Model-based estimation and control for systems over communication networks

Resumen

In the last yearswe havewitnessed the introduction in the control loop of different telecommunication technologies, such as data networks, smart sensors, modern mobile telephony and internet. The control of systems over communication networks constitutes a recent branch of automatic control. The adoption of these new communication capabilities incur in additional issues that ought to be accounted for. In particular, communications delays, packet-based communication, possible data losses, quantization effects, bandwidth limitations and energy consumption, to name a few, are relevant features to be faced in this new paradigm. These issues become critical in real-time applications. This thesis proposes new solutions for the control and estimation of systems over communication networks. Although the thesis is mainly focused in problemswhere bandwidth and energy consumption constraints apply, other effects, such as delays and packet dropouts, will also be considered where appropriate. First of all, the thesis studies the stability of time-delay systems (TDS) and networked control systems (NCS) affected by delays and packet dropouts. A new stability criterion is proposed, achieving less conservative results than existing methods in the literature. Secondly, the thesis presents a novelmethod to design H2/H¥ controllers applicable to TDS and NCS. The method is demonstrated to synthesize controllers that achieve an upper bound of the cost index lower than other approaches. Furthermore, the reduction of the traffic over the network is explored by introducing a model of the plant at the controller end of the communication. A periodic and a self-triggered sampling policy are proposed. Concerning decentralized large-scale systems, the sensor scheduling problem is of great interest when the available bandwidth is severely limited. The thesis proposes two novel solutions in this line: a scheduling based on a predefined periodic pattern and a Kalman-based aperiodic filter. Although the former is a mathematically simpler solution and more energy-efficient, the latter yields better performance. Then, it is shown that, under some assumptions, a priori aperiodic solutions eventually produce periodic patterns, providing with the benefits of both approaches. Finally, the thesis tackles a problem that, despite its importance, has received little attention in the literature: the joint problem of estimation and control for distributed systems. The objective is to propose a design method that ensures the system stability, providing a cost-guaranteed solution with respect to a given quadratic index. The reduction of the bandwidth usage is attained exploiting an event-based communication policy between agents. Most of the contributions of this thesis are of theoretical nature. Notwithstanding, experimental applications have not been forgotten. Two experimental testbeds have been considered, namely a networked control of a direct drive robotmanipulator and an educational four-tank level control system.