Manufacturing systems considered as time domain control systems : receding horizon control and observers

This thesis considers manufacturing systems and model-based controller design, as well as their combinations. The objective of a manufacturing system is to create products from a selected group of raw materials and semifinished goods. In the field of manufacturing systems control is an important issue appearing at various operation levels. At the level of fabrication, for example, control is necessary in order to assure properly working production processes such that products are being fabricated in the desired way. At a higher level in the hierarchy of manufacturing system control, the product streams through the system are controlled in order to satisfy, for example, customer demands in an optimal way. Here, the definition of optimal can be interpreted in various ways, such as "with the least possible costs in terms of money" or "in the shortest possible time". In this research, the attention is focussed on this higher hierarchy level of manufacturing system control. In the literature, many heuristic methods have been developed for the control of a manufacturing system. Nowadays, some heuristicmethods are still being used in combination with operator experience for management of resources and planning of production. However, as the complexity of the manufacturing systems increases rapidly, the (simple) heuristic methods and operator experience will at some point become incapable of finding an optimal control strategy. In this dissertation the potential of consideringmanufacturing system control from a control systems point of view is investigated. The ultimate goal of the research is to eventually obtain a more constructive way to address controller design for manufacturing systems. One control strategy from control systems theory, on which is in particularly focused in this research, is a model-based receding horizon control strategy, known in literature as Model Predictive Control (MPC). Since in manufacturing systems a lot of physical system constraints are involved, like for example finite machine process capacities, finite product storage capacities, finite product arrival rates, etc., the capability for a manufacturing control strategy to handle those constraints is a necessity. One of the key features of model predictive control is the capability of handling constraints in the controller design. This is one of the major motivations to investigate the model predictive control principle as a control strategy for manufacturing systems. Other issues that are important and that the model predictive control design methodology can handle is to enforce optimality, to introduce feedback, and the capability of allowing for mixed continuous and discrete model structures. The later are typically encountered when models of manufacturing systems are derived. The main results that are obtained in this dissertation and that are relevant in the context of manufacturing systems control, but are certainly also relevant beyond this field are: • One has developed an robust computationally friendly nonlinear model predictive control algorithm that can handle model structures with mixed continuous and discrete dynamics. The algorithm can be designed for additive disturbance rejection purposes; • Robustness (with respect to measurement noise) results that are in particulary of interest in the field of nonlinear model predictive control are obtained; • An asymptotically stabilizing output based nonlinear model predictive control scheme for a class of nonlinear discrete-time systems is developed. Results that are relevant in the context of manufacturing systems control are: • It is illustrated howthe aforementioned developed robust computationally friendly nonlinear model predictive control algorithm can be employed to solve a large scale manufacturing control problem in an efficient decentralized manner; • The relation between the so-called event domain modeling approaches for a class of discrete-eventmanufacturing systems to time domainmodels is derived. This results enables one to solve seemingly untractable time domain formulated optimal control problems for a class of manufacturing systems in a tractable manner; • An observer theory for a class of discrete-event manufacturing systems is developed.