Neural networks based fault monitoring scheme for nonlinear systems and its application on robotic systems

This thesis focuses on the neural networks and their application in the fault monitoring. A neural network based fault monitoring system is presented for a class of discrete-time nonlinear systems. The neural network plays an important role of function approximator in the fault monitoring system. Two kinds of neural network approximators are proposed. One is a discrete-time RBF network with a robust gradient descent training algorithm. A fixed dead-zone technique is used to make the network parameters unchanged when the estimation errors of the network is below the upper bound of system uncertainties. It also guarantees the convergence of the estimation errors of both the neural network and the fault monitoring system in the presence of system uncertainties. The effectiveness of the RBF network based fault monitoring system is shown via simulations of a robotic system.Master of Engineerin