Fault diagnosis for nonlinear systems using a bank of neural estimators

An optimal diagnosis problem is cast according to a receding-horizon strategy. Plant, actuator, and sensor fault models are developed and parameters describing such faults are introduced for the purpose of modelling. The estimates of these parameters are made by using a nonlinear estimation technique that consists in minimizing an estimation cost function on line. The solution is obtained by means of a bank of filters designed via neural networks. Simulation results show the effectiveness of the approach.