Fault diagnosis of nonlinear systems using multistep prediction of time series based on neural network (Q2725129)

tew approaches to fault diagnosis of nonlinear systems based on neural networks or multi-sensor data fusion have been widely studied recently. NEWLINENEWLINENEWLINEHere, a novel approach to this problem that utilizes multistep prediction of time series directly based on recurrent neural networks is presented. For this purpose, the historical residual series constructed by sampling values from process sensor and the predicting residual series constructed by desired values for the process sensor are defined. Then evaluation indices are designed to characterize the fault diagnosis. NEWLINENEWLINENEWLINECritically one should mention that it is not clear how such fault detection strategy should work from the paper. It is even not clear how the fault is defined as such. By the way, the example used for a simulation and confirmation of the method presented is unfortunately only in the form of a dynamical mathematical model from which it is not clear if the method works in real situations.