Data-driven optimal ILC for multivariable systems: Removing the need for L and Q filter design

Many iterative learning control algorithms rely on a model of the system. Although only approximate model knowledge is required, the model quality determines the convergence and performance properties of the learning control algorithm. The aim of this paper is to remove the need for a model for a class of multivariable ILC algorithms. The main idea is to replace the model by dedicated experiments on the system. Convergence criteria are developed and the results are illustrated with a simulation on a multi-axis flatbed printer.