Neurofuzzy networks based fault diagnosis of nonlinear systems

An artificial intelligence fault detection and isolation technique based on neurofuzzy networks is developed in this paper. B-spline functions are used as the memberships of the fuzzy variables in the network, yielding a network with its output linear in its weights. Consequently, the networks can be trained online using the well known recursive least squares method. In the proposed technique, a neurofuzzy network is trained first to model the plant under normal operating conditions. Residual is generated online by this network for fault diagnosis, which is modelled by another neurofuzzy network. From fuzzy rules obtained by the second neurofuzzy network, qualitative diagnosis of the faults is extracted, as presented in this paper. The performance of the proposed fault diagnosis scheme is illustrated by applying to diagnose faults in a simulated two-tank system