高维征兆数据特征下的多故障智能诊断方法研究

项目名称： 高维征兆数据特征下的多故障智能诊断方法研究

项目编号： No.61203084

项目类型： 青年科学基金项目

立项/批准年度： 2013

项目学科： 自动化学科

项目作者： 张可

作者单位： 重庆大学

项目金额： 26万元

中文摘要： 复杂过程具有多样性的特点，大量的无关属性使得多故障体系非线性特征明显。不同的故障间互相关联、紧密耦合，导致故障征兆高维稀疏、不易量化、难以区分，传统的智能故障诊断方法难以发现多故障征兆间的线性函数关系。针对该问题，开展基于高维征兆数据分析的多故障体系空间和实体识别鉴定的研究。将多故障诊断视为一种不确定性因果关联表现，以征兆的空间距离、概率、趋势、分布、对象组相似度等参数作为多故障体系的数量定性指标，利用类比、学习、自组织等技术手段，使用联合聚类、非线性映射、目标优化、子空间映射等方法，将高维征兆矢量这类不确定数据在未知的知识范围内进行合理的聚类分析，实现高维征兆数据在较低维中的表征，同时使用映射聚类发现有效离群点，确保多故障诊断结果能完整的位于形成的一个类群之中，建立起具有自学习、自适应、和不确定性问题处理能力的多故障诊断方法。

中文关键词： 多重故障诊断；聚类分析；高维数据；数据驱动；模式分类

英文摘要： Complex process has the characteristics of multifarious, and simultaneity multi-fault is familiar in the area. The faults are correlative and tightly coupled, therefore, high dimension and sparse features are hard to quantified or distinguished.Linear function relationship among the multi-fault features are not easy to be found by traditional intelligent diagnosis methods. Aiming at this problem, a research on hyperspace and physical identification of multi-fault based on high-dimensional feature data analysis will be conducted. In the research, multi-fault diagnosis will be regarded as an indeterminate causal association. Various parameters (e.g. space distance, probability, tendency, distribution, similarity between different objects, etc) will be defined as qualitative indicators. Methods such as co-clustering, nonlinear mapping, target optimization, and subspace mapping are used to make a cluster analysis on the high-dimension feature vector within an unknown scope of knowledge, combining a few technical measures (e.g. analogy, self-study, self-organization ) will be combined. Simultaneously, by using the above methods effective outliers will be found by mapping clusters, thus ensuring multi-fault diagnosis can be completely located in a class . Finally, multi-fault diagnosis methods, which are self-learning

英文关键词： multiple fault diagnosis；cluster analysis；high dimensional data；data-driven；pattern classification