Collaborative Research: Robustness of Networked Model Predictive Control under Scheduling Constraints

点击翻译按钮获取中文摘要

Connecting multiple actuators, controllers, and sensors over shared data networks is a common means of reducing cost and increasing maintainability in modern industrial applications, including automobiles, aircraft, and manufacturing facilities. In most of these applications precise timing is necessary for proper system function, and timing deviations have the potential to cause detrimental and even life-threatening deterioration of performance. However it is inherent to shared networks that contention may occur, meaning that more than one connected device wants to transmit data over the network at the same time. This project will develop real-time networked controllers that resolve contention while achieving desired control objectives. Furthermore, they will be robust to perturbations of the physical system and the network itself. The focus of the project is on network architectures that are common in industry, and the results will apply particularly to automotive control and robotic applications. As reflected in the expertise of the PIs, the project combines insightful engineering with sophisticated mathematics, towards the goal of producing practically useful controllers that have rigorous performance guarantees. Through a series of outreach activities, the project will help broaden participation of underrepresented groups in STEM research.This project will address among the most challenging and important networked systems problems. It will entail fundamental research to overcome current limitations of model-based control of industrial networks. The project will use a new robust model predictive control framework and event-triggered timing model that combines the strengths of autonomous control and optimization. The work will develop an event-triggered timing model for receding horizon model predictive control of a real-time network, that will handle task dependency and timing variations and adaptively compensate for contentions and time delays. This will allow multiple sensor and actuator nodes for each control loop, a necessity for state-of-the-art networked industrial applications. The controller will respect state and input constraints, optimize cost criteria, predict timing variations, and ensure robustness to perturbations. It will provide least-conservative estimates of robust positive invariant sets in the workspace, and overcome the conservativeness of the best existing results, where the state space is usually chosen to be a sublevel set of a Lyapunov function whose boundary is determined by the supremum of the perturbations. Instead, the controller will seek maximal perturbation bounds that can be allowed before state constraints are violated. Much of the specific implementation, as well as the experimental validation, will emphasize CANbus networks. Because CANbus is popular for real-time industrial control applications, and is the standard protocol for the automotive industry, this will maximize the immediate impact of the results.