Fault diagnosis based on knowledge extracted from neurofuzzy networks using binary and real-valued fault databases

In this paper, an online fault diagnosis scheme for nonlinear systems is derived from fuzzy rules extracted from the neurofuzzy network that models the residuals of the system. As the neurofuzzy network is updated online by the recursive least squares method, the proposed technique is able to diagnose faults online. To initiate the fault diagnosis scheme, a binary or real-valued fault database is constructed first from fuzzy rules extracted from each of the neurofuzzy networks that model the possible faults in the system. Faults are diagnosed online by comparing the currently extracted fuzzy rules with those in the fault database using a classifier. As an illustration, a nonlinear DC motor control system is used to illustrate the implementation of the proposed diagnosis schemes, and the performance of the binary and real-valued fault databases is compared.