteristic properties and laws of the non-steady condition under the non-steady condition. .The research achievements of this program will provide a theoretical basis and some reference data for the application research of the system under the non-steady condition, and therefore it possess some positive significance for the KDD (Knowledge Discovery in Database) field.

系统在起动、制动以及运动过程突变等状态下的工作状况称为非稳定工况。由于非稳定工况下故障率相对较高，监测困难，相关的技术研究也日益受到关注。在系统非稳定工况下，由于变量多、变化幅度大、响应速度快、变量间相关性关系复杂等，特别是当系统运动方程未知，又得不到专家系统的知识援助时，无法建立定量模型。并且稳定工况的方法难以对系统非稳定工况进行有效的检测和诊断。.因此，课题组拟从海量信息中获得知识的思路出发，开展适于系统非稳定工况的故障检测和分类诊断方法的研究:1)归纳演绎系统非稳定工况的特征属性和普适性规律；2)建立传统方法在系统非稳定工况的故障检测控制限，进而研究新的检测方法；3)引入有效主元的思想，根据系统非稳定工况的特征属性和规律，结合时空聚类分析方法，研究非稳定工况的故障分类诊断方法。该项研究的开展，将为系统非稳定工况的应用研究提供理论依据和部分参考数据，对于知识发现领域也有一定的积极意义。

随着科学技术和生产力水平的不断提高，各种大型系统变得日益复杂，系统的可靠性、可维护性、安全性越来越受到人们的关注。然而，复杂系统在运行过程中，处于非稳定工况的时间虽然不长，但发生的故障率是最高的，但由于非稳定工况系统的变量多，变化幅度大，响应速度快，变量间相关性关系复杂等，特别是当系统的运动方程未知，且又得不到专家系统的知识援助时，就无法建立定量模型。传统的故障检测方法不能对非稳定工况进行有效的故障检测，因本项目主要研究适于非稳定工况的故障检测与分类方法。. 在非稳动工况故障检测方法方面，概述了非稳定工况的特性，并将非稳定工况划分为周期非稳定工况和非周期非稳定工况；提出了一系列针对周期非稳定工况故障检测方法；在此基础上，提出了2种非周期非稳定工况的故障检测方法；由于非稳定工况下检测限是动态变化的，提出了几种非稳定工况的动态限，配合动态限，提出了基于相对主元分析的移动数据窗口故障检测方法；提出了基于动态峰谷限的故障检测方法；提出基于动态限的周期非稳定工况的实时故障检测模型。针对非稳定工况中非高斯数据的处理方面，提出了几种解决非稳定工况下非高斯数据处理的方法，并提出了正态PCA的故障检测方法；进而提出了基于自适应控制限的纵向标准化多周期主元分析故障检测策略等；针对不同工况，故障的检测限不一样，各种检测难度不一，我们提出一种KPCA-HSSVM-ELM自学习混合策略；并针对SD系统的突变故障，提出等价空间残差产生器的混合优化方法。. 非稳定工况的故障分类诊断，以风机、逆变器等为研究对象，提出了一系列故障分类诊断方法。融合了PCA和SVM方法的优点，提出一种基于PCA-SVM的故障诊断策略，并应用到五电平逆变故障诊断中；引入MRVM分类模型来实现非稳定工况系统的多分类；进而提出了PCA-mRVM故障诊断策略；根据RPCA算法和SVM模型的优点，将这两种方法结合，提出基于RPCA-SVM模型的级联H桥逆变故障诊断方法对逆变器进行故障诊断。在非稳定工况故障诊断的参数识别里，针对传统白盒建模方式限制条件多，黑盒建模方式的非表达性问题，提出了具有结构模拟能力的局部连接BP神经网络。. 上述研究已经发表了18篇论文，其中SCI/EI检索16篇，申请发明专利3项，实用新型专利1项。这些研究成果将为系统非稳定工况故障检测和诊断提供理论基础和策略方法。