Adaptive neural network control of FES in arm movements and its applications based on a resonant converter

Three types of muscle model are discussed in this paper. Because they are time-dependent and highly nonlinear, its performance with the functional electrical stimulator (FES) must be studied so as to ensure the robustness to changes in system gain, since the gain can vary substantially and abruptly with changes in operating point. The paper is to study the use of neural network adaptive control to regulate the FES output in order to achieve functional restoration and activities. A multiple level neural network with recurrent neural networks as system inversive identification is examined. Simulation experiment shows that it has stability, self-tuning, robust and adaptive. The network is then applied to an extended period quasi-resonant converter for the application as an FES. The circuit was originally used as a power converter, but is now used in this application as an electrical stimulator because it has three degrees of freedom. That is duty-ratio, frequency and amplitude. The advantage of the proposed system is that no transformer is needed and the variation of electrical pulses is mainly relied on the resonant components and the extended-period resonant principle. Experimental results show that the system behaves satisfactory.