引用本文:张沐光,宋执环.一种基于独立元贡献度的子空间故障检测方法[J].控制理论与应用,2010,27(3):296~302.[点击复制]
ZHANG Mu-guang,SONG Zhi-huan.Subspace fault detection method based on independent component contribution[J].Control Theory and Technology,2010,27(3):296~302.[点击复制]
一种基于独立元贡献度的子空间故障检测方法
Subspace fault detection method based on independent component contribution
摘要点击 2107  全文点击 941  投稿时间：2009-01-18  修订日期：2009-06-12
查看全文  查看/发表评论  下载PDF阅读器
DOI编号  10.7641/j.issn.1000-8152.2010.3.CCTA090063
  2010,27(3):296-302
中文关键词  独立元分析  子空间  故障检测  独立元贡献度
英文关键词  ICA  subspace  fault detection  IC contribution
基金项目  国家自然科学基金资助项目(60774067, 60736021).
作者单位E-mail
张沐光 浙江大学 工业控制研究所 工业控制技术国家重点实验室 mgzhang@iipc.zju.edu.cn 
宋执环* 浙江大学 工业控制研究所 工业控制技术国家重点实验室 zhsong@iipc.zju.edu.cn 
中文摘要
      针对工业过程故障检测问题, 本文定义了独立元贡献度和贡献度矩阵, 提出一种改进的子空间检测算法. 首先, 利用独立元分析(independent component analysis, ICA)算法提取过程变量的独立元信息, 通过计算各个独立元在过程变量上的贡献度, 构造贡献度矩阵; 然后根据贡献度的大小, 挑选出对应的变量组成反映不同“源”信息的子空间, 并在这些子空间上建立故障检测模型; 最后综合以上的多个检测模型, 根据实际的需求或者故障的传播特征, 确定集成策略, 对工业过程进行故障检测. 通过在TE(Tennessee Eastman)过程上对21种故障工况和1种正常工况的仿真研究, 说明提出的改进算法是有效的.
英文摘要
      In order to handle the problem of fault detection for industrial processes, an improved subspace method is proposed based on the definitions of the independent component (IC) contribution and the contribution matrix. First, the appropriate independent components are extracted by independent component analysis(ICA), and then, the contributions of different independent components on the process variables are calculated to construct the contribution matrix. According to the contribution, a set of suitable subspaces, which can reflect different root causes, are constructed by corresponding variables. The fault detection models are established on these subspaces. Finally, combining all the above fault detection models and choosing the proper ensemble strategy based on the actual requirement or the spread characteristic of the faults, we make the ensemble decision for the fault detection of industrial processes. A case study on the Tennessee Eastman(TE) process for each mode (1 normal and 21 faulty) illustrates the effectiveness of the proposed method.