A neurofuzzy approach for fault diagnosis in dynamic systems

A system structure based on fuzzy set models of neurons and networks is introduced. The aim is to provide a learning approach for pattern classification. The focus of the paper is on its application to fault diagnosis of dynamic systems viewed as a specific category of pattern classification. The main features of the system include: automatic rule generation; learning capability; processing time independent of the input space partition, if the number of inputs is fixed; and no need of process models if fault patterns are available. Simulation results concerning a seventh order, nonlinear time variant system are presented. The system successfully detected and diagnosed fifteen faults. Its response time in diagnosing suggests the feasibility in real-time applications