A combined monitoring scheme with fuzzy logic filter for plant-wide Tennessee Eastman Process fault detection. (21st September 2018)

• Record Type: Journal Article
• Authors: Ammiche, Mustapha; Kouadri, Abdelmalek; Bakdi, Azzeddine
• Is Part Of: Chemical engineering science. Volume 187(2018)
• Page Start: 269
• Page End: 279
• Publication Date: 2018-09-21
• Subjects: Principal Component Analysis -- Moving Window Principal Component Analysis -- Adaptive thresholds -- Fault detection and diagnosis -- Fuzzy logic filter -- Tennessee Eastman Process; Chemical engineering -- Periodicals; Génie chimique -- Périodiques; Chemical engineering; Periodicals; Electronic journals; 660
• DOI: 10.1016/j.ces.2018.05.001
• Languages: English
• ISSNs: 0009-2509

Abstract

Highlights: A Modified Moving Window PCA (MMWPCA) with Fuzzy Logic Filter (FLF) is developed. The developed monitoring scheme is a combination of PCA and MWPCA. MMWPCA with FLF provides adaptive thresholds with fixed PCA model. Hold-one MMWPCA with FLF is applied to TEP and compared to other works. The results reveal the effectiveness of the proposed method in process monitoring. Abstract: Principal Component Analysis (PCA) is the most common Multivariate Statistical Process Control (MSPC) method that is widely used for Fault Detection and Diagnosis (FDD). Since early abnormality detection with high accuracy is required for safe and reliable process operation, False Alarms Rate (FAR), Missed Detection Rate (MDR) and the detection time delay are the major factors that must be taken into consideration when developing any process monitoring scheme. Unfortunately, the PCA performance, with fixed limits, is weak in terms of the stated factors. In contrast, conventional Moving Window PCA (MWPCA) is an adaptive technique which updates both the PCA model and the thresholds once a new normal observation is available. Yet, MWPCA methodology still does not reduce the MDR and the detection delay. In this paper, a Modified MWPCA (MMWPCA) with Fuzzy Logic Filter (FLF) is proposed to enhance the monitoring performance of PCA. It is an adaptive approach with a fixed model that combines both aforementioned techniques. The aim of using FLF is to ensure robustness to false alarms without affecting the Fault Detection (FD) performance. The application of the proposed method has been carried out on the Tennessee Eastman Process (TEP). Hold-one and hold-five MMWPCA with FLF are applied and compared to recent FDD work in the literature. The obtained results demonstrate the superiority of the proposed technique in detecting different types of faults with high accuracy and with shorter time delay.