Title :
Sensors´ FDD by quadruple and modified ART-1 ANNs
Author :
Khan, Muhammad Rafiq
Author_Institution :
Pakistan Atomic Energy Commission, Islamabad, Pakistan
Abstract :
An approach to continuous online detection and diagnosis of sensors multiple simultaneous faults with various degrees of deviations is presented. The FDD system is composed of a feature detector, a novel artificial neural quadruple network capable of performing rule-based operations and a modified ART-1 that can memorize the faults´ history. An additional backward intermediate term flush memory is employed in the ART-1 to memorize faults history to eliminate external disturbances and noise. The feature detector is developed such that it is capable of providing a set of vectors of digital residuals over a full range and for various combinations of simultaneous faults. The system is successfully employed for a nuclear power plant waste treatment system´s sensors FDD
Keywords :
ART neural nets; fault diagnosis; feature extraction; multilayer perceptrons; nuclear power stations; recurrent neural nets; sensors; waste disposal; artificial neural quadruple network; backward intermediate term flush memory; continuous online fault detection; continuous online fault diagnosis; digital residuals; feature detector; modified ART-1 ANN; nuclear power plant waste treatment system; rule-based operations; Detectors; Fault detection; Fault diagnosis; Hardware; History; Neural networks; Power generation; Production; Redundancy; Robustness;
Conference_Titel :
Neural Networks Proceedings, 1998. IEEE World Congress on Computational Intelligence. The 1998 IEEE International Joint Conference on
Conference_Location :
Anchorage, AK
Print_ISBN :
0-7803-4859-1
DOI :
10.1109/IJCNN.1998.682274
