Model-free repetitive control design for nonlinear systems

This contribution presents a novel algorithm for tuning controllers for nonlinear plants. The controller itself can be linear or nonlinear but it should be smooth in the parameters that are to be tuned. The algorithm iteratively minimizes a control criterion of LQG type without using any model of the plant, i.e., the tuning method is model free. The method relies on a series of experiments with almost identical reference signals. If the initial controller stabilizes the plant, then similar trajectories will be obtained in the state space for each experiment. The plant can be described as a linear time-variant system along one trajectory and since the trajectories in the experiments are close to each other they can all be described by the same linear time-variant system. The description as linear time-variant system is used to develop the algorithm