Experimental validation of a sensitivity-basec learning-type controller for a linear time-varying model of a flexible high-speed rack feeder

In many control applications, it is desired to repeat the same control task numerous times with identical reference trajectories. As long as only model inaccuracies and disturbances, which are practically identical for each new execution, appear as external uncertainty, such tasks are well suited for the implementation of learning-type strategies. The fundamental idea of such controllers is not only to stabilize the dynamics during a single execution of the control task but also to improve the tracking behavior successively from execution to execution. This can be achieved by an adaptation of the control signal on the basis of the time history of the tracking error from a previous execution. Well-known learning-type approaches, summarized as ILC techniques (iterative learning control), make use of linear dynamic system models which are assumed to be time-invariant in many cases. If techniques for differential sensitivity analysis are employed for the purpose of control design, also nonlinear and time-varying system models can easily be handled in the frame of a learning-type control design. This paper summarizes corresponding design techniques and presents the effectiveness of the proposed approach for the control of a time-varying model of a flexible high-speed rack feeder system.