Fast and Distributed Model Predictive Control - Tailored Solutions for Mechatronic Systems

The ever increasing desire of humans to automate tasks that are laborious and repetitive has made control systems prevalent in today's society. Applications are widespread in industry, transportation, agriculture and in the household. Novel challenging control applications and the continual quest to improve efficiency, have led to the rise of advanced control methods relying on optimization, such as model predictive control (MPC). Although MPC is a well-investigated control approach, new challenges keep arising due to emerging applications. This thesis aims at developing novel MPC approaches to meet some of these challenges, with a particular focus on mechatronic applications. As for any feedback control approach, the effectiveness of MPC demands an update rate that is sufficiently fast compared to the system dynamics. This renders MPC particularly challenging for fast systems, which are omnipresent in mechatronics. Furthermore, such fast control laws should often be implemented on resource-constrained hardware, embedded on the controlled system. To meet this challenge, the thesis proposes a fast MPC algorithm, based on a real-time implementation of the proximal gradient method. It results in simple arithmetics which are rapidly executed on resource-constrained hardware. For linear systems, a closed-loop stability proof is provided. Advanced control applications typically lead to complex control problems. A particular example is free motion planning of motion systems in an obstructed environment, which generally leads to a high dimensional nonconvex problem. In order to efficiently solve such problems, an approach is proposed which parameterizes the motion trajectories as polynomials in a B-spline basis and which exploits B-spline properties to efficiently translate the motion planning problem into a small-scale nonlinear program. A software toolbox, called OMG-tools, is developed to provide a user-friendly interface for modeling, simulating and deploying spline-based motion planning problems. Enhancements in communication technology and miniaturization of computing devices, have led to the rise of so-called multi-agent systems, i.e. systems comprising different interacting intelligent components. In order to reflect such system's modularity on control level, a distributed control architecture is generally adopted. Such approach provides each subsystem with its own controller, while communication between the controllers allows them to cooperate in order to accomplish a common goal. This thesis proposes a novel distributed MPC approach for controlling autonomous vehicles moving in formation. The algorithm is based on a real-time implementation of the alternating direction method of multipliers, and allows high control rates with limited inter-vehicle communication. Furthermore, the thesis presents a demonstration of a factory of the future, in which distributed MPC for transportation is combined with distributed localization and coordination.

Jaar van publicatie:2018

Toegankelijkheid:Open