Fault diagnosis based on variable-weighted separability-oriented subclass discriminant analysis. (4th October 2019)

• Record Type: Journal Article
• Authors: Zeng, Yong; Jia, Zidi; Liang, Wei; Gu, Shifu
• Is Part Of: Computers & chemical engineering. Volume 129(2019)
• Publication Date: 2019-10-04
• Subjects: Fault diagnosis -- Fisher discriminant analysis -- Variable weighting -- Subclass separation; Chemical engineering -- Data processing -- Periodicals; 660.0285
• DOI: 10.1016/j.compchemeng.2019.106514
• Languages: English
• ISSNs: 0098-1354

Abstract

Abstract: Fisher discriminant analysis (FDA) is a classical method for discriminative dimensionality reduction, which has been extensively studied in the pattern classification field and widely used for fault diagnosis in industrial process monitoring. The classification performance of original FDA may degrade in overlapping, non-Gaussian or nonlinear data, and may be unable to extract sufficient features to interpret the data when the number of classes is small. To overcome these limitations, this paper proposes a combined variable weighting (VW) and separability-oriented subclass discriminant analysis (SSDA) scheme (denoted as VW-SSDA) for fault diagnosis. VW-SSDA utilizes K-means clustering to divide a class into subclasses using a separability oriented criterion, and all fault data are weighted by the corresponding weight vectors before applying FDA optimisation using re-defined scatter matrices. The proposed approach is applied to the Tennessee Eastman process. The results demonstrate that VW-SSDA shows better fault diagnosis performance than the several improved variants of the conventional FDA.