An Approach to Fault Diagnosis for Non Linear Dynamic Systems Using Neural Networks

Abstract This paper proposes an approach to fault detection and identification (FDI) for non linear dynamic systems using neural networks. A fault is considered as a variation of physical parameters; therefore the FDI problem can be reduced to the parameter identification of a non linear system and to the association of the set of the estimated parameters with the mode of operation of the plant. Under some assumptions, this can be efficiently accomplished by a combined use of a linear parameter estimator and a bank of neural classifiers. Each neural network is trained to perform the diagnosis in a certain working point of the plant; a supervisor is introduced to allow interpolation between the working points, which the system has been trained with. The FDI scheme has been tested by simulation on a non linear mechanical oscillator.