Projeto e análise de estabilidade de sistemas reais de controle via redes

Abstract

In a control system with a loop closed via a data communication network, better known by the acronym NCS (which stands for Networked Control System), delays are introduced in the transmissions of messages and consequently in the execution of the control signals. The delays in NCSs, closed over random access networks, as the Internet, are highly dependent on the network traffic conditions and present time-varying statistical parameters, e.g., average time and variance. Analyses of actual delays measured in real NCSs, implemented over the Internet, reveal that a single probability distribution is not able to represent the observed behavior. The occurrence of variable delays causes parametric variations and uncertainties in the representation of the dynamics of the loop, which increase the complexity of the analysis and synthesis of the controllers, as well as the difficulty of simulating the real conditions by means of computational algorithms. The analysis and simulation tools, in general, have limitations on the incorporation of real phenomena present in the NCS. In addition, they can lead to extremely conservative synthesis conditions, to the point of jeopardizing the performance for ensuring stability guarantee in any supposedly feasible condition given the established limits of delays. In this PhD Dissertation, the effects of the delays distribution and the most likely regions of stability in NCSs, subject to variable delays are analyzed. Some synthesis and analysis tools available in the literature have been evaluated and, due to some limitations for practical implementation to a real case, a new analysis methodology is proposed. Based on the Monte Carlo method and on the analysis of time response, it provides conditions for analysis of the effects of the variable delays and the indices (which can be used as parameters in the controller synthesis), e.g., the maximum variable delay supported by the NCS with certain delay distributions. This methodology, when compared to counterparts found in the literature (such as, for example, the over-approximation method based on polytopic and norm limited approaches), presents results with lesser degrees of conservatism. The proposed method has also been applied for analysis of physical plants. NCSs were implemented in three distinct physical plants with time constants ranging from milliseconds to hours, including processes with restrictions that require more elaborated control solutions. Different control strategies have been proposed, implemented and evaluated. This study revealed that the use of Adaptive Smith Predictors to compensate for the effects of delays brings significant performance improvements, but requires that the network messages have timestamps so that the effective values of the delays can be determined. The study also showed that the strategy referred to as Explicit Compensation of the Control Signal is very attractive for implementation of NCSs in systems where the messages do not have timestamps. A commercial refrigeration equipment has been one of the case studied, in which, by using the proposed control strategy, implemented over the CAN-Internet network, called Modulated Control with Restriction, it has been possible to achieve significant energy savings (around 34%) and also a significant reduction in internal temperature fluctuations. The many experimental results presented (hundreds of execution hours and more than one million of data samples collected in real NCS) confirm the thesis that it is necessary to consider the delay statistics for the analysis of networked control systems. This study emphasizes the need for more research on the representations of the statistics of the delay and its effect on the design of real networked control systems.