Design of fault tolerant neurocontrollers using immunization technique

Simulation results show that the overall fault tolerance of a neural network used to control a linear plant can be improved when training incorporates the immunization technique. Fault tolerance is improved further when the momentum gain, α, is reduced from 0.4 to 0. The results for the neurocontroller of the nonlinear plant are, however, different, with little or no improvement in fault tolerance of the weights when training incorporates noise. It may be that the architecture for the neurocontroller is unsuitable for the control of the nonlinear plant