This paper presents a hybrid approach to online controller tuning for systems with structured yet uncertain time-varying dynamics. The method integrates model-based Linear Parameter-Varying (LPV) control design with data-driven Bayesian Optimization (BO) to achieve sample-efficient performance tuning under uncertain plant conditions. A key contribution lies in the development of a problem-specific kernel for Gaussian Process Regression (GPR), which incorporates prior system knowledge derived from an LPV model to accelerate convergence of the BO procedure. The proposed method is experimentally validated on a servo pneumatic system subject to artificial leakages, showing that the custom kernel consistently outperforms standard approaches in convergence speed.

Mechatronic application operating in dynamic and unstructured environment can benefit greatly from use of online optimization i.e. non-linear model predictive control (NLMPC). Unfortunately, the deterministic time implementation of NL-MPC on typical industrial automation hardware remains an open challenge, as well as guaranteeing performance in full operational behaviour. This article documents an implementation of an NL-MPC tool-chain. The developed methods are used on a highly dynamic parallel SCARA robot as a performant example that can benefit from the use of the proposed approach. Through us of NL-MPC, energy optimal path planning is demonstrated to operate robustly in all defined experimental conditions, while not violating the prescribed computational time. The resulting system performance is then benchmarked to the conventional industrial automation solution, and the improvement in performance is highlighted showing an improvement of up to 36% in energy efficiency.

Industrial Direct Current (DC) grids are increasingly considered since they can have economical advantages, and can lead to an improved sustainability, flexibility and reliability. However, to meet these challenges, an Energy Management System (EMS) is needed to decide when and how much to charge or discharge the storage elements. In this paper we present an EMS formulation based on Model Predictive Controller (MPC). Such methods have been applied to other energy management applications, and in this paper we adapt and apply them to the use case of industrial DC grids. We further propose a novel method to extend our EMS to ensure long term charge-sustainability is achieved, with less tuning difficulties than existing methods face. Both the application of EMS to industrial DC and the extension towards charge sustainability are illustrated on a set of simulation examples.