Fault detection and isolation in non-linear systems by using oversized neural networks

This article is about fault detection and isolation (FDI) methods by the use of neural networks. The system to supervise is modelised by using a particular structure of neural networks where the hidden layer presents some additive neurons which are only connected to the output neuron and one of the input neurons. These supplementary neurons permit to perform an estimation of each input of the network and the comparison of this estimation with the actual one permits, by using a statistical test, to detect and locate the presence of fault on one input. The proposed method is tested on a simulation example.