Abstract: A novel fault diagnosis architecture for a class of nonlinear time-delay systems is studied.As the input to neural networks, the estimated states and time-delay states are used to approximate the faults.A state observer is proposed, and unmeasured states estimated on-line using the output of system based on nonlinear approximate ability of neural networks.The estimated states are input to the fault estimator whose outputs are an estimated fault model.Based on the Lyapunov theory, the system stability and convergence of the parameters are analyzed.The condition of solution to ensure existence of algorithm is given.Simulation results show effectiveness of the proposed approach.

Key words: nonlinear time-delay systems, state observer, fault estimator, fault diagnosis, neural network