On-line fault diagnosis and accommodation for system failures

Scope and method of study. The major interest of this research work is to investigate on-line fault accommodation control problems for unanticipated catastrophic system failures in the general format. Being control engineers, we would like to know how to properly control the system behavior in time to prevent the failure from causing more serious lost, if the system under failures is still controllable at that time, and/or possibly recover the system performance based only upon imprecise or insufficient information.; Findings and conclusions. The necessary and sufficient conditions for system on-line stability under catastrophic failures have been derived based upon discrete-time Lyapunov stability theory. An on-line fault accommodation control framework is proposed to deal with the desired trajectory-tracking problem for systems suffering from various unknown and unanticipated catastrophic failures. Artificial Neural Network is suggested as the on-line estimator to learn and approximate the unknown system failure dynamics. Effective control signals to accommodate the dynamic failures are then computed through the realization of the neural network estimator for the unknown failures based only upon the partially available information of the faults. A more complete architecture for intelligent fault diagnosis and accommodation has also been presented by incorporating the developed intelligent fault tolerant control technique with a cost-effective fault detection scheme and a multiple-model based failure diagnosis process to efficiently handle the false alarms, the accommodation of the anticipated failures, and to reduce the unnecessary control effort and computational complexity in on-line situations. Extensive simulation studies have been completed to validate the proposed on-line control framework under various multiple failures, noisy environments, and false alarm situations. Real-time hardware experiments and the simulations show encouraging results and demonstrate the effectiveness of the developed control methodology for proper accommodation of catastrophic system failures in on-line real-time situations.