Model based predictive networked control systems /

Official URL: http://risc01.sabanciuniv.edu/record=b1160393 (Table of Contents)

Advantages of networked control systems (NCS) are very diverse and NCS’s address many of the demands of industrial development. As more and more sophisticated problems arise, networked control systems will not only become a convenience or an advantage but they will become an indispensable necessity. However usage of networked control systems introduces different forms of timedelay uncertainty in closed-loop system dynamics. These time delays are caused by the time sharing of the communication medium as well as computation time necessary for control algorithms and digital to analog conversions and have a destabilizing effect on system performance. Computational power of computers has increased dramatically; networks speed has also increased. Although both the network and computer architectures have tended to improve throughput over time, their real-time characteristics have not evolved to match the requirements from a control point of view. New control methodologies that cope with these factors and even take advantage of them are emerging. This work first examines some current methods in design and implementation of networked control systems that try to improve existing methods. Then a novel networked control system architecture that runs under non ideal network conditions with packet loss and noise is introduced. The proposed network control system architecture uses a model to predict the plant states into the future and generate corresponding control signals, then an array of the predicted control signals is sent to the actuator node side of the NCS rather than a single control signal like in basic networked control systems. This array of signals can control the plant if they are applied consecutively with sampling time intervals. However this is not the case under ideal conditions, where the network is lossless. Only the first control signal in each array is applied to the plant as a newer packet arrives every sampling period. The remainder of the array of predicted control signals is only used when packet loss occurs. This approach enables the system to be controlled in a pre-simulated manner and stability can be maintained even with high packet loss probabilities. Synchronization of the network elements becomes a major problem in this approach since models are involved. Synchronizing the actuator and controller nodes is done by an algorithm that can identify control signal arrays that have trustable information. Also the controller has a distributed architecture; some parts of the controller are implemented in the sensor node. This is to ensure that sensor to controller synchronization is not broken. The proposed model based predictive networked control system architecture was tested on a DC motor. The effects of packet loss were examined to reveal that the packet loss does not cause destabilization of the system, when packet loss occurs and the control packet cannot be sent to the actuator node, which prevents the changes in reference from being applied to the plant. The overall effect is the retardation of the response of the plant to the reference. Effects of noise are also examined. Under low packet loss conditions noise does not have an unusual effect on the system but when packet loss increases noise cannot be tolerated because the feedback loop is interrupted due to packet loss. Finally a method for determining the number of predictions to be made at the controller node (the prediction horizon) is suggested. The systems settling time is examined and the settling time is taken as the basis for the prediction horizon. The transmission of a single array of control signals from the controller node to the actuator node will enable the system to reach the desired reference. However this approach is only valid for open loop stable systems.