A framework for on-line trend extraction and fault diagnosis

• Published in: Engineering applications of artificial intelligence 2010-09, Vol.23 (6), p.950-960
• Main Authors: Maurya, Mano Ram, Paritosh, Praveen K., Rengaswamy, Raghunathan, Venkatasubramanian, Venkat
• Format: Article
• Language: English
• Subjects: Algorithms; Extraction; Fault detection; Fault diagnosis; Faults; Matlab; On-line; On-line systems; Qualitative trend analysis; Tennessee Eastman process; Trends
• ISSN: 0952-1976; 1873-6769
• DOI: 10.1016/j.engappai.2010.01.027

Qualitative trend analysis (QTA) is a process-history-based data-driven technique that works by extracting important features (trends) from the measured signals and evaluating the trends. QTA has been widely used for process fault detection and diagnosis. Recently, Dash et al. [2004. A novel interval-halving framework for automated identification of process trends. AIChE Journal 50 (1), 149–162] presented an interval-halving-based algorithm for off-line automatic trend extraction from a record of data, a fuzzy-logic based methodology for trend-matching and a fuzzy-rule-based framework for fault diagnosis (FD). In this article, an algorithm for on-line extraction of qualitative trends is proposed. A framework for on-line fault diagnosis using QTA also has been presented. Some of the issues addressed are: (i) development of a robust and computationally efficient QTA-knowledge-base, (ii) fault detection, (iii) estimation of the fault occurrence time, (iv) on-line trend-matching, and (v) updating the QTA-knowledge-base when a novel fault is diagnosed manually. A prototype QTA-based diagnostic system has been developed in Matlab ® . Results for fault diagnosis of the Tennessee Eastman process using the developed framework are presented.