The Research On Multi-model Based Active Fault-tolerant Control Methods

An active fault-tolerant control method based on multi-model including the possible system fault statuses is proposed aiming at nonlinear systems and some simulations are done on its validity and robustness with the precondition that some data about the systems are known. The main research work on it is done as follows:1) An active fault-tolerant control method based on NN multi-model is put forward. NN are used to model the nominal and the pre-known fault types of system. Each controller parameters are trained offline respectively based on NN-PID method. The model bank is formed of the NN models and corresponding controller parameters for all possible running modes. The method employs a supervision mechanism replacing the FDD and other units in classical active FTC methods which belongs to an hidden FDD methodology. The regulation of the controller parameters in system running is according to the continually real-time compute and analysis of the two supervision indices which are defined as the system performance tolerant index and the model mismatch index, thus the fault-tolerant control is obtained when the known fault occurs. The approach is validated by MATLAB simulations and it is validated to linear system likewise.2) Considering the incompletion of the model bank and the demand of real-time tuning of the PID parameters in case of the unknown fault occurring, an improved training method of controller parameters is put forward by introducing a chaos mechanism to NN weigh space. The simulations indicate the improved method has a better convergence and global optimal property in parameters tuning and its control effect is better than the former one.3) There is some brief analysis on the performance of the FTC system and some simulations on systems which contain some uncertainty due to the environment in the thesis. The simulations indicate the FTC systems have some robustness in the condition of a limited disturbance and small faults, which means that the supervision does not have false judge and false controller reconfiguration and the controller itself has some robustness.