联合指标独立成分分析在多变量过程故障诊断中的应用

Translated title of the contribution: Combined indices for ICA and their applications to multivariate process fault diagnosis

樊继聪, 王友清*, 秦泗钊

*Corresponding author for this work

Research output: Journal PublicationsJournal Article (refereed)peer-review

27 Citations (Scopus)

Abstract

作为主成分分析(Principal component analysis, PCA)和因子分析(Factor analysis, FA)的扩展, 独立成分分析(Independent component analysis, ICA)已经在多变量过程故障诊断中得到了很多的应用和发展. ICA的监测指标通常有三个(I2、Ie2和SPE), 使用起来不如一个指标方便, 且分散了故障信息.本文利用三个指标的加权和, 提出了两种联合的ICA监测指标. 本文进一步对比分析了不同指标的统计意义和物理意义, 并在仿真数据中验证了联合指标的优势, 在TE过程中验证了其检测和诊断特性.

As a development of principal component analysis (PCA) and factor analysis (FA), independent component analysis (ICA) has been applied effectively to multivariate process monitoring and fault diagnosis and has got many excellent achievements. Usually, ICA has three indices for monitoring and diagnosis, i.e., I2, Ie2, and SPE, and the multi-indexes make the monitoring and diagnosis inconvenient and also decentralizes the fault influence. In this paper, two combined indices for ICA are developed, both of which are weighted sums of the three indices. The statistics and physical meanings of all indices are analyzed and compared. Based on the simulation tests on a numerical example and TE process, the proposed combined indices have some advantages compared with the traditional multi-indices. Copyright © 2013 Acta Automatica Sinica. All rights reserved.
Translated title of the contributionCombined indices for ICA and their applications to multivariate process fault diagnosis
Original languageChinese (Simplified)
Pages (from-to)494-501
Number of pages8
Journal自动化学报/Acta Automatica Sinica
Volume39
Issue number5
DOIs
Publication statusPublished - May 2013
Externally publishedYes

Keywords

  • 多变量过程
  • 故障诊断
  • 独立成分分析
  • 联合指标
  • Multivariate process
  • fault diagnosis
  • Independent component analysis
  • combined index

Fingerprint

Dive into the research topics of 'Combined indices for ICA and their applications to multivariate process fault diagnosis'. Together they form a unique fingerprint.

Cite this