(143f) Networked Predictive Control Of Process Systems | AIChE

(143f) Networked Predictive Control Of Process Systems


Muñoz de la Peña, D. - Presenter, University of California, Los Angeles
Liu, J. - Presenter, University of California, Los Angeles
McFall, C. - Presenter, University of California, Los Angeles
Ohran, B. - Presenter, University of California, Los Angeles
Davis, J. F. - Presenter, University of California - Los Angeles

Most control systems are designed under the assumption of flawless communication at the sensor-controller and controller-actuator links. This assumption holds in most applications where point-to-point communication links are used. However, nowadays there are an increasing number of industrial processes controlled via a shared communication network. Control systems which operate over a communication network (wired or wireless) are known as networked control systems (NCS) and can substantially improve the efficiency, flexibility, robustness and fault-tolerance of an industrial control system as well as reduce the installation, reconfiguration and maintenance costs. However, in addition to dealing with complex process dynamics (e.g., nonlinearities and uncertainty) and enforcing certain optimality properties in the closed-loop system, NCS have to account for the dynamics introduced by the communication network. In general, network dynamics are modeled as time-varying delays, data quantization or data losses.

This work focuses on networked predictive control of nonlinear process systems subject to data losses. In order to regulate the state of the system towards an equilibrium point while minimizing a given performance index, we propose a Lyapunov-based model predictive controller (LMPC) which is designed taking data losses explicitly into account, both in the optimization problem formulation and in the controller implementation. The central idea is to implement the computed predictive control manipulated input trajectory over the time interval in which communication between the controller and the actuators or sensors is lost and the loop opens. The proposed controller allows for an explicit characterization of the stability region and guarantees that this region is an invariant set for the closed-loop system (under data losses) if the maximum time of open-loop operation is shorter than a given duration. The length of this period depends on the parameters of the system and the Lyapunov-based controller that is used to formulate the optimization problem. The theoretical results are demonstrated through a series of chemical process examples. First, the proposed networked predictive control approach is applied to a reverse osmosis desalination process model. Using the proposed LMPC, the reverse osmosis system is demonstrated to maintain a desired level of permeate water quality when experiencing network connectivity problems or data losses in the controller-actuator and/or controller-sensor communication links. The second example focuses on a continuous crystallization system where the LMPC controller is designed on the basis of an approximate moment model and is shown to stabilize an open-loop unstable steady state of the population balance model in the presence of input constraints and data losses. Finally, the networked predictive control approach is applied to a polyethylene reactor example to achieve stabilization of an unstable steady state in the presence of data losses.