A supervised multisegment probability density analysis method for incipient fault detection of quality indicator. (August 2022)

• Record Type: Journal Article
• Authors: Tao, Yang; Shi, Hongbo; Song, Bing; Tan, Shuai
• Is Part Of: Journal of process control. Volume 116(2022)
• Page Start: 53
• Page End: 63
• Publication Date: 2022-08
• Subjects: Incipient fault detection -- Quality indicator -- Process monitoring -- Supervised multisegment probability density analysis; Process control -- Periodicals; Fabrication -- Contrôle -- Périodiques; Process control; Periodicals; Electronic journals; 660.281
• DOI: 10.1016/j.jprocont.2022.04.006
• Languages: English
• ISSNs: 0959-1524

Abstract

Abstract: The quality indicator monitoring has received widely attention and research in recent years, however, the detection of indicator-related incipient fault is still a challenging topic. In this paper, a supervised probability density analysis algorithm is proposed to detect the incipient fault in quality indicator. Firstly, the core process variable filter is introduced, and the regression model is constructed to extract the indicator-related information from process variable. Secondly, the data distribution extension and subsegment division strategy are presented, and a probability density estimation method is put forward for the indicator-related latent variable. Through the proposed symmetric divergence index, the distribution discrepancy between the online sample and the reference sample set is evaluated, which can be used for the incipient fault detection. Finally, a numerical example and the Tennessee Eastman process are used to demonstrate the effectiveness of the proposed method. Highlights: A novel supervised multisegment probability density analysis algorithm is proposed, which can achieve the online incipient fault detection for the quality indicator. A supervised feature analysis method is presented to extract the indicator-related information from process variable. A data distribution interval extension and subsegment division method is introduced for the probability density estimation of the indicator-related latent variables. A symmetric divergence index is presented to evaluate the distribution discrepancy between the online sample and the reference sample set, which has high sensitivity for the incipient fault.