Data-Based Tuning of Reduced-Order Inverse Model in Both Disturbance Observer and Feedforward With Application to Tray Indexing

Abstract

Performance of traditional model-based control relies upon accurate modeling. In motion control of flexible systems, it is desirable to use the reduced-order model for ease of trajectory planning and pole placement, but its performance is constrained by modeling inaccuracies due to the existence of friction and multiple flexible modes. To improve the tracking performance, we have developed a data-based method for iterative tuning of the parameters in the reduced-order inverse model within a three-degree-of-freedom composite control structure. The proposed method solely makes use of the input-output data obtained during closed-loop experiments to fine-tune the inverse system model, and accurate system modeling is not required. Unbiasedness of the cost function gradient estimation is proven under reasonable assumptions of stochastic properties of the perturbations. Simulation and experiments are conducted to further illustrate the proposed method and show its practical appeals in industrial applications.