Learning Control Without Prior Models: Multi-Variable Model-Free IIC, with application to a Wide-Format Printer

Abstract

Learning control enables performance improvement of mechatronic systems that operate in a repetitive manner. Achieving desirable learning behavior typically requires prior knowledge in the form of a model. The prior modeling requirements can be significantly reduced by using past operational data to estimate this model during the learning process. The aim of this paper is to develop such a data-driven learning control method for multi-variable systems, which requires that directionality aspects are properly addressed. This is achieved by using multiple past experiments to estimate a frequency response function of the inverse dynamics while ensuring smooth convergence by using smoothed pseudo inversion. The developed method is successfully applied to an industrial wide-format printer, resulting in high performance.