Multi-phase batch process monitoring based on multiway weighted global neighborhood preserving embedding method. (September 2018)

• Record Type: Journal Article
• Authors: Hui, Yongyong; Zhao, Xiaoqiang
• Is Part Of: Journal of process control. Volume 69(2018)
• Page Start: 44
• Page End: 57
• Publication Date: 2018-09
• Subjects: Batch -- Process monitoring -- Multi-phase -- Global-local -- Probability weighted -- Gaussian mixture model; Process control -- Periodicals; Fabrication -- Contrôle -- Périodiques; Process control; Periodicals; Electronic journals; 660.281
• DOI: 10.1016/j.jprocont.2018.06.012
• Languages: English
• ISSNs: 0959-1524

Abstract

Highlights: The process is divided into different operation phases, and it is easy to carry out statistical analysis for each phase. Global and local structures are fully extracted to preserve more features in each phase. The probability density of the extracted structures is estimated to amplify useful information and suppress noise. Abstract: A multi-phase batch process monitoring method based on multiway weighted global neighborhood preserving embedding (MWGNPE) is proposed. MWGNPE has three advantages. Firstly, for the multi-phase feature of batch process, gaussian mixture model (GMM) method is used to divide phases by clustering characteristics. Secondly, after the multiple phases have been divided, global and local structures are extracted by using global neighborhood preserving (GNPE) method. Thirdly, probability density estimation characteristic of GMM is introduced to estimate the probability density of the extracted global and local structures. It can amplify useful information and suppress noise. These three advantages make MWGNPE well suit for batch process monitoring. A full MWGNPE model is combined with the cluster and the density estimation characteristic of GMM concurrently to improve the effect of fault detection in batch process monitoring. The effectiveness and advantages of proposed method are verified by a numerical system and the penicillin fermentation process. The results show that the proposed method can effectively capture the fault information hidden in process data and has the superiority compared with other conventional methods.