An intelligent controller design based on system parameter estimation

To enhance the performance of an intelligent control system an automated, online procedure for failure detection is needed. An interesting method is the use of neural network. The design is achieved by implementing a real-time recurrent neural network. Based on the input-output data, the recurrent neural network is used to simultaneously identify the A and B matrices and to estimate the states of a state-space model of a linear time-invariant MIMO discrete dynamic system. The network consists of one layer of linear activation neuron functions and delay operators. In this paper, the intelligent controller architecture, the off-line training results of the neural model, and the online testing of the neural network ability to learn a failed plant dynamics will be presented