Novel common and special features extraction for monitoring multi-grade processes. (June 2018)

• Record Type: Journal Article
• Authors: Liu, Jingxiang; Liu, Tao; Chen, Junghui; Qin, Pan
• Is Part Of: Journal of process control. Volume 66(2018)
• Page Start: 98
• Page End: 107
• Publication Date: 2018-06
• Subjects: Multi-grade processes -- Process monitoring -- Limited samples -- Common feature extraction -- Subspace division; Process control -- Periodicals; Fabrication -- Contrôle -- Périodiques; Process control; Periodicals; Electronic journals; 660.281
• DOI: 10.1016/j.jprocont.2018.03.001
• Languages: English
• ISSNs: 0959-1524

Abstract

Highlights: Novel feature extraction with very limited samples for monitoring multi-grade processes. Each production grade is divided into common, special and residual part for model building. Three indices are defined for on-line monitoring of each part. Common and special variables are identified for understanding multi-grade processes. Abstract: Since industrial plants manufacture different specifications of products in the same production line by simply changing the recipes or operations to meet with diversified market demands, it often happens that very limited samples could be measured for each grade of products, thus inadequate to establish a model for monitoring the corresponding process. To cope with the difficulty for monitoring such multi-grade processes, a novel feature extraction method is proposed in this paper to establish process models based on the available data for each grade, respectively. Firstly, a common feature extraction algorithm is proposed to determine the common directions shared by different grades of these processes. Based on the extracted common features, the principal component analysis is then used to extract the special directions for each grade, respectively. Consequently, each grade of these processes is divided into three parts, namely common part, special part, and residual part. Three indices are correspondingly introduced for on-line monitoring of each part, respectively. A numerical case and an industrial polyethylene process are used to demonstrate the effectiveness of the proposed method.