Automated fault detection in nonlinear systems using an OLA method combined with TAF-MFNN

This paper presents a robust fault detection (FD) scheme for detecting and approximating state faults occurring in a class of nonlinear dynamical systems. In the presence of a failure, the values exported by the on-line approximator (OLA), are used as an estimate of the real nonlinear fault function. The general inspiration for constructing OLA model in FD is based on the radial basis function (RBF) neural network technology. Here we adopt a novel tunable activation function multi-layer forward neural network (TAF-MFNN) to construct the OLA due to its strong learning capability is proposed in this paper, and a systematic procedure for constructing nonlinear estimation algorithms is developed. Eventually, the simulation studies are used to illustrate the results.