Uncorrelated component analysis on manifold for statistical process monitoring. (23rd June 2017)

• Record Type: Journal Article
• Authors: Jia, Qilong; Zhang, Yingwei; Zhai, Lirong; Feng, Lin
• Is Part Of: Journal of chemometrics. Volume 31:Number 10(2017)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2017-06-23
• Subjects: data‐driven modeling -- manifold regularization -- statistical process monitoring -- uncorrelated component analysis; Chemistry -- Mathematics -- Periodicals; Chemistry -- Statistical methods -- Periodicals; 542.85
• DOI: 10.1002/cem.2918
• Languages: English
• ISSNs: 0886-9383

Abstract

Abstract : A novel process modeling approach referred to as locally invariant uncorrelated component analysis (LIUCA) is proposed for the purpose of statistical process monitoring. The contributions are as follows: (1) LIUCA intends to find a part‐based representation subspace in which two data points are close to each other, if they are close in the k ‐nearest neighbor graph; (2) LIUCA can exploit the geometrical structure of the data space, which will improve the algorithm's modeling performance in real‐world applications; (3) LIUCA‐based multivariate statistical process monitoring scheme is proposed. (4) In contrast to traditional process modeling algorithm such as principal component analysis, LIUCA imposes no restriction on data distribution. In addition, both a multivariate numerical example and a hot galvanizing pickling waste liquor treatment process are taken to evaluate the feasibility of the proposed process monitoring scheme. Experiment results demonstrate the effectiveness of the proposed method. Abstract : A novel process monitoring technique including fault detection and identification approaches is proposed. Geometrical structure of the data space is incorporated into the monitoring model leading to an improvement of the fault‐detection performance. After process faults are detected, one can use the new fault‐identification approach to recognize the dominant inducements of the detected abnormal event.