An adaptive confidence limit for periodic non-steady conditions fault detection. (May 2016)

• Record Type: Journal Article
• Authors: Wang, Tianzhen; Wu, Hao; Ni, Mengqi; Zhang, Milu; Dong, Jingjing; Benbouzid, Mohamed El Hachemi; Hu, Xiong
• Is Part Of: Mechanical systems and signal processing. Volume 72/73(2016)
• Page Start: 328
• Page End: 345
• Publication Date: 2016-05
• Subjects: Fault detection -- Periodic non-steady conditions -- Longitudinal-Standardization -- Multiperiod PCA -- Adaptive confidence limit; Structural dynamics -- Periodicals; Vibration -- Periodicals; Constructions -- Dynamique -- Périodiques; Vibration -- Périodiques; Structural dynamics; Vibration; Periodicals; 621
• DOI: 10.1016/j.ymssp.2015.10.015
• Languages: English
• ISSNs: 0888-3270

Abstract

Abstract: System monitoring has become a major concern in batch process due to the fact that failure rate in non-steady conditions is much higher than in steady ones. A series of approaches based on PCA have already solved problems such as data dimensionality reduction, multivariable decorrelation, and processing non-changing signal. However, if the data follows non-Gaussian distribution or the variables contain some signal changes, the above approaches are not applicable. To deal with these concerns and to enhance performance in multiperiod data processing, this paper proposes a fault detection method using adaptive confidence limit (ACL) in periodic non-steady conditions. The proposed ACL method achieves four main enhancements: Longitudinal-Standardization could convert non-Gaussian sampling data to Gaussian ones; the multiperiod PCA algorithm could reduce dimensionality, remove correlation, and improve the monitoring accuracy; the adaptive confidence limit could detect faults under non-steady conditions; the fault sections determination procedure could select the appropriate parameter of the adaptive confidence limit. The achieved result analysis clearly shows that the proposed ACL method is superior to other fault detection approaches under periodic non-steady conditions. Highlights: Longitudinal-Standardization is proposed to convert non-Gaussian data to Gaussian ones. The multiperiod PCA algorithm is proposed to improve the monitoring accuracy. The adaptive confidence limit (ACL) could detect faults under non-steady conditions. The fault sections determination procedure is proposed to select ACL parameter.