Hybrid Methods for Detection and Identification of Faults in Dynamic Systems

• Published in: Asian journal of control 2015-09, Vol.17 (5), p.1831-1847
• Main Authors: Martins, Rodrigo S., Vale, Marcelo R. B. G., Maitelli, Andre L.
• Format: Article
• Language: English
• Subjects: Decision trees; Dynamical systems; Fault detection and identification (FDI); Fault diagnosis; Identification systems; Mean square errors; mean squared error; state observer
• ISSN: 1561-8625; 1934-6093
• DOI: 10.1002/asjc.1039

Since the 1960s, when automation became essential to productivity, methods for the detection and identification of faults have been proposed. Physical systems are diversified and can be mechanical, electrical, pneumatic, electronic, or a combination of these. In addition, real plants have a large number of these devices, which are for its own operation, sensoring or control. Therefore the solutions given for detection of faults are generally very specific or particular. This paper aims to describe and analyze two hybrid methods of detection and fault identification based on residue and to check whether their inclusion with other methods, combining different techniques, can produce a better fault detection and identification system. The methods use the state observers for the generation of residues, which serve for the detection and identification and the set called the bank of signatures to identify the faults. Thereafter, the methods use different approaches to diagnose the fault: the first uses the approach of the mean square error, and the second uses a decision tree.