Hopfield/ART-1 Neural Networks based Fault Detection and Isolation

A new approach to detect and isolate faults in linear dynamic systems is proposed. System parameters are estimated by Hopfield type neural network, while the system is in certain operating stage. When the system dynamics changes, estimated parameters go through a transition period and this is used to detect faults. But the estimates are not reliable enough to be used for isolating faults. The judgement on the instance at which the system moves out of the transition zone is made through a user specified threshold on the sum of squares of residuals in a moving window. Once the system is out of the transition zone and settles to a new operating level, the estimated parameters are classified using an ART-1 based network. The proposed scheme is implemented to detect faults in a position servo system.