A Novel Hybrid Method Integrating ICA-PCA With Relevant Vector Machine for Multivariate Process Monitoring

• Published in: IEEE transactions on control systems technology Vol. 27; no. 4; pp. 1780 - 1787
• Main Authors: Xu, Yuan, Shen, Sheng-Qi, He, Yan-Lin, Zhu, Qun-Xiong
• Format: Journal Article
• Language: English
• Published: New York IEEE 01.07.2019; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Bayesian analysis; Covariance matrices; Fault detection; Feature extraction; Gaussian process; Independent component analysis; independent component analysis (ICA); Machine learning; Mathematical model; Matrix decomposition; Monitoring; Multivariate analysis; Normal distribution; Principal component analysis; principal component analysis (PCA); Principal components analysis; Process variables; relevance vector machine (RVM); Support vector machines; Tennessee Eastman (TE) process
• ISSN: 1063-6536; 1558-0865
• DOI: 10.1109/TCST.2018.2816903

This brief proposes an independent component analysis-principal component analysis (ICA-PCA) integrating with relevance vector machine (RVM) for multivariate process monitoring. Given the fact that the distribution of industrial process variables is mostly non-Gaussian and PCA cannot well deal with the non-Gaussian part. A hybrid ICA-PCA method is proposed to simultaneously extract the non-Gaussian and Gaussian information of multivariate processes. ICA is first used to monitor the non-Gaussian part of the process and then the Gaussian part of the residual process can be extracted using PCA. After feature extraction, a Bayesian-based classifier named RVM is established to make fault detection for the sake of both preventing the chosen of threshold as in traditional method and compensating for the single statistic. The performance of the proposed approach is validated using the Tennessee Eastman process. Simulation results verified the effectiveness of the proposed method.