Abstract

Control of mechatronic systems poses typical challenges i.e. nonlinear and fast dynamics with unknown optimal reference trajectories which vary significantly in course of operation. In this context, learning control for production machines has recently garnered attention from the research community and special sessions in conferences and journals have been organized. We tackle such systems by a two-level learning control approach with (non)linear model-based control at lower level tracking references learned/adapted by a high level iterative leaning controller.

Moreover, many complex machines are non-collocated vibrating systems i.e. actuators and sensors do not act at the same point. We show that classical compensators are incapable to control the induced oscillations and that model-based controller synthesis and tuning for can deliver the required performance.

The above two methodologies are experimentally validated over a wet-clutch engagement and harvester height control. If time permits, new ideas on extensions of closed-loop formulations of MPC for constrained systems will be touched on.