(740b) Quasi-Decentralized Networked Output Feedback Control of Process Systems Using a Dynamic Communication Policy

Chemical plants are large-scale dynamical systems that consist of a large number of distributed interconnected subsystems. Traditionally, the controller synthesis problem for such plants has been addressed within either the centralized or decentralized control frameworks. An approach that provides a compromise between the complexity of centralized control schemes and the performance limitations of decentralized control approaches is quasi-decentralized control [1] which refers to a distributed control strategy in which most signals used for control are collected and processed locally, while some signals are transferred between the local units and controllers to adequately account for the interactions and minimize the propagation of process upsets. A key consideration in this strategy is to enforce the desired closed-loop stability and performance objectives of the plant with minimal cross communication between the component subsystems. This is an appealing objective particularly when the communication medium is a resource-constrained wireless sensor network and conserving network resources is key to prolonging the service life of the network. Information transfer between the plant units is kept to a minimum by embedding within the local control systems dynamic models that provide the local controllers with estimates of the states of the neighboring units when communication is suspended, and updating the states of those models when communication is restored.

A key feature of the communication logic used in [1] is that it is static in the sense that the allowable communication rate is constant and can be computed off-line prior to plant operation. More recently, we developed in [2] a feedback-based communication policy in which the necessary communication rate can be determined and adjusted on-line (i.e., during plant operation) based on the evolution of the state of the plant. The key idea is to locally monitor the evolution of the state of each unit and request model updates from the rest of the plant only when a state-dependent stability bound is breached. This approach is robust to unpredictable disturbances and allows the plant to respond quickly in an adaptive fashion to a unit that requires immediate attention. Another advantage of this approach is that it ultimately leads to a more efficient utilization of network resources since the communication rate is increased only when necessary to maintain closed-loop stability. In most practical applications, however, direct measurements of the full-state are typically unavailable, and this introduces a number of challenges that need to be accounted for at the local control level, as well as at the plant-wide communication level. At the local control level, for example, the lack of full-state measurements necessitates the design of suitable state observers to generate estimates of the states of each unit from the available measurements. Each observer must be designed to enforce quick decay of the estimation error in the presence of interconnections between the plant units. This is critical given that the observer-generated estimates are needed to implement the local control action and must also be transmitted to the neighboring units to perform model updates at communication times. Observer estimation errors also require placing additional restrictions on the communication logic to account for these errors explicitly in the stability threshold and for the fact that only the observer-generated estimates can be monitored and used to assess the stability properties of the individual subsystems.

Motivated by these considerations, we present in this work a quasi-decentralized networked control structure with a dynamic communication logic for plants with limited state measurements and interconnected units that exchange information over a shared, resource-constrained communication network. The structure brings together ideas from model-based feedback control, high-gain observers, and feedback-based communication. The objective is to find a strategy for establishing and terminating communication between the local control systems in a way that minimizes the utilization of network resources without jeopardizing closed-loop stability. To this end, we initially synthesize for each unit a dynamic output feedback controller that enforces practical stability and ultimate boundedness of the states in the absence of communication suspension. The design is realized by combining a robust Lyapunov-based state feedback controller with a state observer that generates estimates of the local state variables from the measured outputs. The observers are constructed using high-gain observer design techniques to ensure an arbitrarily fast convergence of the state estimation error in the presence of interactions between the plant units. To reduce information transfer over the network, a set of models are included within each local control system to provide estimates of the states of the neighboring units when communication is suspended. To determine when communication must be re-established, the evolution of the observer-generated state estimates is monitored locally within each unit such that if it begins to breach a pre-specified time-varying stability threshold at any time, the neighboring units are prompted to send their data over the network to update the corresponding models. The stability threshold is determined using Lyapunov techniques and can be tightened or relaxed by proper selection of the controller and observer design parameters. Finally, the stability and performance properties of the proposed networked control structure are illustrated using a chemical plant example.

References:

[1] Sun, Y. and N. H. El-Farra, ``Quasi-decentralized model-based networked control of process systems," Comp. Chem. Eng., 32:2016-2029, 2008.

[1] Sun, Y. and N. H. El-Farra, ``Quasi-decentralized networked process control using an adaptive communication policy," Proceedings of American Control Conference, to appear, Baltimore, MD, 2010.

->