引用本文:张成,吕佩琛,伊海迪,李元.随机傅里叶特征相异度的故障检测算法[J].控制理论与应用,2022,39(7):1251~1260.[点击复制]
ZHANG Cheng,LV Pei-chen,YI Hai-di,LI Yuan.Fault detection approach using random Fourier feature dissimilarity[J].Control Theory and Technology,2022,39(7):1251~1260.[点击复制]
随机傅里叶特征相异度的故障检测算法
Fault detection approach using random Fourier feature dissimilarity
摘要点击 1082  全文点击 421  投稿时间：2021-08-24  修订日期：2022-06-24
查看全文  查看/发表评论  下载PDF阅读器
DOI编号  10.7641/CTA.2022.10781
  2022,39(7):1251-1260
中文关键词  随机傅里叶特征  相异度  非线性过程  动态过程  故障检测
英文关键词  random Fourier features  dissimilarity  nonlinear process  dynamic process  fault detection
基金项目  国家自然科学基金项目(61673279), 辽宁省自然科学基金项目(2019–MS–262), 辽宁省教育厅基金一般项目(LJ2019013)资助
作者单位E-mail
张成 沈阳化工大学 zhangcheng@syuct.edu.cn 
吕佩琛 沈阳化工大学  
伊海迪 沈阳化工大学  
李元* 沈阳化工大学 li-yuan@mail.tsinghua.edu.cn 
中文摘要
      针对大规模非线性动态过程故障检测问题, 提出随机傅里叶特征相异度(RFF–DISSIM)的故障检测方法. 首先, 利用RFF对原始数据进行映射, 获得特征空间中的数据集; 然后, 在特征空间中应用滑动窗口技术并结合相异 度指标对特征空间中的数据集进行过程状态监控. 本文方法通过RFF快速捕获数据的非线性结构并结合相异度指 标消除样本间自相关性的影响, 有效地提高了过程监控性能. 通过一个数值例子和连续搅拌釜反应器(CSTR)的仿 真实验并与传统的核主元分析、动态主元分析等方法对比分析, 仿真结果进一步证明了本文所提方法的有效性.
英文摘要
      Aiming at the problem of fault detection in large-scale nonlinear dynamic processes, a fault detection approach based on random Fourier feature dissimilarity (RFF–DISSIM) is proposed. Firstly, the RFF is used to obtain the new dataset by mapping the original dataset into the feature space. Then, the sliding window technology and dissimilarity index are utilized to monitor the status of this process in the feature space. In this paper, the nonlinear structures of dataset are quickly captured by RFF, and the influence of autocorrelation between samples is effectively eliminated by dissimilarity index. The performance in process monitoring can be effectively improved through the combination of these two methods. The efficiency of the proposed strategy is implemented in a numerical case and the continuous stirred tank reactor system (CSTR). The experimental results indicate that the proposed approach has higher detection rate than conventional methods, such as kernel principal component analysis and dynamic principal component analysis.